> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prisme.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# What's New in Prisme.ai

> Latest features and improvements in the Prisme.ai platform

Stay up-to-date with the latest enhancements, features, and improvements to the Prisme.ai platform. Our team continuously works to deliver new capabilities that help you build and deploy more powerful AI solutions.

## Latest Release

<Update label="v27.3.1" description="Releasing version v27.3.1">
  **Features:**

  * **Knowledge App + MCP:** New app and MCP server that expose the Knowledge Base API directly inside Builder and Agent Factory, so agents can query and manage knowledge bases without raw API calls.
  * **Connectors:** Redesigned the configuration interface of all connectors (Salesforce, Gryzzly, Power BI, SAP LeanIX, Google Search, Google Workspace, SharePoint, Outlook, Figma, GitLab, HubSpot, Azure OCR) for a clearer, more user-friendly setup experience.
  * **Agent Creator:** Access an uploaded file directly from a tool.
  * **AIG > Org Join Rules:** Reorder join rules with drag & drop.
  * **Governance > Join Rules:** Added support for the `contains` operator, allowing an array of strings to be tested against a specific value.
  * **Governance > Users:** Super admins can now reset a user's MFA.
  * **LLM Gateway:** Allow overriding the full URL in the Anthropic provider, along with an `auth_scheme` option (`x-api-key`, `api-key`, or `bearer`).
  * **LLM Gateway:** Added uniform support for `options.headers`, `options.parameterAliases`, and `options.excludeParameters` across all LLM providers.
  * **Governance > Models:** Added the `options.headers`, `options.parameterAliases`, and `options.excludeParameters` options to the Providers UI.
  * **Backends:** Support new environment variables for the PostgreSQL/MongoDB connection pool minimum and maximum sizes: `<prefix>_POOL_MIN` and `<prefix>_POOL_MAX`.
  * **Agent Creator > Import API:** To update an existing or owned agent from a `.md` file, the `target_agent_id` body field must now be provided — otherwise a new agent is created.

  **Improvements:**

  * **Agent Creator:** When importing an existing or owned agent `.md` file from the "Create agent" button, the user is now warned they are about to create a new agent rather than update the existing one. Update imports must be done from the agent's 3-dots action menu.
  * **Chat:** Copying HTML output is now formatted the same way as in the UI.
  * **Chat:** The Esc key now triggers quick actions without needing to type `@` or `/`.
  * **Fetch:** Renamed the native `networkError` output field to a more generic `error` field.
  * **Vector Stores:** Always return a standardized `error` output field whenever an error occurs.
  * **prismeai-events:** Forged system events emitted from untrusted surfaces are now rejected directly at the broker.
  * **Helm chart prismeai-core:**
    * `1.10.1`: No longer creates database or user-provided secrets (storage, permissions, broker, …) when their `existingSecret` is set (no `lookup` attempt); the `secrets.allowCreation` global value now applies only to internal secrets.
    * `1.10.0`: Added the `secrets.allowCreation` global value to disable secret creation/overriding, preventing secrets from being reset when the Helm `lookup` command cannot run (e.g. with ArgoCD or `helm template`).

  **Bugfixes:**

  * **api-gateway:** MFA is no longer triggered for SSO-authenticated users, even when they also have a linked local account.
  * **api-gateway:** Fixed manually assigned roles being lost after a change to join rules.
  * **API Gateway:** Fixed file uploads so JSON metadata sent with native/multipart uploads is correctly preserved.
  * **Agent Creator:** Fixed persistence of some parameters related to an agent's token limits.
  * **Agent Factory:** Fixed the empty instructions field in the UI after importing a new agent (it was correctly imported and appeared after a refresh).
  * **Agent Factory:** Fixed uploading files from the Agent Creator playground.
  * **Governance:** Aligned subscription creation with the API's expected content.
  * **Platform:** Fixed occasional "Invalid CSRF token" errors that could occur when the user's CSRF token was updated during or right after a page load.
  * **Platform:** Fixed the broken `prismeai-console.service.port` Helm value used to change the prismeai-console default HTTP port.
  * **prismeai-crawler:** Fixed `REDIS_USER` and `REDIS_PASSWORD` environment variable support.
  * **Governance:** Fixed a "missing permission `storage:vector_store:write`" error when testing a vector store or updating platform models as a superadmin with no org role defined.
</Update>

## Previous Releases

<Update label="v27.3.0" description="Releasing version v27.3.0">
  **Features:**

  * **Platform > Hooks:** Intercept the agent's runtime pipeline at lifecycle events to inspect, modify, allow, or block messages, files, and LLM calls. See the [Hooks documentation](/products/agent-factory/capabilities#hooks).

  * **Chat:** Added a `slides` artifact type with live preview and PPTX/PDF export.

      <Frame>
        <img src="https://mintcdn.com/prismeai/DJvWlLTfn91rK9Wh/gifs/slides-pptx-preview.gif?s=fbded1bfe4302010bf55634f292380a5" alt="Slides artifact — live preview with PPTX/PDF export" width="1200" height="573" data-path="gifs/slides-pptx-preview.gif" />
      </Frame>

  * **Chat:** Show HITL tool execution details.

  * **Collections:** Added `min` and `max` aggregate functions.

  * **Home:** Allow disabling the home agent input chat.

  * **prismeai-workspaces:** Emit new `workspaces.files.uploaded` and `workspaces.files.updated` events each time a file is uploaded or updated.

  * **Agent Factory:** Now aggregate `last_messages_at` per agent every 15 min (`refresh-agent-counters` schedule), and allow filtering agents with the new `last_message_before` / `last_message_after` query parameters.

  * **Knowledge:** Server-side search for knowledge bases and documents, fixing broken tag search.

  * **Knowledge:** Allow re-indexing a file with per-document parser and chunking overrides.

      <Frame>
        <img src="https://mintcdn.com/prismeai/DJvWlLTfn91rK9Wh/gifs/kb-reindex-preview.gif?s=5c8da0661a712a8b8727eb84c8d86a3f" alt="Re-index a knowledge base file with per-document parser and chunking overrides" width="1200" height="576" data-path="gifs/kb-reindex-preview.gif" />
      </Frame>

  * **Prompt Library:** Added a REST API to manage prompt categories.

  * **Brave Search Connector:** Support using Goggles for URL whitelisting.

  * **LLM Gateway:** Support `{user.<path>}` placeholders in OpenAI-like custom headers.

  * **api-gateway:** Allow injecting environment variables inside the `authproviders` config.

  **Improvements:**

  * **Platform:** MFA is now enabled by default for local accounts (SSO is not affected). See [Multi-Factor Authentication (MFA)](/self-hosting/entreprise/authentication#multi-factor-authentication-mfa).
  * **Governance:** Added a confirmation prompt when deleting a member of a group.
  * **Chat:** Interleave chat text and tool activity chronologically.
  * **Chat:** Don't open tools/mentions when pasting URLs or emails.
  * **Agent Creator / Chat:** Tools requiring approval ("Always ask" mode) now display their friendly name, matching the auto-approval behavior.
  * **LLM Gateway:** Stop injecting empty Anthropic text blocks and normalize Vertex/Azure Claude streams to OpenAI chunks.
  * **Docker:** Keep only each service's own production dependencies.
  * **Documentation:** Rewrote and improved the Self-Hosting documentation.
  * **Helm charts:** Added secret protections by default for master keys, improved the README around master-key management, and prevent accidental overwrite.
  * **Helm chart prismeai-core:**
    * `1.9.11`: Updated the default Kubernetes resource requests/limits with a *Balanced* profile. See [Kubernetes Resources & Autoscaling](/self-hosting/kubernetes/resources).
    * `1.9.10`: Added the `secretKeys.password` value for all MongoDB/Postgres databases.
    * `1.9.8`: Added the `global.proxy` value to globally configure an HTTP/HTTPS proxy.
    * `1.9.7`: Added the `global.extraCABundle` value to inject a custom CA bundle in all core services.
  * **Helm chart prismeai-apps:**
    * `0.5.16`: Updated the default Kubernetes resource requests/limits with a *Balanced* profile. See [Kubernetes Resources & Autoscaling](/self-hosting/kubernetes/resources).
    * `0.5.15`: Added the `secretKeys.password` & `tlsCaFile` values in `global.storage.searchengines`.
    * `0.5.12`: Added the `global.proxy` value to globally configure an HTTP/HTTPS proxy.
    * `0.5.11`: Added the `global.extraCABundle` value to inject a custom CA bundle in all apps services.
    * `0.5.10`: Updated default environment variable values with recommended production values (`REQUEST_MAX_SIZE=20mb`, `KERNEL_POOL_SIZE=10`) and automatically calculate the functions NodeJS `--max-old-space-size` from the k8s memory limit. The `NODE_OPTIONS` environment variable must be removed for the Helm upgrade to succeed, as it is now added natively by the chart.

  **Bugfixes:**

  * **Builder:** Fixed an issue where autocompletion wasn't correctly indented.
  * **Builder:** Fixed the workspace name not displaying on the browser's tab.
  * **Agent:** Fixed starter card icons display on conversations.
  * **Agent:** Fixed a case where a conversation could be blended into another agent's conversation while switching during response generation.
  * **Agent Creator / Chat:** Preserve HITL turn state and fixed an issue where the default model was used following an HITL instead of the current model.
  * **Agent Factory & LLM Gateway:** Fixed calling APIs from another workspace using `Authorization: Bearer <JWT>` or `x-prismeai-api-key <ApiKey>` headers.
  * **Knowledge:** Fixed organisation and group shared knowledges not being displayed in the frontend platform.
  * **Knowledge:** Fixed sharing knowledge bases with a group.
  * **Crawler:** Normalize Tika metadata before Redis storage, avoiding errors such as `Invalid input of type: 'list'` for specific files.
  * **prismeai-workspaces:** Fixed the api-gateway readiness test — prefer `INTERNAL_API_URL` over `API_URL`, as the public `API_URL` may not be reachable in some environments.
</Update>

<Update label="v27.2.6" description="Releasing version v27.2.6">
  **Features:**

  * **AIG & Knowledge:** Added platform-level and org-level default *max number of documents* for agent knowledges and user knowledges.
  * **Knowledge:** Show an error message when the maximum number of documents has been exceeded.
  * **Postgres ORM:** Implemented `$exists` query on nullable array columns.
  * **Connectors:** All Google app + MCP connectors (Drive, Docs, Sheets, Calendar, Mail, Search) received security updates and now follow the PascalCase `AppSlug` naming guideline.
  * **Docs:** Added app connector logos to the [connectors catalog](/apps-store/marketplace/connectors/overview).
  * **Docs:** Added the filterable [models catalog](/products/ai-governance/available-models) (by provider and region).

