Tutorials & Example Apps

The Tavora Go SDK ships with self-contained example apps. Each is its own Go module — cd examples/<name> && go run . — and demonstrates a concrete integration pattern you can copy into your own backend.

Walkthroughs

The two walkthroughs below build on each other. Read them in order if it’s your first Tavora integration; treat them as standalone references once you’re past the basics.

1. Build the tasklist agent — the canonical code-first + MCP integration template, end-to-end. Start here if it’s your first integration.
2. Extend the agent with a module skill — add a require()-able JS module that wraps an MCP call + an LLM sub-call in one sandbox step. The “skills vs MCP” decision worked out concretely.

Reference examples

Example	What it shows
chat	Multi-turn agentic REPL — one AgentSession reused across turns.
support-bot	RAG-augmented chat over a documents folder using Conversation + SendMessage.
research-assistant	Single-turn agent with search + list_indexes tools.
knowledge-base	Document upload, index management, semantic search.
tasklist	End-to-end app template — local SQLite app exposes its domain to a Tavora agent via MCP. See the walkthrough.

The TS SDK ships parallel examples under tavora-sdk-ts (ts-chat, ts-chat-browser).

The integrator template: tasklist

If you’re wiring Tavora into a Go app for the first time, start with tasklist. It’s the smallest end-to-end demonstration of the code-first integration pattern Tavora is built around:

1. Your app hosts an MCP server at /mcp exposing your domain tools (in the example: SQLite-backed task lists).
2. You author the agent under tavora/agents/tasklist/ — the agent.jsonc → mcp block declares your /mcp endpoint + the bearer secret name. tavora deploy ships it.
3. The agent writes require('tasklist-example').add_task({...}) inside a think step. Tavora’s MCP client POSTs JSON-RPC tools/call to your /mcp endpoint; your handler dispatches it and returns the result.
4. The result flows back into the sandbox trace, streams over SSE, and renders in your UI.

One think step can fire many MCP calls — the agent doesn’t spend one LLM turn per tool call.

Hosted tools

For deployable CLI tools (interactive chat surface, CI gates, RAG eval suites), see tavora-tools:

• tavora deploy --dry-run — CI validate gate for the local tavora/ folder.
• tavora rag-eval formats --gate — verify the RAG pipeline accepts every supported document format.
• tavora rag-eval judge --gate — LLM-as-judge RAG eval against structured ground truth.
• tavora-tui — interactive Bubble-Tea chat surface against a configured agent.

What’s next

• Quickstart — first code-first agent.
• MCP servers — the integration mechanism the tasklist template uses.
• Skills — require()-able JS modules and prompt skills.