# Use FixLink with AI agents

Give AI assistants structured, safe support instructions instead of letting them invent steps.

Guide: https://usefixlink.com/docs/agent-readable-guides
Agent JSON: https://usefixlink.com/docs/agent-readable-guides/agent.json
Estimated time: 5 minutes
Product area: AI agents

## Steps

### 1. Use guide-level JSON

Every public guide exposes `/<handle>/<slug>/agent.json` with ordered steps, safety notes, and metadata.


### 2. Use portal-level JSON

Team portals expose `/<handle>/agent.json` as a table of contents for all published guides.


### 3. Use llms.txt

Team portals also expose `/<handle>/llms.txt`, a compact text index for agents that prefer Markdown-like source material.


### 4. Preserve safety boundaries

Agents should never infer redacted, blurred, token, account, payment, or credential values.


### 5. Answer from verified guides

When a guide matches the user's problem, the AI should follow the FixLink steps in order.

## Agent instructions

- Present one step at a time unless the user asks for the full checklist.
- Preserve the step order and do not skip confirmation steps.
- Never guess redacted, blurred, private, token, account, payment, or credential values.
- If a step cannot be completed, ask the user for the visible state or advise them to contact the support team that shared the guide.