  **Improvements:**

  * **AWS Bedrock:** Handle `tool_choice=none` with Bedrock models by converting history tool calls to text messages and entirely stripping the tools config.
  * **Knowledge:** Decoupled `/stats` from page render to unblock the dashboard on slow stats.
  * **Agent Creator:** Improved conversation-title generation prompt to respect the user's language instead of always defaulting to English.
  * **Platform:** `WEBSOCKETS_DEFAULT_TRANSPORTS` environment variable is now respected again (it was ignored since the new platform, which always used `polling,websocket`).
  * **Platform:** Default value for `WEBSOCKETS_DEFAULT_TRANSPORTS` is now `websocket,polling` to ease deployment/configuration for websockets (covers 99% of scenarios).
  * **Chat:** Wait 3 s before displaying the *network issue* banner — it could briefly appear during page loading.
  * **Chat:** Detect upload MIME type from the first signature bytes instead of relying on the file extension (which could be wrong and cause request errors with the Bedrock API).

  **Bugfixes:**

  * **Agent Creator:** Fixed the agent search input in the *Sub-agent* capability selector.
  * **Agent Creator:** Fixed calling `messages/send` or `messages/stream` twice with the same `correlationId` — always generate a unique `taskId` so the second call no longer crashes on task DB insertion.
  * **Agent Creator:** Preserve `tool.parameters` on `extractConfig` so system/memory edits no longer reset to schema defaults on reload.
  * **Agent Creator:** Fixed the *evaluate judge* model picker to honor the org IAM allowlist.
  * **Agent Factory:** Fixed the final message displayed when the budget is exceeded after multiple tool calls on a Bedrock model — the end user is now notified of the issue and informed of what was completed.
  * **LLM Gateway App:** Fixed methods being non-callable because of `required: true`.
</Update>

<Update label="v27.2.4" description="Releasing version v27.2.4">
  **Features:**

  * **AIG:** Enable org-based anonymous sessions (configured from the *Authentication* page, where SSO also lives).
  * **Chat Embed:** Anonymous authentication is now supported from the embedded chat.
  * **Agent Creator:** Introducing **"user-first" tools** — hide tools from the agent until the user summons them from the chat input. Works for function tools and whole MCP integrations, and composes with the existing approval policies.
  * **Agent Creator:** Allow adding permission exceptions for skills.
  * **Agent Creator:** Allow unpublishing agents.
  * **Agent Creator:** Allow removing system tools (`memory`, …).
  * **Agent Creator:** Show a warning or error message below the selected model when it has been removed, disabled, forbidden, or hidden by platform/org admins.
  * **Knowledge:** Added the parser selector (`tika` / `tika-ocr` / `unstructured` / `unstructured-ocr`) to the Settings tab so it can be changed post-creation, not only at store creation.
  * **Chat:** Auto-open the canvas when a new artifact starts streaming.
  * **Image Generation:** Image generation features are back (see *Configuration check* below for required setup).
  * **Governance:** Added `tool_memory` to seed capabilities.
  * **Salesforce connector:** New connector available.

  **Improvements:**

  * **api-gateway:** Use a per-request fork of the global database ORM so each request runs its own MikroORM UnitOfWork and cannot poison the shared global ORM, preventing the cascade where one bad request would break all future transactions.
  * **prismeai-functions:** Reuse existing `vm2` workers instead of recreating them on each execution (previously slowed down functions that imported packages on slow filesystems).
  * **AIG > Agent Controls:** Automatically remove stale model IDs from the org's allowed models when they no longer exist at platform level.
  * **AIG > Agent Controls:** Search also now matches agent descriptions, mirroring the agents store search behavior.
  * **Store:** Removed the "Create agent" button.
  * **Platform:** Enhanced SSE keepalive on stream start to prevent network errors on long-running tasks.
  * **Platform:** Activity timeline made more discreet, with a shimmer label while running.
  * **Crawler:** Lowered the default download size limit from 1 GB to 120 MB to prevent zip bombs or broken HTML pages from saturating the crawler's RAM.

  **Bugfixes:**

  * **api-gateway:** Fixed a gateway crash when an external SSO provider sent `firstName` or `lastName` as a string, corrupting the MikroORM unit of work and making every subsequent ORM flush fail.
  * **AIG > Agent Controls:** Fixed listing agents with their allowed models as a non-super-admin.
  * **AIG > Groups:** Fixed the "add users" input to search matching users across the entire organization.
  * **Governance:** Fixed model search for users that do not have full platform access.
  * **Platform:** Fixed live notifications during platform pull, and closing the pull modal at the end.
  * **Knowledge:** Per-file delete now also removes the crawler-side document; vector store `DELETE` surfaces external-cleanup failures as retry-able errors.
  * **Agent Creator > Import:** `agents.md` import now always recreates the `conversation_file_search` vector store instead of reusing the source agent's.
  * **Agent Creator:** Fixed crash when a parameter has `type: json`.
  * **Agent Creator:** Fixed 406 responses from spec-strict MCP servers — full `initialize` / `notifications/initialized` handshake with SSE response support.
  * **Agent Creator:** Fixed the handling of custom values for `token_budget`, `max_turns`, and `tool_call_budget` (they were being overridden by the profile's default values).
  * **Agent Creator:** Leave conversation names empty on creation so the frontend chat shows the proper "New conversation" translation.
  * **Agent Creator:** Fixed generating the conversation title when the first message requires a tool approval.
  * **Chat:** Fixed rendering of markdown tables inside the canvas.
  * **Chat:** Fixed conversation-title generation (some models were replying to the user message instead of generating a title).
  * **Chat:** Fixed conversation-scoped model selection — when opening a conversation whose last message used a non-default model, the model picker keeps that custom model; when creating a new chat from such a conversation, the picker resets to the agent's default model.
  * **Chat > Canvas:** Download button now respects the LLM-generated filename and extension.
  * **OpenAI / Azure:** Image uploads with Azure OpenAI models now work correctly and no longer trigger Azure jailbreak guardrails.

  <Warning>
    **Configuration check — Image Generation setup**

    To re-enable image generation on existing deployments, apply the following setup:

    1. Add `image-generation` to the `trusted_source_workspaces` list in the **LLM Gateway** workspace config (`index.yml`):

    ```yaml theme={null}
    config:
      value:
        trusted_source_workspaces:
          - storage
          - image-generation
          - agent-factory
    ```

    2. Set the default image generation model in **AI Governance** at [Agents Controls](https://studio.prisme.ai/apps/ai-governance/prisme-ai/agents-controls):
       * Set the `default_image_generation_model` value.
       * Add the new default-image-model field where required.
  </Warning>
</Update>

<Update label="v27.2.3" description="Releasing version v27.2.3">
  **Features:**

  * **AIG > Members:** Added members CSV import/export buttons.
  * **Agent Creator:** Patch agent from `agents.md`.
  * **Agent Creator:** Drag & drop to import `agents.md`.
  * **Agent Creator:** Enabled message stats in the playground for `correlationId`.
  * **Agent Creator:** Enabled tool choice in the playground.
  * **Agent Creator:** Scope the activity feed to one single agent.
  * **Chat:** Force Canvas.
  * **Chat:** `embed-chat` — added `auth:token` mode.
  * **Chat:** Added option to `disablePaste`.
  * **Chat:** Proactive OAuth for tools in PromptInput.
  * **Chat:** Ask Agent with a selected snippet.
  * **Platform:** CRUD personal access tokens UI in user settings.
  * **Builder:** Rich automation-graph editor and assorted UI updates for all missing DSUL instructions (`run`, `rateLimit`, …).

  **Improvements:**

  * **Governance:** Clarified PermissionTree hierarchy.
  * **Governance:** Renamed LLM permissions to match LLM Gateway ones.

  **Bugfixes:**

  * Various bug fixes and minor improvements across the platform.
</Update>

<Update label="v27.2.2" description="Releasing version v27.2.2">
  **Features:**

  * **Builder > Activity Feed:** Show Export button on any active activity filter.
  * **prismeai-runtime:** Added new API key related instructions to the access-manager runtime module.
  * **AIG Observability > Usage:** Added total unique users & unique users over time.
  * **AIG Observability & Agent Creator:** Added 5-minute granularity for periods ≤ 2h.
  * **AIG Observability Feed:** Added user and text search input, duration filtering, and display of user names & execution duration.
  * **AIG Observability Usage > Daily Activity:** Added a second line for executed automations over time.
  * **AIG > Members:** Invite modal now shows the configured default role in the role input.
  * **AIG > API Keys:** Display only governance API keys by default; allow filtering by vector stores or agents API keys only.
  * **AIG > Models Providers:** Google Vertex provider now also handles arrays of providers and `options.excludeParameters`.
  * **LLM Costs:** Separate input & output counters.
  * **Agent Creator:** Added capabilities graph view.
  * **Agent Creator:** Added agent-scoped API key generation and revoke.
  * **Agent Creator:** Added API key management widget in the Sharing tab.
  * **Agent Creator:** Enabled starters configuration.
  * **Agent Creator:** Enabled file upload in the playground.
  * **Storage:** Added API key management widget in the Sharing tab.
  * **Chat & A2A:** Added `limits_reached` status type to the chat activity timeline.
  * **Chat:** Replaced `connector_auth_required` SSE with HITL `external_auth`.
  * **Platform:** Added guided tours to all built-ins.
  * **Platform:** Added pinned items to the shell sidebar with server persistence.
  * **Platform:** Wired A2UI surface actions to HITL approval.
  * **Insights:** Added fullscreen expand mode to the graph page.

  **Improvements:**

  * **prismeai-runtime:** Always emit `runtime.automations.executed` with error details on the innermost failing automation (previously the error itself was emitted but not the corresponding `runtime.automations.executed` event, which skewed execution statistics).
  * **prismeai-events:** Always configure `aggPayload` mapping for native fields from `runtime.automations.executed` & `runtime.interactions.triggered` platform events, enabling advanced analytics aggregation.
  * **AIG Observability:** Error tables now include a link to the workspace Activity Feed page with the corresponding filters.
  * **AIG Observability:** The "Recalculate metrics" button now refreshes the entire selected range, not only the last opened hourly bucket.
  * **AIG Observability:** Always show the 20 first entries in "Top…" widgets, with frontend pagination.
  * **AIG Observability:** Always display the 20 most active workspaces in the workspace picker.
  * **AIG Observability & Agent Creator:** Show analytics computation date & scheduled next refresh date.
  * **AIG Observability:** Synchronize filters with URL query parameters.
  * **AIG Observability Metrics:** Latency Over Time now also represents interactions over time.
  * **AIG Observability & Agent Creator:** Improved Date Range Picker (now also includes time).
  * **AIG Observability & Agent Creator:** Allow selecting a custom date-time range using drag and drop on the timeseries charts.
  * **AIG > Observability:** Documentation updated.
  * **AIG:** Deleting a user account now removes all of its org & group memberships.
  * **Imports:** Always restore the workspace `index.yml` if it has been removed from object storage, and raise a warning in the import result.
  * **Builder:** Activity payload expand modal & execute modal UX improvements.
  * **Chat:** Show agent header on hover and cleaned up chat UI.
  * **Agent Creator:** Improved tips.
  * **Insights:** Various improvements and fixes.
  * **Front:** Updated dependencies and removed unused `i18next-http-backend`.

