Built-in Tools
Eight tools are registered in the chat runtime, letting the chatbot act inside conversations — not just generate text. Up to 30 sequential tool-use steps per request, with dedicated inline UI components for every tool result.
Web Search
google_search — Serper API
Search the web in real time from inside a conversation. Returns up to 10 results with titles, snippets, and URLs that the AI can reason over.
- Powered by the Serper API (Google search results)
- Returns up to 10 results per query
- AI synthesizes results into a coherent answer with citations
- Inline UI component shows search results as clickable cards
Snippet Search
snippet_search — Exa API
Find specific highlighted excerpts from across the web. Better than full-page retrieval when you need a precise answer to a factual question.
- Powered by the Exa API for semantic web search
- Returns highlighted text excerpts, not full pages
- Ideal for factual lookups, definitions, and quick references
- Inline UI shows extracted snippets with source attribution
Webpage Retrieval
retrieve_webpage — multi-provider fallback
Fetch and read the full content of any webpage. Uses a three-provider fallback chain to maximize reliability, with automatic summarization for long pages.
- 3-provider fallback: Serper → Exa → Jina
- GPT-4.1-mini auto-summarizes content that exceeds context limits
- Handles paywalls, JS-rendered pages, and PDFs where possible
- Inline UI shows the retrieved page title, URL, and content preview
Code Execution
Provider-native sandboxed Python
Run Python code in a sandboxed environment directly from the chat. Supports calculations, data processing, and verification. Enabled per-model through provider-native execution.
- Sandboxed execution via Anthropic, OpenAI, and Google runtimes
- Supports calculations, data transforms, and quick prototyping
- 6 models have code execution enabled
- Results displayed inline with formatted output and error handling
Memory Tool
memory — 5 modes
The AI can persist, search, update, delete, and list memories during conversations. Facts and context are stored per-user and scoped by organization.
- 5 modes: search, create, update, delete, list
- Organization-scoped isolation for multi-tenant setups
- Full-text search with PostgreSQL GIN index and ts_rank scoring
- Mode-specific UI cards: blue (search), emerald (create), amber (update), red (delete), violet (list)
Skills Tool
AI skills retrieval
The model can inspect and retrieve user-authored reusable AI skills from the database. Skills let users define custom behaviors and templates the chatbot can follow.
- Lets the AI look up user-created skill definitions
- Skills defined in frontmatter + markdown format
- Parent-child skill relationships for composability
- Shareable across organizations and teams
Google Workspace
Google Drive, Docs, Sheets, Calendar, Gmail
Live read-only access to a user's connected Google Workspace accounts. The AI can search files, read documents, check calendars, and browse emails mid-conversation.
- Drive: search with filters (MIME, folder, date, starred, trashed)
- Docs: structural API with markdown conversion (headings, lists, tables)
- Sheets: spreadsheet data retrieval
- Calendar: date-range queries, attendee extraction, meeting-link detection
- Gmail: thread grouping, sender/text filtering, full-body retrieval
Microsoft 365
Outlook, Calendar, OneDrive, SharePoint
Live read-only access to a user's connected Microsoft 365 accounts. Same capabilities as the Google connector, but for the Microsoft ecosystem.
- Outlook: thread grouping by conversationId, sender/text filtering
- Calendar: date-range queries, Teams/Zoom meeting-link detection
- OneDrive: file search, metadata lookup, content download with text extraction
- SharePoint: site-level file access with the same parsing pipeline
All 8 tools included and ready to use.
Every tool has a matching inline UI component for displaying results in the chat.
