Prompt Management

Use the built-in Prompt Library to quickly start conversations with ready-made, expert-crafted prompts — and customize them per organization or per agent.

The Prompt Library gives every user instant access to a curated set of prompt templates right from the Chat home screen. Instead of writing a prompt from scratch, you can browse categories, preview a prompt, and send it in one click.

How It Works

When you open the Chat home screen, category pills appear below the input box. Each pill represents a prompt category. Clicking a category opens a panel with all available prompts in that category.

1. Choose a category

   Click on one of the category pills below the input box (e.g., Write, Learn, Code).

   A floating panel opens with the list of prompts available in that category.

2. Preview a prompt

   Hover over any prompt in the list. The first line of the prompt template is previewed directly in the input box, so you can see what it will send before committing.

3. Select and send

   Click the prompt you want. The full prompt template is loaded into the input box. You can then:

   • Edit the text to add your own context or details before sending
   • Press Enter to send the prompt as-is to the AI agent

Default Categories

The Prompt Library ships with the following default categories, each containing ready-to-use templates. The list is shared per deployment and can be customized through the Category Management API below — translate the labels, add a new pill, or remove one you don’t want.

Write

Professional writing and communication: executive summaries, client emails, meeting notes, business proposals, release notes.

Learn

Knowledge extraction and understanding: explain a concept, summarize a document, compare options, industry briefs, Q&A generation.

Code

Software development assistance: code review, debugging, API design, SQL query building, test case generation.

Coach

Professional guidance and mentoring: constructive feedback, presentation prep, difficult conversations, career strategy, interview preparation.

Brainstorm

Ideation and creative problem solving: idea generation, SWOT analysis, user stories, name generation, risk assessment.

The shipped configuration also includes empty categories ready to be populated: Sales, Support, Marketing, Engineering, and General.

Prompt Scopes

Prompts can live at three different scopes. Users always see the union of the scopes that apply to them.

Scope	Visible to	Editable by	Use case
Global	Everyone on the platform	Platform administrators only	Curated, expert-crafted templates shipped with the product
Organization	All users in that organization	Organization administrators	Org-specific prompts (internal style guides, company processes)
Agent	Users interacting with that specific agent	Agent owners	Prompts that only make sense for one agent (e.g. an HR-onboarding agent’s intake questions)

When a user opens the chat home, they see global prompts + their organization’s prompts. When they open a specific agent that has its own prompts, they additionally see that agent’s prompts.

Global prompts are read-only — organizations and agent owners cannot modify or delete them, only add their own on top.

Prompt Template Structure

Each prompt template includes:

Field	Description
Name	Unique slug used to identify the prompt (e.g. write-executive-summary)
Title	Short, descriptive name shown in the category panel
Description	Brief explanation of what the prompt does
Content	The full prompt text, crafted with expert instructions for the AI
Arguments	Placeholders you can fill in (e.g., topic, context, code)
Category	The category pill it appears under
Tags	Searchable keywords for discovery
Scope	Which agents or organization the prompt belongs to

Some prompts also include tool guidance — they can suggest or force the AI to use specific tools (e.g., web search, knowledge base) to produce better answers.

Initial Setup

The Prompt Library starts empty on a fresh deployment or after a migration. Before users can see the category pills on the Chat home screen, an administrator must seed the library with prompt templates.

Prerequisites

• Access to the Prompt Library workspace in Builder (slug: prompt-library)
• A workspace API key for that workspace (or admin access to trigger automations)

Step 1: Seed the Default Templates

The Prompt Library workspace ships with a built-in automation — Admin/Seed Prompts (seed-prompts) — that creates 25 ready-to-use prompt templates across five categories: Write, Learn, Code, Coach, and Brainstorm.

From Builder

1. Open Builder and navigate to the Prompt Library workspace
2. Go to Automations and find Admin/Seed Prompts
3. Click Run (or trigger it via the activity panel)
4. Optionally pass an agent_id to scope all 25 prompts to a specific agent (see Step 2 below)

The automation creates 5 prompts per category. Running it again on an already-seeded library will attempt to insert duplicates — it is safe to run once per environment.