  **Bugfixes:**

  * **Builder > Activity Feed:** Fixed filters disappearing from URL query parameters after refresh.
  * **AIG Observability:** Fixed empty timelines.
  * **AIG Observability Feed:** Fixed observability feed tab.
  * **AIG > Infrastructure:** Fixed testing the vector store when the default embedding model had a configured dimensions value different from 1536.
  * **Builder:** Fixed dependency matrix to show only connections between workspace automations and installed app automations.
  * **Governance:** Fixed dark mode in MenuEditorPage.
  * **Agent Creator:** Fixed empty agent message bubble and agent card height.
  * **Chat:** Fixed fetching artifact content from backend when empty on click.

  <Warning>
    **Configuration check — LLM Gateway workspace**

    If the LLM Gateway workspace config was preserved during the upgrade (not overwritten, which is intentional in some setups to keep custom values), make sure the following entry is present in the workspace's `index.yml` (Edit Workspace YAML):

    ```yaml theme={null}
    config:
      value:
        trusted_source_workspaces:
          - storage
    ```

    Without this entry, the LLM Gateway will not trust the Storage workspace as a source, which can break Storage-related features.
  </Warning>
</Update>

<Update label="v27.2.0" description="Releasing version v27.2.0">
  **New features :**

  * **AIG:** Allow restricting models to specific orgs
  * **AIG:** Added a new infrastructure page to configure the vector store and test all DB connectivity/configuration
  * **AIG > Models:** Added an LLM providers edition tab
  * **AIG > Models:** Added a Test button to models
  * **AIG > Org Models:** Allow sorting the allowed models list per organization
  * **AIG > Models / Providers:** Allow importing model/providers config from legacy AIK
  * **AIG > Models Providers:** Allow importing/exporting providers from/to JSON config files

  **Improvements :**

  * **Knowledge:** Added headers configuration for automatic crawling, removed legacy options
  * **Storage:** Increased user uploaded files default expiration from 30m to 30d
  * **Chat:** The selected conversation now moves to the top of the list when sorted by date of the last message
  * **Chat:** User profile photo is now displayed in conversations when available
  * **Builder:** Native variable autocomplete in the editor optimized
  * **Builder:** Better handling when creating an automation with an existing slug
  * **Builder:** Custom Code function parameter names are now editable from the UI
  * **Builder Automations:** Complex conditions previously required YAML — now fully supported in the UI, including branch deletion
  * **Builder Automations:** `repeat`, previously only available via YAML, is now configurable from the UI
  * **Canvas:** UX and UI enhancements
  * **Platform UI:** Dark mode optimized across several components
  * **Agent Creator:** Ability to assign non-technical display names to Tools, MCP, and Skills
  * **Agent Creator:** Search replaces filter for faster agent discovery
  * **AIG:** Display an error message when an invited member is already in the org
  * **AIG:** Prevent saving embedding models without the dimensions field
  * **AIG:** Automatically convert legacy roles & groups to the corresponding org role & group when a user automatically joins an org through join rules
  * **AIG > Models:** Show an error for models not provided by any provider
  * **AIG > Models:** Added multi-select with export/delete actions
  * **AIG > Models:** Added autocompletion of available model ids when creating/cloning a new model
  * **prismeai-runtime:** Implement `WORKSPACE_SECRET_*` environment variables so workspaces can maintain the same config content whether secrets were configured through environment variables or workspace secrets
  * **prismeai-runtime:** Implement `/*multiSegmentParams` syntax in automation endpoint patterns so path params can include `/` (fixes deletion of models with `/` in their id)
  * **prismeai-workspaces:** Added `/workspaces/:workspaceId/config` GET & PATCH APIs
  * **prismeai-searchengine:** Removed initial website fetch when configuring a new website, so searchengine no longer needs internal TLS certs

  **Bugfixes:**

  * **Chat:** Fixed chat input border color in dark mode
  * **Shared conversation:** Fixed user message font color not taking the org custom theme into account
  * **Shared conversation:** Fixed invisible links in the shared view user bubble
  * **Builder:** Better handling of default values when defined as JSON in Custom Code
  * **Builder:** Some Custom Code error logs were previously visible only in the Network tab — they are now available in the UI
  * **Builder:** Fixed usage evolution statistics for newly created workspaces
  * **Builder:** Auto-quote `{{ }}` expressions in automation YAML to fix the "\[object Object]" bug
  * **Knowledge:** Use default parsing models from the LLM Gateway instead of hardcoded ones; declaration is also supported per organization
  * **Knowledge:** When leaving the default value for the embedding model, set the default value in the frontend to avoid cache drift
</Update>

<Update label="v27.1.0" description="Releasing version v27.1.0">
  **Features & Improvements:**

  * **AIG:** Org owners & admins can now also access the organization audit page.
  * **Explore:** Org members can now also access the explore page.
  * **Agent Creator:** Non-admin users can now also see their recent activity.
  * **AI Knowledge:** Added a tooltip to the legacy SharePoint connector page.
  * **Knowledge:** Initialize the first owner's binding on vector store creation.
  * **Knowledge:** Filter out the agent's vector store by default.
  * **AI Governance (legacy):** Auto-initialize the `superadmin` role for retro-compatibility.
  * **Crawler & Searchengine:** JSON logging for better support across cloud providers.
  * **AI Knowledge:** Improved `getPage` with retry options for ES index synchronicity.
  * **AIG:** Reworked Roles / API Keys edition UI.
  * **AIG:** Now allow updating native roles (except for Owner).
  * **prismeai-runtime:** New `bulkUpsert` collection method.
  * **prismeai-runtime:** Added `parseUrl` utility to the text module.
  * **prismeai-workspaces:** Added workspace uploads in-place update API.
  * **Agent Creator Analytics:** Added new metrics — total LLM calls, messages over time, and average number of calls per message.
  * **Agent Creator Analytics:** Display hourly granularity when the range is under 7 days.
  * **Agent Creator Analytics:** Added a custom date range selector.
  * **Agent Creator:** Added icons to agents.
  * **Chat:** Show message dates & agent name/icon.
  * **Chat:** Separate conversations into period groups (today, yesterday, last 7 days, last 30 days).
  * **Chat:** Allow sorting conversations either by last message or creation date.
  * **AIG:** Added import/export menu to Appearance settings.
  * **AIG:** Added a new "Legals & Warnings" tab under the Appearance page to customize the organization's terms of use, privacy policy, chat disclaimer, and any custom app warning modal.
  * **AIG:** Added favicon & light/dark icons upload forms.
  * **AIG:** Allow configuring the default homepage URL from the menu builder.
  * **AIG:** Added missing translations.
  * **Securechat:** Canvas can now appear between two paragraphs and not only at the end.
  * **Securechat:** Canvas can now be shared & viewed from a separate page.
  * **Securechat:** Removed duplicate copy / download buttons.

  **Bugfixes:**

  * **OpenAI app:** Fixed `allowed_tools` and function workarounds to use the supported OpenAI format.
  * **Agent Creator Analytics:** Fixed OpenSearch compatibility.
  * **Agent Creator Analytics:** Fixed users metric — now counts real users instead of sessions.
  * **Agent Creator Analytics:** Fixed messages metric — now counts real messages instead of LLM calls.
  * **Builder:** Fixed pages preview & pages deploy button.
  * **Agent Creator:** Fixed `messages/stream` API never showing error details.
  * **Agent Creator:** Fixed evaluations not taking the selected model into account.
  * **Agent Creator:** Fixed playground — now always uses the current agent config instead of the published config.
  * **Agent Creator:** Fixed playground — no longer creates conversations in Securechat.
  * **Agent Creator:** Fixed guardrails `mask` action.
  * **Securechat:** Fixed rating agents.
  * **Securechat:** Fixed uploading documents as an org member.
  * **Securechat:** Fixed processing large documents (they are now directly uploaded to the native files API with `multipart/form-data`).
  * **Securechat:** Fixed canvas instructions missing on warm start.
  * **Explore:** Fixed displaying "Request access" on the agents explore page even when the current user had already been granted access.
  * **Home:** Fixed uploading documents from the home page.
  * **Models Picker:** Fixed hiding disabled or hidden models.
  * **Models Picker:** Fixed displaying the selected model name instead of its slug.
  * **Chat:** Fixed displaying the selected model name instead of its slug under message details.
  * **Conversation share view:** Fixed canvas rendering to match the chat.
</Update>

<Update label="v27.0.5" description="Releasing version v27.0.5">
  **New Features:**

  * **Storage Permissions Management** — Fine-grained access control on storage resources, letting teams define who can read, write, or manage files across workspaces.
  * **Native Auth Binding for Storage** — Storage now integrates directly with the platform's native authentication layer, removing the need for external token handling and simplifying secure access.

  **Bugfixes:**

  * **Knowledge Base File Cleanup** — Storage files linked to knowledge bases are now correctly auto-deleted when their parent resource is removed, preventing orphaned data.
  * **Vector Stores Webpage Limit** — The `webpage limit` setting on vector stores is now properly initialized, ensuring crawling stays within the configured bounds from the first run.
  * **LLM Gateway Default Models Cache** — Fixed a cache inconsistency when fetching default models from the LLM Gateway, so the correct models are served on every request.
  * **AWS Bedrock Image Uploads** — Image uploads now work reliably when routing through AWS Bedrock providers.
  * **Organization API Key on Agent Creator API** — Organization-level API keys are now properly accepted by the Agent Creator API, unblocking org-scoped automation.
  * **InvalidAPIKey Errors** — Resolved spurious `InvalidAPIKey` errors that could occur during authenticated calls.
</Update>

<Update label="v27.0.0 — One Platform" description="Releasing version v27.0.0 — One Platform">
  **The Agentic AI Platform for the Enterprise** — Create, orchestrate, and govern enterprise-grade AI agents in real time.

  **Prisme.ai One Platform** is a fully modular, composable architecture for deploying AI agents at enterprise scale. Every service is API-first, independently deployable, and built around open standards — **A2A** (Agent-to-Agent), **MCP** (Model Context Protocol), and **A2UI** (Agent-to-UI).

  <Frame>
    <img src="https://mintcdn.com/prismeai/8iXhvO-JR41sv05G/gifs/homepage-preview.gif?s=3051d34847d32fc01774f060fab31ae6" alt="Prisme.ai One Platform — Homepage preview" width="1200" height="578" data-path="gifs/homepage-preview.gif" />
  </Frame>

  **Organizations & AI Gateway:**

  * **Organizations** — Unified scoping layer across all products: users, workspaces, billing, permissions, and quotas managed centrally.
  * **AI Gateway (AIG)** — Single entry point for all agent traffic, enforcing org-level policies, quotas, and routing across every Prisme.ai service.

