AI Agent

The AI Agent is the most powerful feature of Jambo Studio. Unlike the basic chat (/schema, /data, /all), the agent has access to 8 real tools that let it read your schema, create collections, generate content in bulk, translate entries, and even produce AI images — all with a single prompt.

Agent mode vs. Basic mode

Feature	Basic mode (`/schema`)	Agent mode
Schema generation	✅ JSON preview → manual apply	✅ Auto-created + saved
Content generation	3–5 entries per collection	up to 100 entries in one shot
Multi-locale	Manual per locale	auto, all configured locales at once
Images	Manual upload via admin	AI‑generated + auto‑upload
Bulk operations	Not available	✅ Create, update, delete dozens of entries
Schema exploration	Static context string	✅ Real-time `explore_schema()` tool
Translation	Not available	✅ `translate_entries()` tool
Plan preview	Not available	✅ PlanCard with action list before execution

Provider capabilities

The agent automatically detects what your configured AI provider can do and adapts its behavior. A capabilities badge is displayed above the chat input at all times:

🟢 Text (claude-sonnet-4-6)  🟢 Images (dall-e-3)

If a capability is missing, the agent warns you and suggests how to enable it:

⚠️ No image provider configured. I’ll use professional SVG placeholders instead. To get AI‑generated images, add OpenAI or Gemini in Admin → App Settings → AI Providers.

Provider	Text	Images	Notes
OpenAI	✅ GPT-4o, o1, etc.	✅ DALL‑E 3	Full capabilities
Anthropic	✅ Claude Opus / Sonnet	❌	Text only
Google Gemini	✅ Gemini Flash / Pro	✅ Imagen	Full capabilities
Ollama	⚠️ Quality limited	❌	Local model
DeepSeek / Mistral / Groq / xAI / Perplexity / Qwen / OpenRouter	✅	❌	Text only

The 8 agent tools

Each tool is a server-side function the AI can invoke. The agent chooses which tools to use based on your request.

Read tools (always available, read-only)

Tool	Description
`explore_schema`	Returns all collections with their fields, types, required flags, and relations. The agent calls this at the start of every session.
`read_entries`	Reads entries from a collection. Supports pagination and locale/status filters.

Write tools (execution rules)

Tool	Confirmation	What it does
`create_collections`	Auto	Creates collections + fields directly in the database. Includes relation targets and enumeration values.
`create_entries`	Auto	Bulk creates entries. Supports up to 100 entries × N locales in a single call.
`update_entries`	Preview required	Modifies existing entries by UUID. Shows a diff before executing.
`delete_entries`	Preview + Confirm	Soft-deletes entries by UUID. Shows the list of affected entries before confirming.
`generate_images`	Auto	Generates images via DALL‑E or Gemini, uploads them to the media library, and returns UUIDs. Falls back to SVG placeholders if no image provider is available.
`translate_entries`	Preview required	Translates entries from a source locale to N target locales using the AI. Preserves slugs and structured data.

Confirmation rules

Action	Behavior
Create anything	Executed immediately, no confirmation needed
Modify existing data	Preview shown first — click “Execute” to apply
Delete existing data	Preview shown first — click “Confirm” to delete

How to use the agent

1. Simple requests — automatic detection

For small requests (1–2 collections, 3–5 entries), use the standard /schema, /data, or /all commands. The agent will respond in classic JSON preview mode with “Apply” buttons.

/schema Create a blog with posts, categories, and comments
/data Generate 5 professional blog articles in French
/all Build a complete e-commerce schema with products

2. Bulk requests — agent mode

When you ask for bulk operations, the agent switches to plan mode:

Create a full agency website with a portfolio (6 projects),
testimonials (8 quotes), and a blog (5 articles).
Everything in French, English, Spanish, and Arabic.
Add professional hero images for each section.

The agent responds with a PlanCard:

┌────────────────────────────────────────────────┐
│ 📋 Plan                                        │
│                                                │
│ Create agency website with portfolio,          │
│ testimonials, blog — 4 languages               │
│                                                │
│ 1. explore_schema                              │
│ 2. create_collections  3 coll.                 │
│ 3. create_entries     76 entrées × 4 langues   │
│ 4. generate_images     12 images               │
│                                                │
│ [Execute plan]                                 │
└────────────────────────────────────────────────┘

Click Execute plan and the agent runs all actions sequentially, showing an ExecutionLog:

✅ explore_schema: OK
✅ create_collections: 3 collection(s)
✅ create_entries: 76 created, 0 errors
✅ generate_images: 12 images generated (DALL-E)

3. Explore and read

Ask the agent to analyze your project before making changes:

Explore my schema and tell me what's missing for a blog
Read the current testimonials and suggest improvements
What collections should I add for an e-learning platform?

Tool execution endpoint

If you’re building a custom integration, you can call the agent programmatically:

POST /api/projects/{uuid}/studio/ai-execute
Content-Type: application/json
Authorization: Bearer <token>

{
  "actions": [
    { "tool": "explore_schema", "params": {} },
    { "tool": "create_entries", "params": {
        "collection": "testimonials",
        "entries": [{ "quote": "...", "author_name": "..." }],
        "locales": ["fr", "en", "es", "ar"]
    }}
  ],
  "auto_confirm": true
}

Capabilities endpoint

GET /api/projects/{uuid}/studio/ai-capabilities
→ { "text": true, "images": true, "provider": "openai", "model": "gpt-4o", "limits": [] }

Image generation

The agent can generate images when an image-capable provider (OpenAI or Gemini) is configured.

Generate a dark dashboard screenshot for the hero section
Create blog cover images for 3 articles about PHP performance
Add author avatars for the 6 testimonials

When no image provider is available, the agent falls back to professional SVG placeholders with the Jambo emerald color scheme.

Tips

Be specific — “Create 6 testimonials from freelance developers about switching from Strapi” works better than “Add testimonials”
Mention locales — “In French and English” or “In all configured locales” triggers multi-locale generation
Iterate — “Keep the first 3 testimonials, replace the last 3 with better ones”
Explore first — “Analyze my schema and suggest blog improvements” before making changes
Images — Describe the visual you want: “A futuristic dark dashboard with green data panels and analytics charts”