Via API

You can also trigger the seeding programmatically:

curl -X POST "https://api.{your-env}.prisme.ai/v2/workspaces/slug:prompt-library/webhooks/prompt-library/seed-prompts" \
  -H "Authorization: Bearer <workspace-api-key>" \
  -H "Content-Type: application/json" \
  -d '{}'

To scope the seeded prompts to a specific agent, pass agent_id in the body:

curl -X POST "https://api.{your-env}.prisme.ai/v2/workspaces/slug:prompt-library/webhooks/prompt-library/seed-prompts" \
  -H "Authorization: Bearer <workspace-api-key>" \
  -H "Content-Type: application/json" \
  -d '{ "agent_id": "<your-agent-id>" }'

Replace {your-env} with your deployment domain (e.g. studio.prisme.ai for the cloud environment, or your custom domain for on-premise deployments).

Step 2: Scope to an Agent

By default, seeded prompts are visible to all users on the Chat home screen (organization scope). To show them only when a user is chatting with a specific agent, pass the agent’s ID when seeding — or update individual prompts afterwards:

curl -X PATCH "https://api.{your-env}.prisme.ai/v2/workspaces/slug:prompt-library/webhooks/prompt-library/v1/prompts/write-email" \
  -H "Authorization: Bearer <workspace-api-key>" \
  -H "Content-Type: application/json" \
  -d '{ "agent_id": "<your-agent-id>" }'

Setting agent_id on a prompt makes it visible only when users interact with that agent, and invisible on the general home screen.

Step 3: Verify

After seeding, open the Chat interface and confirm the category pills appear below the input box. If nothing appears:

• Confirm the workspace API key used belongs to the correct Prompt Library workspace
• Check that at least one prompt has "enabled": true in its payload
• Verify the user’s organization matches the scope of the seeded prompts

Example: Using a Prompt Template

Here’s a typical workflow using the Write > Executive Summary prompt:

1. Click the Write category pill
2. Click Executive Summary in the panel
3. The input box fills with the prompt template
4. Replace the placeholder with your topic or paste your document content
5. Press Enter to get a professional executive summary

The AI will follow the template’s expert instructions — in this case, producing a concise summary focused on key findings, business impact, and recommended next steps, in a tone appropriate for a C-level audience.

Agents Using Prompts as Tools

Beyond the Chat home screen, agents themselves can call the Prompt Library at runtime. The library exposes a prompt_library tool that an agent can use to:

• Search for relevant prompts by keywords or category
• Fetch a specific prompt by name
• Apply the prompt’s expert guidance to its own answer

This means even when the user types a free-form question, the agent can decide on its own to consult the library and produce a higher-quality answer following an expert template.

When the agent calls the tool, the search is automatically scoped to:

• The user’s organization prompts
• Global prompts
• The current agent’s own prompts (if any)

The user does not see the tool call directly — they just get a better answer.

For Administrators

The Prompt Library is fully manageable through a public REST API. Each operation is authenticated and scoped to the caller’s organization.

List & search

Paginated listing with filters by category and agent. Returns total count, page, and limit.

Create

Add a new prompt scoped to your organization, optionally pinned to a specific agent.

Update

Edit any field of an organization prompt — title, content, arguments, category, tags, or enabled state.

Delete

Remove an organization prompt. Global prompts cannot be deleted via the API.

Endpoints

Method	Path	Action
GET	/v1/prompts	List prompts (with page, limit, category, agent_id filters)
POST	/v1/prompts	Create a new prompt
GET	/v1/prompts/:name	Fetch a single prompt by name
PATCH	/v1/prompts/:name	Update fields on an existing prompt
DELETE	/v1/prompts/:name	Delete a prompt

All endpoints require authentication. Organization prompts are visible only to members of the same organization; global prompts are visible to everyone but read-only.