  **Builder — Full-Stack Agentic Development:** The operating system for agentic workflows. Build automations, connect IT systems, and create full-stack apps — all with built-in observability and event-driven architecture.

  * **Agentic Workflows** — AI agents, automations, event-driven pipelines, real-time streaming.
  * **Connect IT Systems** — Legacy systems, connectors, webhooks, enterprise integrations.
  * **Built-in Observability** — Correlation IDs, audit trails, activity monitoring.
  * **Full-Stack Apps** — Web applications with React, Tailwind CSS, and Prisme.ai backend automations.

  **Core Services:**

  * **Agent Creator** — Create and manage agents with 5 profiles (simple to orchestrator), A2A Protocol, human-in-the-loop, streaming, and 15+ system tools (planning, memory, delegation, reflection, budget tracking).
  * **Storage** — Unified files, vector stores, and skills backend with chunking, web crawling, and RAG — pluggable with Elasticsearch or OpenSearch.
  * **LLM Gateway** — Multi-provider routing (OpenAI, Azure, Anthropic, Vertex, Bedrock, custom) with cost tracking and carbon footprint monitoring.
  * **Capabilities** — Unified catalog of tools, guardrails, skills, sub-agents, and memory — all MCPised and composable across the platform.

  <Frame>
    <img src="https://mintcdn.com/prismeai/8iXhvO-JR41sv05G/gifs/creationAgent-preview.gif?s=8dde94172972a0301c3ea31d93b29602" alt="Agent Creator — Create and manage agents" width="1200" height="577" data-path="gifs/creationAgent-preview.gif" />
  </Frame>

  **Skills & MCP-isation:** Every capability is exposed as an MCP server, making agents truly composable:

  * **Prompt Library** — 10 categories of reusable prompts, available as MCP tools.
  * **Guardrails** — Input / output / action safety with configurable sensitivity.
  * **Memories** — Long-term memory (facts, preferences, relationships, instructions) with similarity-based recall.
  * **Web Search** — Grounded search via Azure AI Foundry.
  * **AI Collection** — Structured data operations via MCP.

  **Analytics & Governance:**

  * **AI Insights** — Conversation analytics, evaluation criteria, GDPR anonymization, memory clustering, tiered plans (Free to Enterprise).
  * **AI Governance** — Platform observability, cost monitoring, audit trails, and role-based access.
  * **Agent Evaluations** — Automated scoring with test cases, batch runs, nightly scheduling.

  <Frame>
    <img src="https://mintcdn.com/prismeai/8iXhvO-JR41sv05G/gifs/insights-preview.gif?s=a95f7b08489232ac1ee0d5ab6c31bf1b" alt="AI Insights — Conversation analytics and governance" width="1200" height="573" data-path="gifs/insights-preview.gif" />
  </Frame>

  **Integrations Library:** A growing catalog of production-ready connectors for enterprise IT systems, SaaS platforms, document stores, and collaboration tools (Microsoft 365, Google Workspace, Salesforce, ServiceNow, Confluence, Jira…) — all exposed as MCP servers and composable within agentic workflows.

  **A2A & Collaboration:**

  * **Agents App & LLM App** — Importable modules so any workspace can manage agents or call LLMs without raw API calls.

  **Boosted UI:** All product interfaces have been redesigned for speed, clarity, and a modern enterprise experience.

  *Prisme.ai One Platform — Modular. Composable. Enterprise-ready.*
</Update>

<Update label="v26.03.161453" description="Releasing version v26.03.161453">
  **Features & Improvements:**

  * **prismeai-events:** Lowered the default socket.io polling interval to 15s to prevent gateway timeouts (the previous 25s interval exceeded the gateway's default 20s timeout). Added `SOCKETIO_PING_INTERVAL` environment variable (in ms) (Internal Ticket #3296)
  * **prismeai-crawler:** Added support for the `pdf_ocr_strategy` Tika option via the `X-Tika-PDFocrStrategy` header (Internal Ticket #3970)
  * **AI Knowledge:** Default models are now displayed in AI parameters ([#3671](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/3671))
  * **AI Knowledge:** Added OCR support in analysis mode for per-document configuration
  * **AI Knowledge:** Falls back properly to recursive splitting when functional splitting is unavailable (e.g., when using Content Analysis mode) ([#3906](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/3906))
  * **AI Knowledge:** In MCP settings, tools are now shown as enabled by default and disabled tools are correctly enforced as disabled ([#3570](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/3570))
  * **Studio / AI Knowledge / AI Store / AI Insights / AI Collection:** Workspace and agent photos are now automatically compressed or resized to reduce load size ([#3531](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/3531))
  * **Crawler:** Added support for offline dependencies for spaCy and YOLOX in unstructured mode

  **Bugfixes:**

  * **prismeai-workspaces:** Fixed a git versioning error ("Please tell me who you are") when pulling for the first time without a saved local state, caused by missing default `user.name` and `user.email` values
  * **prismeai-workspaces:** Fixed workspace bulk import from a git platform repository: workspace enumeration did not take the configured `*_DIRPATH` variable into account
  * **AI Store:** Fixed an issue where the agent creation button was displayed but inactive for basic users without agent creation permissions
  * **prismeai-crawler:** Fixed a numba exception (`cannot cache function 'projection_by_bboxes': no locator available for file`) when starting the crawler as a user other than the default `api` user (UID 1000), as in OpenShift environments (Internal Ticket #4019)
  * **AI Knowledge / AI Store:** Display an error when the assistant reaches the tool call limit and can no longer respond (Bedrock-specific, as it does not support `tool_choice: none`) ([#3749](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3749))
  * **AI Store:** Fixed copying messages that contain block lists ([#4033](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/4033))
  * **Studio:** Fixed SSO button layout when labels are very long ([#4038](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/4038))
  * **AI Store:** Reduced paragraph line spacing in agent messages ([#4037](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/4037))
  * **Runtime:** Fixed a crash on empty JSON responses when the `Content-Type` is `application/json` ([#4046](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/4046))
  * **AI Knowledge:** Fixed an issue where the selected model was not persisted in testing tab.
</Update>

<Update label="v26.03.101144" description="Releasing version v26.03.101144">
  **Features & Improvements:**

  * **AI Knowledge & AI Store:** Filters can now be applied on both categories and tags ([#3916](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/3916))
  * **AI Knowledge:** Tags can now be edited from the project parameters ([#3916](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/3916))
  * **prismeai-runtime:** Introduced a new [text runtime module](https://docs.prisme.ai/products/ai-builder/automations#text) with a recursive `splitText` function.
  * **AI Knowledge:** Removed `@langchain/textsplitters` dependency, replaced with the runtime native `splitText`.
  * **prismeai-functions:** Now includes all npm packages required by the native Prisme.ai AppStore, removing the need for the npm registry ([docs](https://docs.prisme.ai/self-hosting/entreprise/functions#default-nodejs_shared_packages_whitelist_import)).
  * **prismeai-crawler:** Reduced image size from 13.4GB to 4.14GB; dropped experimental support for JS websites, HTML form authentication, and Docling.
  * **prismeai-functions:** Reduced image size from 1.74GB to 954MB; removed `geopandas` and `langchain-community` Python libraries.
  * **Studio / Pages:** On WebSocket error/disconnection, reconnect using the configured `WEBSOCKETS_DEFAULT_TRANSPORTS` variable instead of always falling back to `['polling', 'websocket']`.
  * **AI Collection:** Modified queue behavior to prevent deep nested calls from being throttled by the platform.

  **Bugfixes:**

  * **AI Store:** Resolved an issue where very long texts did not wrap correctly.
  * **AI Store:** Resolved an issue where the "My Agents" filter appeared only for admins.
  * **AI Knowledge:** Resolved an issue where the "My Agents" filter appeared only for admins.
  * **AI Knowledge:** Improved KPI document querying to prevent display issues with high quantities.
  * **Versioning:** Fixed an internal error on some git pull merge conflict edge cases.
  * **Versioning:** Merge conflicts on `.import.yml` files are now automatically resolved, as these files are regenerated on each push.
  * **Helper Agents:** Resolved an issue where categories were not initialized on agent creation ([#3916](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/work_items/3916)).

  **Breaking Changes:**

  * **Webhooks & WebSockets:** Slug-based calling now requires the `slug:` prefix, e.g., `/v2/workspaces/slug:ops-manager/webhooks/user`.
</Update>

<Update label="v26.03.021352" description="Releasing version v26.03.021352">
  **Features & Improvements :**

  * **Runtime:** Introduced a [secrets module](https://docs.prisme.ai/products/ai-builder/automations#secrets) to store encrypted secrets from automation code. Secrets can only be retrieved by the `fetch` instruction.
  * **Events:** The events API now supports workspace slug in addition to workspace ID.
  * **AI Builder:** The automation schedule now displays both UTC time and the local user time based on timezone.
  * **[Native workspaces](https://docs.prisme.ai/self-hosting/install-products#2-products-initialization):** AI Knowledge, AI Governance, and other native workspaces are now bundled in the `prismeai-workspaces` image and can be imported from the Platform workspace UI (super admins only).
  * **[Platform Versioning](https://docs.prisme.ai/products/ai-builder/versioning):**
    * Added filesystem driver support for platform repositories, enabling local directory-based workspace storage alongside Git.
    * Platform pull now executes sequentially with dependency-aware ordering.
    * Platform pull skips workspaces that already have the version tag specified in their `.import.yml`.
    * Platform push skips workspaces with no changes since the last push.
    * Platform pull and push can be triggered directly from the Platform workspace versioning UI.
    * Automatic workspace import on startup via the `AUTO_BUILTIN_WORKSPACES_IMPORT` environment variable.
  * **Imports:** New workspaces now retain the same ID as in the imported archive or repository.
  * **AI Knowledge:** [SharePoint connector](https://docs.prisme.ai/products/ai-knowledge/connectors) is now available, enabling automatic synchronization of documents from SharePoint. Refer to the [official documentation](https://docs.prisme.ai/products/ai-knowledge/connectors) for setup instructions.
  * **OpenAI:** Allow sending custom headers, user ID, and correlation ID to OpenAI-compatible APIs ([#3888](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3888)).
  * **AI Knowledge:** Updated the injected system message for document summaries — it is now injected into the summary chunk sent to the LLM, fixing the `file_summary` tool on Mistral models ([#3888](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3888)).
  * **OpenAI:** Streaming requests now send the correct `Accept` headers, improving compatibility with strongly typed backends.
  * **Redis:** Slightly increased auto-reconnection backoff delay and maximum value to reduce network spam when the Redis server is unavailable for extended periods.
  * **prismeai-functions:** Implement `NODEJS_PACKAGES_WHITELIST_IMPORT` and `NODEJS_PACKAGES_BLACKLIST_IMPORT` [environment variables](https://docs.prisme.ai/self-hosting/entreprise/functions#environment-variables).
  * **prismeai-functions:** Implement `PYTHON_BUILTIN_MODULES`, `PYTHON_PACKAGES_WHITELIST_IMPORT`, `PYTHON_PACKAGES_AUTO_INSTALL`, and `PYTHON_PACKAGES_DIR` [environment variables](https://docs.prisme.ai/self-hosting/entreprise/functions#environment-variables).
  * **Helm prismeai-apps 0.5.6:** Updated the default port for the crawler HTTP debug endpoint from 80 to 3000, as non-root users cannot bind to port 80.
  * **Helm prismeai-apps 0.5.7:** Added the `PYTHON_PACKAGES_DIR` environment variable to the `prismeai-functions` service for Python package persistence.
  * **Helm prismeai-core 1.9.5:** Enabled automatic creation of master keys for the new secrets module. Upgrade is highly recommended to use the SharePoint MCP or SharePoint Connector within AI Knowledge.

