Guides
Guides
Task-oriented walkthroughs for common Featherlane AI workflows.
Guides are task-oriented. Start here when you know what you want to do and need the shortest path through the repo.
Available Guides
| Guide | Use it when |
|---|---|
| Coding-Agent Tool Gates | You want Claude Code, Codex, or OpenCode tool calls checked before execution. |
| Hosted MCP Access | You want employee AI clients to use assigned remote MCP tools through Featherlane AI. |
| Python Agent Frameworks | You want to guard AG2 or Agno local tools and final plain-text output without rewriting agent flow. |
| AI Usage Cost Caps | You need per-caller LLM spend limits, 80% alerts, and hard preflight enforcement. |
| Policy Authoring | You need to write, validate, push, or pull policy YAML. |
| Guard Modes | You need to decide whether an SDK integration should block, rewrite, or regenerate unsafe drafts. |
Common Workflows
Guard An Agent Reply
- Create an agent and runtime key in the Featherlane AI dashboard.
- Install the SDK package for your runtime.
- Decorate the agent object once at construction.
- Keep calling its
reply()method normally. - Use the resulting trace to inspect the event later.
For AG2 and Agno, follow Python Agent Frameworks to protect both exposed local tools and final plain-text output.
Build A Demo
Use the demo package when you want a clean product moment:
agent drafts risky output
-> Featherlane AI checks it
-> user receives only the allowed or rewritten output
-> dashboard shows the traceFor local demos, use pnpm demo:chat or the LiveKit example under
demo/livekit.
Add A New Runtime Capability
Follow the SDK-driven rule:
- Add or update the shared Rust wire type in
tl-core. - Add the Rust server endpoint or behavior.
- Regenerate contracts.
- Expose the capability through Rust, TypeScript, and Python SDKs.
- Update examples and docs in the same change.