Permissions

Operation	Global prompts	Organization prompts	Other org’s prompts
Read (list / get)	Visible	Visible	Not visible
Create	Not allowed (admin-seeded only)	Allowed	Not allowed
Update	Not allowed	Allowed	Not allowed
Delete	Not allowed	Allowed	Not allowed

Use Cases

Internal Style Guide

Create org-wide prompts that enforce your company’s tone, formatting, and terminology in every generated email or document.

Per-Agent Specialization

Pin prompts to a specific support, HR, or sales agent so they only appear when users interact with that agent.

Compliance Templates

Provide vetted templates for regulated workflows (legal disclosures, customer notifications) that users can run in one click.

Bulk Onboarding

Seed a new agent with a curated set of prompts at creation time, scoped to that agent only.

Prompt and category management are entirely driven by the REST API documented on this page. There is no dedicated admin UI today — call the endpoints from your own tooling, scripts, or an HTTP client.

Category Management

The category pills shown on the chat home screen are read from the deployment’s configured category list. Administrators can list them, rename them (for example to translate the default English labels), add new ones, or remove ones they don’t want.

List & inspect

Read the current category list with GET /v1/categories. Each item is a string used both as the slug stored on prompts and as the label rendered in the UI.

Rename

Use PATCH /v1/categories/:value to replace a category by its translated label (e.g. rename write to Écrire).

Add

Use POST /v1/categories with { "value": "<string>" } to append a new category to the list.

Remove

Use DELETE /v1/categories/:value to remove an entry from the list.

Category Endpoints

Method	Path	Action
GET	/v1/categories	List configured categories (returns { items: string[], total })
POST	/v1/categories	Add a new category. Body: { "value": "<string>" }. 409 if already present
GET	/v1/categories/:value	Fetch a single category. 404 if unknown
PATCH	/v1/categories/:value	Rename a category. Body: { "value": "<new>" }. 404 if unknown, 409 if the new value is already taken by another entry
DELETE	/v1/categories/:value	Remove a category. 404 if unknown. 204 No Content on success

Renaming a category with PATCH updates the configured list only — it does not rewrite the category field of existing prompts. To fully localize the library, also update each prompt with PATCH /v1/prompts/:name so its category matches the new value.

Example: translate the default categories

To switch a deployment from the default English labels to French, do it one category at a time so users never see a half-translated UI:

1. Pick one category (e.g. write).
2. PATCH /v1/prompts/:name on every prompt that has category: write to set the new value (e.g. Écrire).
3. PATCH /v1/categories/write to rename the pill itself.
4. Move on to the next category.

If you do all the /v1/prompts updates first and the /v1/categories updates after, there is a window where users click the old pill (write) and see nothing because every prompt has already moved to Écrire. Doing it per category, or during a low-traffic window, avoids that.

# Base URL convention used in all examples below
export PRISME_API_BASE="https://api.studio.prisme.ai/v2/workspaces/slug:prompt-library/webhooks/prompt-library"
# (use https://api.sandbox.prisme.ai/... on the sandbox environment)

curl -X PATCH "$PRISME_API_BASE/v1/categories/write" \
  -H "Authorization: Bearer $WORKSPACE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "value": "Écrire" }'

Permissions

• GET endpoints (list / single) accept any authenticated caller (user session or workspace API key).
• POST, PATCH, DELETE require a workspace API key (Authorization: Bearer <workspace-api-key>). User sessions are rejected with 403, even for admin users, because the configured list is shared across the whole deployment and must be managed by a service credential.
• Concurrent mutations are serialized: a second admin call that arrives while another mutation is in flight is rejected with 429. Retry after a few seconds. In practice this only matters if you script bulk renames in parallel.

The configured list is shared across the whole deployment — every user in every organization sees the same pills.

Privacy & Security

Prompts are stored in a secured library with strict scoping rules:

• A user only sees prompts that belong to their organization, or globals shared with everyone.
• An organization cannot read or modify prompts of another organization.
• An agent’s prompts are only visible when interacting with that agent.

Authentication is required for every prompt operation, both from the chat UI and from the API.