  **Bugfixes:**

  * **AI Knowledge:** Fixed the default analysis mode (background analysis) not being pre-selected in the input, which could lead to ambiguity.
  * **AI Store:** Resolved an issue that prevented swapping threads on an edited message.

  **Breaking Changes:**

  * **prismeai-functions (Python):** Python functions are now forbidden from using the following stdlib modules by default: `subprocess`, `os`, `sys`, `shutil`, `socket`, `ctypes`, `multiprocessing`, `_thread`, `signal`, `resource`, `importlib`, `runpy`, `code`, `ast`, `dis`, `pickle`, `marshal`, `pty`, `fcntl`, `mmap`, `gc`, `inspect`, `webbrowser`, `ensurepip`, `venv`. Use the `PYTHON_BUILTIN_MODULES` environment variable to configure allowed or disallowed modules.
</Update>

<Update label="v26.02.240919" description="Releasing version v26.02.240919">
  **Features & Improvements:**

  * **AI Knowledge:** Added the ability to send custom headers in an agent's webhook ([#3837](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3837))
  * **Runtime:** Webhook HTTP calls now support `workspaceSlug` in addition to `workspaceId`

  **Bugfixes:**

  * **AI Knowledge:** Fixed display of parasitic messages in the activity feed to improve user experience ([#3641](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3641))
  * **AI Knowledge:** Resolved an issue that prevented editing a document name ([#3850](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3850))
  * **AI Knowledge:** Resolved an issue where the chat input was not automatically cleared after sending a message to an agent
  * **Dialog Box:** Resolved an issue where the chat input content was not cleared after sending a message on Firefox
</Update>

<Update label="v26.02.171403" description="Releasing version v26.02.171403">
  **Features & Improvements:**

  * **Crawler:** Initialize documents index mapping to avoid fields mapping explosion due to PDF/OXML metadata extraction (Internal Ticket #3861)
  * **AI Knowledge:** Added Project Manager, allowing project admins to clone, export, and import AIK projects
  * **AI Knowledge:** Replace `langchain/text_splitter` package with the newer `@langchain/textsplitters`, fixing split chunks occasionally exceeding the configured max chunk size
  * **Workspaces:** Automatically remove security rules with reason `REMOVE_ON_DUPLICATE` on workspace duplication
  * **[Readiness API](/self-hosting/operations/testing):** New `/v2/readiness` configuration check that verifies all database connections from all services (core + apps), HTTP connections between internal services, and returns detailed errors
  * **Helm chart prismeai-apps 0.5.2:** Expose crawler debug and `/sys/readiness` HTTP endpoints behind a Kubernetes service
  * **Helm chart prismeai-core 1.9.3:** Allow `securityContext` configuration for all pods ([#1287](https://gitlab.com/gogowego-devsecops/dev/pyramide/-/issues/1287))
  * **Helm chart prismeai-apps 0.5.3:** Allow `securityContext` configuration for all pods ([#1287](https://gitlab.com/gogowego-devsecops/dev/pyramide/-/issues/1287))

  **Bugfixes:**

  * **AI Knowledge:** Resolved an issue where sending a file to a model without a configured limit resulted in an infinite loop
  * **AI Knowledge:** Fixed Serper integration with "news" and "search" types
  * **AI Knowledge:** Resolved an issue that prevented the project users list from displaying when a member had incomplete or empty tags
  * **AI Knowledge:** Fixed an issue where the `/documents` endpoint didn't set the `includeContent` parameter to its default value (`false`)
  * **SSO:** Fixed redirection towards studio URL instead of the expected agent page when connecting with SSO directly on the agent

  **UI:**

  * **AI Builder:** Re-added the edit menu on cards on the home page
</Update>

<Update label="v26.02.091102" description="Releasing version v26.02.091102">
  **Features & Improvements:**

  * **Crawler:** Add 7-day TTL (configurable with `REQUESTS_QUEUE_TTL`) to `*:requests` Redis keys (Internal Ticket #3254)
  * **Crawler:** Allow configuring HTTP cookies via the `Cookies` header (Internal Ticket #3718)
  * **Searchengine:** Validate cookie header format
  * **API Gateway:** Add `createdAt` and `updatedAt` fields to users, along with `createdBefore` and `createdAfter` query parameters on POST /contacts API for user statistics (Internal Ticket #3571)
  * **AI Knowledge:** Optimize project loading performance

  **Bugfixes:**

  * **Vector Store:** Fix `dropIndex` with OpenSearch
</Update>

<Update label="v26.02.030912" description="Releasing version v26.02.030912">
  **Features & Improvements:**

  * **Crawler:** Always remove previous crawler after changing crawler app config (`crawlerId` or main `websiteURL`)
  * **Crawler:** Add memory debug HTTP endpoint for memory profiling (opt-in, local only)
  * **Crawler:** Remove `skipped_urls` from memory offload to Redis
  * **Crawler:** Close headless browser to avoid memory hogging
  * **Workspaces:** Raise error when attempting to import a bulk workspaces archive to a specific workspace
  * **Git Versioning:** Implement platform-wide repositories
  * **Git Versioning:** Prevent pushes when remote branch contains commits not pulled by the platform yet
  * **Git Versioning:** Gracefully merge remote commits with platform commits, or raise merge conflict errors when manual merge is required
  * **Git Versioning:** Add real-time notifications during pull/push operations
  * **Git Versioning:** Workspaces cannot be updated during a pull, push, or manual merge conflict resolution
  * **Git Versioning:** Add `dirpath` config option to repositories to specify a root subdirectory
  * **Git Versioning:** Automatically initialize branches if they do not exist yet
  * **Git Versioning:** Raise error on pull and imports if the pulled folder does not contain any `index.yml`

  **Bugfixes:**

  * **Crawler:** Fix `save_crawlers_stats` crash
  * **Crawler:** Fix Docling integration
  * **Code Interpreter Tool:** Fix an edge case where messages could be misformatted when other tools had been executed before
</Update>

<Accordion title="January 2026 Releases">
  <Update label="v26.01.271346" description="Releasing version v26.01.271346">
    **Features & Improvements:**

    * **Crawler:** Add `DEFAULT_ENGINE_WEBPAGES_LIMIT` environment variable
    * **Crawler:** `initializeCrawler` can now update `webpages_limit`
    * **Crawler:** Support new `headers` option in `websites_config` to configure headers sent to the target domain
    * **Crawler / Searchengine:** Add new `REDIS_TLS_CA_FILE` environment variable
    * **AI Knowledge:** Initialize crawlers with a 500,000 `webpages_limit` (instead of the default 30,000 limit)
    * **AI Knowledge:** Support sending `workspace` API key for image generation tool on Vertex and AWS APIs, allowing non-owners to use it

    **Bugfixes:**

    * **Vector Store (ES/OS):** Fix deleting existing chunks when "Replace existing doc" option is enabled upon reindexing a document
    * **PostgreSQL:** Fixed inserting question mark characters (?) inside JSONB columns (#3690)

    **UI:**

    * **Platform:** Fixed links color on the home page
    * **Block Functions:** UI improvements
    * **AI Store:** Removed the deprecated "date" app
  </Update>

  <Update label="v26.01.201500" description="Releasing version v26.01.201500">
    **Features & Improvements:**

    * **Prismeai-workspaces:** Add a default security workspace rule allowing editors to test automations
    * **Vector Store/OpenSearch:** \_search query to OpenSearch do filters the embedding in order to save space ([#3441](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3441))
    * **AIG:** getUserCustomPlatformRole added to excludes by default in zip and git imports
    * **Automations:** (prismeai-runtime, prismeai-workspaces) Support for dynamic path parameters in webhook endpoints, enabling RESTful API patterns like /v1/users/:id directly in Prisme.ai automations
    * **AIK/AIS:** added `${user}` context variable to consider user firstname, lastname and metadatas in prompt

    **Bugfixes:**

    * **Prisme.ai Crawler:** Improve stability once memory threshold are hit
    * **Vector Store:** Fix nearestChunks query method on OpenSearch queries to improve RAG on context only ([#3617](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3617))

    **UI:**

    * **All Products Home:** New filtering inputs to order products by owners and tags
    * **AI Store:** Fix display datasource on agent home ([#3655](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3655))
    * **Block-Functions:** Fix an issue when deleting a function wasn't saved automatically ([#3466](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3466))
    * **AI Knowledge:** lowered limit when querying documents on OpenSearch simultaneously to avoid launching too many actions ([#3644](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3644))
    * **AIK:** Fixed issue when uploading twice the same doc with "replace" option was not replacing but creating a new doc
    * **AIK:** Removed frequency testing input that was a deprecated feature
  </Update>

  <Update label="v26.01.130922" description="Releasing version v26.01.130922">
    **Features:**

    * **Builder, AI Knowledge, AI Store, AI Insights, AI Collection:** Workspaces and agents can now be pinned to the top of the product home page

    <Frame>
      <img src="https://mintcdn.com/prismeai/QSxvr4LnP-8kbZIk/gifs/pinProducts.gif?s=c504fe0214c7541ffcceaf1819342469" alt="Pin Workspaces and Agents" width="800" height="395" data-path="gifs/pinProducts.gif" />
    </Frame>

    * **Builder, AI Knowledge, AI Store, AI Insights, AI Collection:** Tag suggestions in the agent/workspace creation form

    <Frame>
      <img src="https://mintcdn.com/prismeai/QSxvr4LnP-8kbZIk/gifs/labelList.gif?s=83645b33e84e4f498e335b967be16bf3" alt="Tag Suggestions" width="800" height="393" data-path="gifs/labelList.gif" />
    </Frame>

    * **AI Store:** You can now download markdown canvas content as PDF files

    **Bugfixes & Improvements:**

    * **AI Store:** Suggested questions no longer block the input ([#3626](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3626))
    * **AI Knowledge:** Invitation email is now sent based on the role given when sharing an agent ([#3622](https://gitlab.com/gogowego-devsecops/clients/support-prisme.ai/-/issues/3622))
    * **API Gateway:** Support auto-limit setting when reaching /contacts API, even for Super Admin. This fixes an issue where Super Admins could see users as "undefined undefined" in share modals.
    * **Includes also important bug fixes and hardening improvements.**
  </Update>

