​

AI Builder

​

How It Works

1. You provide a name and description — Tell the AI what kind of agent you want (e.g., “A customer support agent for a SaaS product that can look up orders and send emails”).
2. Optionally select tools and knowledge sources — Pre-select specific tools or attach existing knowledge sources before generating.
3. AI generates a complete configuration — The scaffold includes a system prompt, recommended model, suggested tools, and knowledge types.
4. Review and edit the preview — Inspect and modify the AI-generated configuration before creating.
5. Configure tool credentials — Set up API keys for tools that need them (optional — can be done later in Agent Studio).
6. Create the agent — One click to create the fully configured agent.

​

Using the AI Builder

​

From the Dashboard

1. Click + New Agent to open the Create Agent modal.
2. Select AI Builder from the sidebar.
3. Fill in the fields:

4. (Optional) Click the Wrench icon to pre-select tools from the tool picker.
5. (Optional) Click the Upload icon to attach existing knowledge sources.
6. Click Build with AI.

​

Quick Action Pills

• Voice Receptionist — Front desk agent that greets callers, routes calls, and takes messages
• Sales Caller — Outbound sales agent that qualifies leads and books demos
• Phone Support Agent — Handles customer support calls with troubleshooting and issue resolution

​

Tool Picker

• Search — Filter by name or category
• Category pills — Filter by category (Search, Web Scraping, Data, Finance, Communication, etc.)
• Checkbox selection — Toggle tools on/off
• Badge count — Shows how many tools you’ve selected

​

Knowledge Source Picker

• Select previously created sources (files, URLs, text, etc.)
• Selected sources are automatically attached to the new agent
• If you haven’t created any knowledge sources yet, you can add them later in Agent Studio

​

Preview Step

​

What’s Shown

​

Actions

• Next: Language & Voice — Open the Language & Voice picker (see below).
• Back to Edit — Return to the builder to change your inputs.

​

Language & Voice Step

• Languages — multi-select chips. Pick every language the agent should detect and respond in (Hindi + English is a common combo). Code-mixing like Hinglish is handled automatically. Speech recognition is picked for you based on this list — Sarvam for Indian languages, Deepgram for English and others.
• Voice — paginated, searchable grid. Your own cloned voices appear first, then Aero, then other providers. Filter by gender, language, and provider. Click any card to select; click the play icon to preview.

​

Credentials Step

• Tool name and icon
• Required credential fields (API keys, tokens, secrets)
• Each field shows “Configure in Agent Studio” as a placeholder

​

Via the API

​

Scaffold an Agent

curl -X POST https://api.thinnest.ai/agents/scaffold \
  -H "Authorization: Bearer $THINNESTAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Customer Support Agent",
    "description": "A support agent for an e-commerce store that can look up orders, process returns, and send confirmation emails",
    "selected_tools": ["gmail", "google_sheets"],
    "selected_datasources": ["url", "file"]
  }'

​

Request Body

​

Response

{
  "scaffold": {
    "name": "Customer Support Agent",
    "system_prompt": "You are a professional customer support agent for an e-commerce store...",
    "model": "openai/gpt-4o",
    "tools": [
      { "id": "gmail", "name": "Gmail" },
      { "id": "google_sheets", "name": "Google Sheets" },
      { "id": "duckduckgo", "name": "DuckDuckGo" }
    ],
    "knowledge_types": ["FAQ documents", "Return policy", "Product catalog"],
    "reasoning": "Based on your description, I recommend..."
  },
  "selected_datasources": ["url", "file"]
}

​

Best Practices

• Be specific in your description — “A booking agent for a dental clinic that checks Google Calendar availability and sends SMS confirmations” produces much better results than “A booking agent.”
• Pre-select known tools — If you know you need Gmail and Google Calendar, select them upfront. The AI will incorporate them into the system prompt.
• Review the system prompt — The AI generates a good starting point, but you know your business best. Edit the prompt to add specific details, tone guidelines, or edge case handling.
• Iterate — If the first generation isn’t perfect, go back and refine your description. More detail = better results.

​

Scaffold Flow

curl -X POST https://api.thinnest.ai/agents/scaffold-flow \
  -H "Authorization: Bearer $THINNESTAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Lead Qualification Pipeline",
    "description": "Qualify inbound leads, enrich their data, score them, and route hot leads to sales"
  }'