Important limitations

Keep a documentation assistant read-only

Last materially reviewed 2026-09-25

Quick answerExplain product operations without granting the assistant authority to perform account changes.
Likely to work well when

✓ Public documentation maintainers

✓ Small software teams evaluating a chatbot

✓ Editors building a repeatable answer review

Important limitations

— Private account actions

— Guaranteed factual correctness

— Replacing source owners with a widget

What to know

Write an action boundary

A documentation answer can explain where a setting lives without changing the setting. Put that distinction in the pilot brief. List actions the assistant must not perform, such as account cancellation, payment changes or sending messages. This is not merely a tone instruction. Avoid enabling connectors or tools that are unnecessary for the approved public-information task, and verify the actual configuration before exposing the interface.

What to know

Make the handoff useful

If a reader asks for an account-specific action, direct them to the official authenticated product or support route. Do not ask them to paste credentials into the public chat. The response should explain what information is missing and why the public assistant cannot complete the request. A useful boundary answers the general documentation question while leaving the private transaction in the appropriate system.

What to know

Test a harmless action request

Use a fictional prompt such as asking the bot to delete a sample workspace. The expected outcome is a clear refusal to act, with a link to relevant documentation if appropriate. Record whether it falsely claims completion or produces invented account status. Do not execute the suggested operation or test with a real customer resource merely to see what happens.

What to know

Review scope after configuration changes

A later integration or vendor feature can change what the assistant is capable of doing. Recheck the boundary whenever tools, sources or account permissions change. Keep this separate from ordinary wording improvements. If action-taking becomes a genuine requirement, treat it as a new project with its own access and recovery design rather than quietly expanding a documentation pilot into an autonomous service agent.

  • Keep a visible list of disabled actions and review it whenever the product offers a new connector or automation capability.
Source boundary

Where the safety evidence stops

This guide draws on LangChain: evaluation datasets and reference answers. Merchant-controlled records describe the provider’s own capabilities, terms or standards; they do not independently validate those claims. These records do not establish independent confirmation of the product claims.

Verify any current price, plan limit, label direction, compatibility rule, or commercial term that would materially change the decision. The dated source ledger shows the underlying records so this conclusion can be checked and updated.

Sources used for this page

These records support the facts and comparisons above. Merchant-controlled records are labelled so you can separate product claims from independent evidence.

  1. LangChain: evaluation datasets and reference answers — Platform documentation · docs.langchain.com · Merchant-controlled · checked 2026-09-25