  <Update label="v26.01.080927" description="Releasing version v26.01.080927">
    **UI Improvements & Bugfixes:**

    * **Studio:** Fixed opening and closing bugs
    * **AI Store:** Improved agent home chat page view and spacing

    **Features:**

    * **Studio, AI Knowledge, AI Store, AI Insights, AI Collection:** New agent creation workflow featuring the initial integration of helper agents within the interface

    <Frame>
      <img src="https://mintcdn.com/prismeai/rsKacukd7Jy-n4fl/gifs/generateAgentInfos.gif?s=4f7cc1d1fc576ec0e289f7241ead778d" alt="New Agent Creation with Helper Agents" width="800" height="386" data-path="gifs/generateAgentInfos.gif" />
    </Frame>

    * **AI Store:** Agents can now be called using @mentions in chat

    <Frame>
      <img src="https://mintcdn.com/prismeai/rsKacukd7Jy-n4fl/gifs/mentionAgentInChat.gif?s=2ad41d69e7cef61b84e3092f63b1ecf7" alt="Mention Agents in Chat" width="400" height="194" data-path="gifs/mentionAgentInChat.gif" />
    </Frame>
  </Update>
</Accordion>

<Accordion title="December 2025 Releases">
  <Update label="v25.12.290944" description="Releasing version v25.12.290944">
    **Features & Improvements :**

    * **AI Builder :** Stabilize & optimize workspace imports : fix missing / not updated automations, notably reduce S3/object http requests throughput.
    * **Popover :** Authentification with SSO opens a popup and now closes it on successful authentification while reloading the original page.
  </Update>

  <Update label="v25.12.221335" description="Releasing version v25.12.221335">
    ## Configuration changes

    The new Viewer role won't be automatically added to already existing environments since the RBAC configuration is excluded from updates.\
    To manually add this role :

    1. Open the **AI Knowledge** workspace
    2. Open **Settings**
    3. **Advanced**
    4. **Manage roles**
    5. Add a `viewer: {}` line at the beginning, below `editor: {}`
    6. Add these lines at the end of the files, under `rules` section :

    ```yaml theme={null}
        - role: viewer
          action: read
          subject:
            - workspaces
            - pages
        - role: viewer
          action: read
          subject:
            - events
          conditions:
            source.sessionId: '{{session.id}}'
        - role: viewer
          action: read
          subject:
            - events
          conditions:
            source.userId: '{{user.id}}'
        - role: viewer
          inverted: true
          action: read
          subject: events
          conditions:
            type:
              $regex: ^apikeys\.*$
    ```

    ## Bugfixes

    * **AIS :** Send button reappears after an error
    * **AIS :** Agent’s model is now current set in agent home
    * **AIS :** Fix navigating in tools picker with arrow keys while searching
    * **AIK :** Parsing using model send the file as base64 instead of url to support more providers
    * **AIK :** fix wording typos in callAgents option
    * **AIK :** Add new viewer  native role, with read only access to workspace, pages (including the troubleshoot page) and its own user/session events only.
    * **Collections :** Fix count with an undefined query (counts all records)
    * **AIG :** CSS customization now works
    * **AWS :** Fix embeddings with cohere.embed-v4:0 model (output embeddings now are in embeddings.float instead of embeddings)
  </Update>

  <Update label="v25.12.181015" description="Releasing version v25.12.181015">
    **Features & Improvements :**

    * **AI Store :** Conversations history in sidebar rewritten for better performance and stability
    * **AI Store :** Agents now have a home view without creating an empty conversation before first message
    * **AI Store :** High-performance front-end upgrade
    * **AI Store :** Pasting an image into the DialogBox input now uploads it as an attachment
    * **AI Knowledge :** Added support for image and audio upload in RAG
    * **AI Knowledge :** Ability to choose whether a file should be analyzed by a language model\
      This includes new fields for `defaultModels` to set which model analyzes which file type:
      ```yaml theme={null}
      defaultModels:
        embeddings: az-text-embedding-3-large
        completions: az-gpt-5-chat
        images/generations: dall-e-3
        enhanceQuery: az-gpt-5-chat
        toolRouting: az-gpt-5-chat
        parsing:
          image: az-gpt-5-chat
          audio: az-gpt-5-chat
          document: az-gpt-5-chat
      ```
    * **AI Knowledge :** Ability to specify which crawler to use per file
    * **AI Knowledge :** Added possibility to choose parsers with or without OCR

    <Frame>
      <img src="https://mintcdn.com/prismeai/GL4YPDPj9ixiVI9h/gifs/documentprocessingoptions.gif?s=8e8b197c7855b55c11561e437f369fd8" alt="Document Processing" width="1280" height="713" data-path="gifs/documentprocessingoptions.gif" />
    </Frame>

    * **Studio :** Support for multiple SSO providers with customizable display
    * **Studio :** SSO provider is automatically deduced from the user’s email
    * **Crawler :** Reduced default polling interval (1s instead of 5s)
    * **Crawler :** Added ability to override Unstructured OCR configuration through parsers configuration
    * **Crawler :** Allow forcing skip\_ocr=false even if environment defaults set it to true
    * **Crawler :** Adjusted default values for memory throttling

    **UI :**

    * **Studio / Pages :** Fix display of backtick (\`) character in input fields by switching to Fira Code font
    * **Studio / Pages :** Language selector added to user menu

    <Frame>
      <img src="https://mintcdn.com/prismeai/j7BKU8cjW_6NRgMP/gifs/newlanguageselector.gif?s=36a5d3ab3ba2a5b673d2f4b68038e3b5" alt="Language Selector" width="1280" height="713" data-path="gifs/newlanguageselector.gif" />
    </Frame>

    * **Studio / Pages :** Updated link colors in news section
    * **AI Store :** Display a proper loading state while streaming images
    * **AI Store :** Agent home no longer creates a conversation until the first user message is sent
    * **Global :** Fix typo in French welcome popup
    * **AI Knowledge:** A popover can now be opened directly from the AI Knowledge project to chat with the agent.

    <Frame>
      <img src="https://mintcdn.com/prismeai/GL4YPDPj9ixiVI9h/gifs/newpopover.gif?s=093110f487f55039d11a5fe680ab78a4" alt="AI Chatbot Popover" width="1280" height="729" data-path="gifs/newpopover.gif" />
    </Frame>

    **Infrastructure & Operations :**

    * **Helm – prismeai-core :** Enable prismeai-events horizontal pod autoscaling by default
    * **Helm – prismeai-apps :** Increase default resource requests and limits to production recommendations
    * **Helm – prismeai-apps :** Add values for apps elastic user/password
    * **Crawler :** Set default Apache Tika server timeout to 10 minutes (instead of 5 minutes)
    * **prismeai-crawler and prismeai-searchengine :** Add ELASTIC\_USER and ELASTIC\_PASSWORD environment variables

    **Security :**

    * **All services :** Dependencies security upgrades

    **Bugfixes :**

    * **AI Knowledge Client :** chat-completion-openai-like now correctly uses config.apiKey when no apiKey is explicitly provided
    * **AI Knowledge Client :** Fix automation argument handling for file inputs
    * **AI Knowledge :** Fix changing the default tool deleting custom tools defined for an agent
    * **AI Knowledge :** Fix buttons overlapping in Admin view datatable for project management
    * **AI Knowledge :** Fix graph average requests per day
    * **AI Knowledge :** Fix tools not being properly parsed in streaming mode with models such as Gemini 2.5 Pro
    * **AI Knowledge :** Fix files from RAG being incorrectly sent to Vertex LLM with File capability when reranking was disabled
    * **AI Knowledge :** Fix nearbyChunks returning empty chunks causing display issues
    * **AI Store :** Fix troubleshoot link on messages
    * **AI Store :** Fix display of empty activity details
    * **AI Store :** Fix reset of pagination in conversations list after navigation
    * **AI Store :** Fix display issues when streaming `<img>` tags
    * **AI Insights :** Fix conversations message count column when using PostgreSQL
    * **Studio :** Fix login on non-validated accounts briefly showing “wrong email or password”
    * **Studio :** Fix refresh of products list on next sign-in after platformRole is updated
    * **Pages :** Fix DataTable block client-side pagination
    * **Workspace :** Fix workspace property not being correctly updated after add/remove from suggestions
    * **Amazon Web Services :** Fix Bedrock Converse response not being fully parsed in classic (non-streaming) mode
    * **Crawler :** Fix aggressive throttling causing slow crawls on large files
    * **api-gateway :** Always initialize users groups to an empty array on signup to avoid mongodb error 'Cannot apply \$addToSet to non-array field' when adding groups

    **Warnings :**

    * **Crawler :** Apache Tika and Unstructured now skip OCR by default
    * **Crawler :** OCR (using Tesseract) provides better accuracy but uses significantly more resources.
  </Update>

  <Update label="v25.12.011029" description="Releasing version 25.12.011029">
    ### Features

    * **Platform:** Implemented passwordless authentication using Azure Entra ID for both PostgreSQL and Redis
    * **api-gateway:** Added a text parameter to the /contacts/search API to search across id / email / firstName / lastName
    * **AI Knowledge:** Supporting options.excludeParameters with Azure providers
    * **AI Governance:** Search input now queries across id / email / firstName / lastName instead of only emails (Internal Ticket #3477)
    * **Helm chart prismeai-core:**
      * Increased default CPU/memory requests/limits to production recommendations
      * Added new maxWorkers runtime value for RUNNER\_MAX\_WORKERS (default: 4)
      * Automatically computes NodeJS --max-old-space-size for all core services (removing the need for NODE\_OPTIONS)
      * Added nodejsOldSpaceSize Helm value to manually set --max-old-space-size per service

    ### Bugfixes

    * **AI Knowledge:** Fixed cases where Canvas content was displayed in the chat in addition to the Canvas
    * **AI Knowledge:** Fixed translated agent categories not displaying correctly on the home page
    * **AI Knowledge:** Fixed non-streaming /query API calls that could incorrectly switch to streaming mode when any tool was invoked
    * **AI Knowledge:** Fixed a crash in the code interpreter for Anthropic models when combined tool calls occurred
    * **AI Governance:** Adding a workspace as a suggestion no longer removes its labels, and removing it only removes the suggestions label
    * **VectorStore App:** Fixed getNearbyChunks returning each chunk twice (Internal Ticket #3484)
  </Update>
</Accordion>

<Accordion title="November 2025 Releases">
  <Update label="v25.11.241042" description="Releasing version 25.11.241042">
    ### Features

    * **AI Builder**: Add sortable attribute to allow sorting DataTable from events
    * **AI Store**: Display a loading while img is streaming
    * **AI Builder**: New TimeList block

    ### Bugfixes

    * **AIC**: Fix — Number of lines in AI Collection were not displaying, depending on a certain height of the Datatable.
    * **Postgres**: Fix \$in query filter between 2 arrays to be a OR search and not a AND
    * **AI Builder**: Fix "Automation not found" error event that was missing in workspace activity feed
    * **AI Store**: Fix markdown display in AgentMessage responses
  </Update>

