AI Chat Hooks
Chat hooks let you intercept the LLM pipeline at two points: before the prompt reaches the model and after the model returns a response. Use them to enforce compliance, brand voice, or analytics requirements without forking the chat infrastructure.
Pre-hook#
The pre-hook runs after the reader's question is captured and before the LLM is called. It can:
- Strip PII (emails, account IDs, tokens) from the user's text.
- Inject extra context — current plan, account tier, locale.
- Rephrase the question to match canonical terminology.
- Block the request entirely with a structured error.
// Example pre-hook payload
{
question: "What's the price for team@acme.com?",
context: { plan: "pro_plus", locale: "en" }
}Post-hook#
The post-hook runs after the LLM produces an answer and before it streams to the reader. It can:
- Redact sensitive information that leaked into the response.
- Append a disclaimer or a CTA.
- Reformat code blocks, links, or headings.
- Log the final pair to your own analytics store.
Use cases#
| Scenario | Hook |
|---|---|
| Compliance (GDPR, HIPAA) | Pre + post redaction |
| Brand voice enforcement | Post-hook style rewrite |
| Internal analytics | Post-hook mirror to your DB |
| A/B prompt testing | Pre-hook prompt mutation |
MCP tools#
Manage hooks from Claude Code or any MCP client:
set_chat_hooks # register pre/post-hook URLs and secrets
test_chat_hook # dry-run a hook against sample data
get_chat_system_prompt # inspect the current system promptHooks are HTTPS endpoints. Docsbook signs every call with an HMAC SHA-256 signature in the X-Docsbook-Signature header, the same scheme used by webhooks.
Pricing#
Chat hooks are a PRO+ feature ($59/month).
Related#
- AI Chat — The pipeline the hooks plug into.
- MCP Server — Configure hooks remotely from your editor.