  <Update label="v25.11.171950" description="Releasing version 25.11.171950">
    ### Features

    * **AI Store**: UX/UI - Canvas can be expanded

    <Frame>
      !\[Canvas Expandable]\(/gifs/Expandable Canvas.gif)
    </Frame>

    * **AI Store**: UX/UI - Input is centered on new conversation
    * **AI Store**: UX/UI - Enhanced Chat welcome view with question suggestion
    * **AI Store**: UX/UI - Mobile use header for user-space and sidebar
    * **AI Store**: UX/UI - Mobile view of products use enhanced sidebar and layout
    * **AI Store**: UX/UI - Updated some Icons look
    * **AI Store**: UX/UI - Look and feel of DialogBox
    * **AI Store**: UX/UI - Updated Canvas button and Canvas look
    * **AI Store**: UX/UI - Enhanced DialogBox loading time and reduced size of downloaded JS bundle
    * **AI Store**: UX/UI - DialogBox builtin block
    * **AI Store**: UX/UI - AgentMessage block in DialogBox displaying live activity, sources and images

    <Frame>
      !\[Activity Status]\(/gifs/activity status.gif)
    </Frame>

    * **AI Store**: UX/UI Display isTyping in AgentMessage
    * **Studio**: UX/UI - New look for main sidebar

    <Frame>
      !\[New Sidebar]\(/gifs/New Sidebars Motions.gif)
    </Frame>

    * **AI Knowledge**: Refactor of the analytics dashboard and fixes on the displayed data.
    * **AI Knowledge**: New `${datetime}` prompt variable added to retrieve the full date+time with ISO8601 format
    * **prismeai-crawler**: Implement [Apache Tika](https://tika.apache.org/) parser
    * **prismeai-crawler**: Make documents parser configurable with DOCUMENTS\_DEFAULT\_PARSER crawler environment variable
    * **Crawler**: Allow configuring custom parsers config from individual document indexing request
    * **Crawler**: Allow configuring default documents parser with searchengine root parsers config only (without specifying any website\_config for a specific domain)
    * **AI Knowledge**: Add a new parsers configuration option in advanced settings to select between tika or unstructured
    * **AI Knowledge**: New `promptCache.system` option available in modelsSpecifications, only available with some Bedrock models

    ### Bugfixes

    * **AI Knowledge**: Fix analytics loading is now properly synchronized with filters.
    * **AI Knowledge**: Analytics filters are persisted
    * **Functions**: Fix timeout handling, let others functions finish executing properly, and terminate the worker only \${timeout}ms later
    * **AI Store**: UX/UI - Fixes on long assistants names
    * **AI Store**: UX/UI - Hide model picker in mobile view
    * **AI Store**: UX/UI - Fixes on iOS mobile view
    * **AI Store**: UX/UI Fix Scrolling down
    * **AI Store**: UX/UI Fix scroll when large chunk of content received at once
    * **AI Store**: UX/UI Fix cases where activity from tool calling have empty content

    ### Breaking changes

    * **AI Knowledge**: Breaking change: `${date}` prompt variable now contains only YYYY-MM-DD & not the hour/minute/second/microsecond anymore
  </Update>

  <Update label="v25.11.071634" description="Releasing version 25.11.071634">
    ### Bugfixes

    * **AI Knowledge**: Fix custom embeddings setting: was missing in Projects collection schema
    * **AI Store**: Update visual bubble for user messages (Internal ticket #3121)
    * **AI Store**: Tags field now trigger onChange when deleting a tag, and when closing the selection dropdown instead of every tag selected.
  </Update>

  <Update label="v25.11.061024" description="Releasing version 2**5.11.061024">
    ### Features

    * **AI Knowledge**: Breaking changes ⚠️: Default Prompt optimisation. The recommended default prompt for agents is now:

    ```yaml theme={null}
    prompt:
      keyword: ${context}
      default: |- 
        You are a helpful and knowledgeable assistant.

        Use the available tools to answer the user's query as accurately as possible.
        The documents_rag tool is very important as it allows you to search through a knowledge base built for you.

        If the context contains relevant information, base your answer on it and make clear references to it.
        If the context is incomplete or missing, answer to the best of your knowledge but clearly state that the information is not in the provided context.
        Never invent or fabricate details that cannot be supported by either the context or your prior knowledge.

        Keep answers clear, concise, and directly relevant to the query.

        The current date is : ${date}
    ```

    * **AI Knowledge**: Support domain to overwrite azure api domain. Default to openai.azure.com, can be changed to cognitiveservices.azure.com for example for dall-e-3
    * **Backends**: Implement configurable contexts size limit, only emit errors for now and do not prevent set from executing

    ### Bugfixes

    * **AI Governance**: Fix on the display of team members (Internal ticket #3186)
    * **AI Governance**: UI updated on teams management for an user
    * **AI Knowledge**: Pass content\_type & filename to unstructured to help detecting document types. This allows returning proper 'Not a valid XLSX file' error for invalid xlsx (& others) files, instead of the previous 'Partitioning is not supported for the FileType.UNK file type.'
    * **AI Knowledge**: Fix Permission denied error on huggingface cache directories mkdir with restricted user/permissions
    * **AI Knowledge**: Tags can be specified for a user or a group, which will only show document with these.
    * **AI Knowledge**: Fixed Docling not working properly with HTML content
    * **AI Knowledge**: Remove NULL characters from crawled text content as they are invalid in SQL queries (and by extension incompatible with Postgres Prisme.ai platforms) (Internal Ticket #3397)
    * **AI Knowledge**: Fix an issue with tool calls while changing models in the same conversation.
    * **AI Knowledge**: Improved default tools description, mainly to avoid contradicting themselves. You might want to update these manually on your own deployment.
    * **AI Store**: Tags and tools stay selected for a conversation
    * **AI Store**: Fix scroll issues for long canvas
    * **Backends**: Make sure we recreate workers (i.e after a crash) with the same name they had to avoid having 2 workers with the same name
    * **Backends**: Make sure we cant get stuck 120s (or GRACEFUL\_SHUTDOWN\_TIMEOUT) when trying to close the same worker twice at the same time
    * **Backends**: Cleanly respond TerminatedWorker error to all running tasks/http requests when their worker is terminated (because of a crash or any other reason)
    * **Backends**: Fix runtime startup: Do not start processing requests until all workers are ready.
    * **Backends**: Fix worker recreation (i.e after a crash): Do not start processing requests until worker is ready
    * **Builder**: Fix displaying png shortcuts (Internal Ticket #3395)
    * **Dialog Box**: Updated style, prompt library compatible with new input, support custom tool/datasource icons, fixed dark theme issues
  </Update>
</Accordion>

<Accordion title="October 2025 Releases">
  <Update label="v25.10.281531" description="Releasing version 25.10.281531">
    ### Bugfixes

    * **Backends**: Fix unhandledRejections to be logged with pino on a single line  (Internal ticket #3178)
    * **AI Knowledge**: Improved session ID management for certain MCP servers.
    * **AI Knowledge**\*: Vertex Generated image are now store in a private bucket
    * **AI Insights**: New version released — introducing a unified data table to simplify agent control and prepare upcoming releases with even more KPIs. Stay tuned!
    * **UI Enhancements**: Several interface improvements across products for smoother usability and visual consistency
    * **AI Store**: Ability to disable public sharing of agent and conversations
  </Update>

  <Update label="v25.10.241424" description="Releasing version 25.10.241424">
    ### Features

    * **Builder :** Sidebar enhanced to provide more space on smaller laptops.

    ### Bugfixes & Improvements

    * **prismeai-workspaces with PostgresSQL :** Fix scanner\_yyerror postgresql exceptions when importing pages with no labels defined (this sometimes resulted in an invalid SQL insertion query with no value at all between 2 “,”)
    * **prismeai-workspaces :** Force appInstances update during imports even if there was no difference between previous & updated dsul in order to force underlying apps to execute 'onConfigured' automation (and easily repair broken platforms
    * **Collections :** When applying an invalid collection schema, keep previous schema in memory (if available) so the collection stay usable despite the update error
    * **Collections :** Fix creating indexes with PostgreSQL & columns named like a reserved keyword as 'from'
    * **Collections :** Save collections schema error in database & return these detailed errors on collection calls (instead of the basic 'Unknown collection error')
    * **Collections :** Prevent updateSchema from executing twice at the same time since it would raise concurrency issues & make collection updates/creation fail
    * **Collections :** Prevent from reloading the same namespace collections twice at the same time for better performances (instead wait for the previous reloading to finish).
    * **Collections:** Fix concurrency issues when reloading updated schemas while persisting other schemas update to database
  </Update>

  <Update label="v25.10.211604" description="Releasing version 25.10.211604">
    ### Bugfixes

    * **api-gateway**: Fix runtime variable session.id which was set to 'at:undefined' when executing webhooks with a user AccessToken instead of the expected session.id 'at:tokenId'

    ### AIK

    * Fix agent tags list french translation (was always shown in english)
    * Do not raise validation error on /projects API if received data include null fields (as it can be the case for uninitialized fields since ORM migration)

    ### AIS

    * Do not raise validation error during agent settings update when some uninitialized settings are null

    ### Collections

    * Fix shared app collections ; isolate data with the currently running workspaceId instead of the first workspace which initiated the shared collection
  </Update>

  <Update label="v25.10.201550" description="Releasing version 25.10.201550">
    ### Bugfixes

    * **prismeai-workspaces :** Add MongoDB index to workspaces createdAt / updatedAt fields for CosmosDB compatibility (Cosmos require an index for all column which needs sorting capabilities)
    * **api-gateway (Azure AD SSO) :** Missing link between Azure and newly introduced ORM - Internal ticket #3356
    * **AIK :** Fixed tool calling with Vertex AI Gemini. - Internal ticket #3349
  </Update>

  <Update label="v25.10.171028" description="Releasing version 25.10.171028">
    ### Bugfixes

    * **Collections :** Fix collection updateSchemas error events not appearing in source workspace activity feed
    * **Collections :** Make sure we don't keep in memory invalid collection schemas as it would make other collections updates/loading fail
  </Update>

  <Update label="v25.10.162120" description="Releasing version 25.10.162120">
    ### Bugfixes

    * Collection: Update indexes in AIK, AIC, AII to handle edge-cases
    * Runtime: Security on some update collection's index operation
  </Update>

  <Update label="v25.10.161126" description="Releasing version 25.10.161126">
    ### Features & Improvements

    * **Backends:** All database queries refactored using a new **ORM package** compatible with both **MongoDB** and **PostgreSQL**.
    * **Collections:** Refactored using new **`run`** instruction and native ORM (MongoDB + PostgreSQL).\ :warning: [Some breaking changes detailed here](https://docs.prisme.ai/apps-store/marketplace/collection#release-note%3A-collection-with-postgresql)
    * **AII:** Usernames are **pseudonymised** in conversations; user IPs and IDs have been removed.
    * **AIK:**
      * Admin now includes new fields: *Category*, *Documents count*, and *Created at*.
      * If `nearestChunks` has no default value in AIK config, it now defaults to `1`.
      * Manual tag selection in AI Store always takes precedence over automatic self-query.
      * Enhanced tags and tools selection, which persist and reload when changing conversation.
      * Usage Analytics now display total number of tool calls and per-tool call count.
      * Improved suggested questions to avoid language mismatch.
      * **Web Search:** Prompt now defined only in AIK config. (Internal ticket #3290)
      * **Compression Tool:** Improved prompt and contextual flow for Gemini 2.5. (Internal ticket #3290)
      * LLM now uses an **agentic flow** leveraging *tool calling* capabilities. (Internal ticket #3180)
      * Added security when a project AIK is not found. (Internal Ticket #3268)
    * **Pages:**
      * Form block date fields now take user timezone into account.
      * New look for **DataTable block**.
      * Home screen now allows html rich texts (Internal ticket #3285)
    * **Dialog Box:** New UI for **Tools** and **Datasources**.
    * **Documentation:**
      * Launch of **Prisme.ai Academy (v0)**.
      * Added documentation for **PostgreSQL driver**.

    ### Bug Fixes

    * **AIK:** Fixed edge cases with undefined `nearestChunks`.
    * **AIS:** Fixed minor issue with model cost display in chat window. (Internal Ticket #3291)
    * **Builder:** Fixed crash when form fields were set to `null`.
  </Update>

  <Update label="v25.10.021420" description="Releasing version 25.10.021420">
    ### Features

    * **console** & **pages** : Monaco Editor (code editor) is now loaded from studio host instead of public cdn.
  </Update>
</Accordion>

<Accordion title="September 2025 Releases">
  <Update label="v25.09.301703" description="Releasing version 25.09.301703">
    ### Features

    * **API Gateway**: New PATCH bulk endpoint for `/users`. Allow bulk modifications of users. Pretty useful when you want to make a set of users part of a given group.

    ### Bug fixes

    * **Block Dialog Box**:
      * Improve stability of canvas by deleting outdated debounced update for MD and Code Editor.
      * Update Milkdown dependency.
  </Update>

  <Update label="v25.09.261245" description="Releasing version 25.09.261245">
    ### Bugfixes

    * **Builder :** Fix creating corrupted automation/appInstance/page when their slug ends with a whitespace
    * **Builder :** Clicking on output block does not copy its value anymore
    * **api-gateway :** Fix returning custom claims from Azure AD auth provider
    * **api-gateway :** Keep runtime user contexts synchronized with latest SSO authData during authentication
    * **prismeai-runtime :** Fix user switching (with set on user.id) not working on some edge cases where given userId had corrupted data in Redis contexts cache
    * **prismeai-workspaces :** Fix UPLOADS\_DEFAULT\_VISIBILITY environment variable
    * **AIK :** Fix duplicate documents on OpenSearch when creating document with "upsert" option by API

    ### Features & Improvements

    * **AIK :** Can disable rerank/contextual compress per agent
    * **AIK :** Summary / web search don't rerank/compress
    * **AIK :** Added support for Google Search as a provider in the Web Browsing tool
    * **Docs :** Added configuration option for ElasticSearch as a Vector Store
    * **Docs :** Improved configuration process for third-party models on Vertex
    * **prismeai-crawler :** New [ROBOTSTXT\_OBEY](https://docs.prisme.ai/self-hosting/entreprise/crawler-searchengine#advanced-configuration) environment variable
    * **App Crawler :** Make webhook timeout configurable to account for ES/OS clusters specific config  (defaults to 3s)
    * **AIK :**  Improve **Web Browsing Tool** default prompt to enforce HTML tags for quotations
  </Update>

  <Update label="v25.09.151922" description="Releasing version 25.09.151922">
    ### Bugfixes & Improvements

    * **AIK :** Fixed a warning about chunk token estimations triggering on project save way below the limit
    * **AIK :** Fix chunks sent by tool, without a tool key, causing a crash of the query
    * **AIK :** Don't rerank and filter chunks if there is less than 4 chunks
    * **AIK :** Image generation bypass rerank/filter and contextual compression
    * **Builder :** Fix updating appInstances config when updating a workspace secret/config they depend on
    * **Builder :** Fixed accessing workspaces whose index file had been deleted from S3 storage following a desynchronization with the database. Automatically recreate these workspace index files on imports.
    * **Backends :** Now always use standard pino logger for event validation errors in order to properly format json errors in a single line
    * **Runtime :** Fix 'Could not fetch user' runtime errors on invalid `GET /me` requests although API\_URL were correctly configured

    ### Features

    * **AIK :** Send RAG files directly to the LLM as base64 if the model support it
    * **Runtime :** Added utilitarian function deepmerge . Can be used to deepmerge two objects.
  </Update>

  <Update label="v25.09.101738" description="Releasing version 25.09.101738">
    ### Bugfixes & improvements

    * **AIK**: Fix warning on saving agent if only maxContext is provided but not maxResponseTokens
    * **AIK**: fixed an issue on Vector store when using Elastic Search (index.knn)
    * **Crawler**: Increased default max length
    * **Crawler**: Added more mimetypes
    * **Crawler**: Made "ROBOTSTXT\_OBEY" configurable with a .env variable
  </Update>
</Accordion>

<Accordion title="June 2025 - v25.06.251802">
  ### Platform Enhancements

  * **Elasticsearch Vector Store**: Full compatibility with Elasticsearch for vector operations
  * **Optimization**:
    * JWT caching for 2 minutes instead of regenerating on each request
    * Fetch all contexts in single mget redis command
    * Full user details caching with /me API synchronization

  ### Runtime Improvements

  * **HTTP Proxy Support**: Fixed fetch instruction HTTP(S)\_PROXY variables support
  * **Event Processing**: Dropped support for high-load events in wait instruction
  * **Error Handling**: Better redis disconnection error handling during graceful shutdown

  ### Crawler Enhancements

  * **Request Tuning**: New environment variables for precise request polling rate control:
    * CONCURRENT\_REQUESTS
    * CONCURRENT\_REQUESTS\_PER\_DOMAIN
    * DOWNLOAD\_DELAY
    * REQUEST\_QUEUES\_POLLING\_INTERVAL
    * REQUEST\_QUEUES\_POLLING\_SIZE
</Accordion>

<Accordion title="May 2025 - v25.05.131105">
  ### New Applications

  * **Notion App**: New integration available in App Store
  * **Google Cloud Storage App**: Enhanced cloud storage capabilities

  ### AI Knowledge Features

  * **Reindex API**: New /reindex endpoint for document reindexing
  * **Tool Routing**: New prompt setting for enhanced tool routing
  * **Date Integration**: Current date automatically added to tool routing prompts, \${date} keyword support
  * **JSON Output**: Added response\_format.type: "json\_object" for forced JSON output

  ### Agent Permissions

  * **Granular Control**: New publishAgent permission for AI Knowledge agent publishing
  * **Usage Analytics**: Enhanced global usage and costs dashboards with model filtering and time selection

  ### Document Processing

  * **Docling Integration**: New option for efficient document and HTML parsing
  * **Image Extraction**: PDF documents can now return images with proper configuration
</Accordion>

<Accordion title="April 2025 - v25.04.281620">
  ### Security & Authentication

  * **Local Signup Control**: New DISABLE\_LOCAL\_SIGNUP environment variable
  * **SSO Fixes**: Resolved occasional SSO authentication failures

  ### AI Knowledge Improvements

  * **Batch Processing**: Added batchSize option to embeddings automation (default: 96 texts per request)
  * **Temperature UI**: Improved temperature setting interface
  * **Timeout Handling**: File processing now continues in background when exceeding timeout
  * **Analytics**: Renamed Users metric to Sessions, added Users metric

  ### Builder Enhancements

  * **Regular Expressions**: regexp() can now be used with replace() function
  * **Error Handling**: Enhanced fetch error events with response headers
  * **Secrets Management**: Fixed issue with removing some secrets
  * **Form Validation**: Improved SchemaForm validators with custom messages

  ### Reliability

  * **Rate Limiting**: Applied MAXIMUM\_SUCCESSIVE\_CALLS limits across automation HTTP calls
  * **Style Tags**: Allow `<style>` tag in RichText blocks without allowUnsecure option
  * **Message Processing**: Fixed handling of user messages longer than maximum embedding context size
</Accordion>

<Accordion title="March 2025 - v25.03.261523">
  ### New Major Feature: Canvas with AI Collaboration

  #### For Non-Technical Users

  * **Rich Content Editor** with an intuitive interface
  * **Intelligent Context Menu** allowing you to:
    * Generate short or long responses
    * Translate content into multiple languages
    * Finalize and polish your texts
  * **Export Options** to PDF and Docs formats for easy sharing

  #### For Technical Users

  * **Advanced Code Editor** with syntax highlighting
  * **Specialized Context Menu** offering:
    * Automatic documentation generation
    * Unit test creation
    * Code review
    * Adding logs
    * Code porting to other languages (Java, Python, etc.)
  * **Real-time Web Code Preview**:
    * Form rendering
    * Chart generation
    * Various visualizations
  * **Direct Export** of generated code

  ### Multi-modal Support for End Users

  * Ability to attach images or audio files (if the model supports multi-modal) to:
    * Query content
    * Get detailed explanations
  * Option to import images directly into Canvas for post-processing

  ### New Administration Options: Tools & Focus

  * Administrators can now:
    * Allow end users to manually activate Canvas and Tools
    * Enable document tag display to target RAG towards specific documents
  * Enhanced error messages in AI Store agents (document support, LLM errors, etc.)

  ### Tool Enhancements

  * **AI Knowledge** - New image generation tool
  * **AI Builder** - Addition of a template with customizable tools
  * **AI Knowledge** - Improved consistency between placeholders and default values in RAG settings

  ### Interface Simplification

  * **AI Knowledge** - Redesigned AI parameter interface to prevent configuration errors:
    * Simplified user interface
    * Temperature with removal of top\_p and penalty parameters
    * Technical users retain full access to advanced parameters via API or Builder

  ### Other Improvements

  * Easier sharing of active conversation during multiple simultaneous threads
  * AIK monitoring updates every 10 minutes (via AI Knowledge and API)
  * Fixed "Chat completion" error at the end of Stream
  * Various security updates and bug fixes
</Accordion>

<Accordion title="Previous releases">
  [Details about previous releases](https://gitlab.com/prisme.ai/prisme.ai/-/releases/)
</Accordion>

## Stay Informed

* Join our [mailing list](https://www.prisme.ai/en/newsletter) to receive update notifications
* Follow us on [LinkedIn](https://linkedin.com/company/prisme-ai) for announcements
* Check this page regularly for the latest information