Claude (Anthropic)

Claude, made by Anthropic, is one of the two AI providers Limn can use. Limn runs Claude on your computer, on your own Claude Code login or your own Anthropic API key. There is no Limn account in between, and Limn stores no Claude credentials. When both Claude and Codex are signed in, Limn uses Claude by default and runs reviews on both.

#What you need

One of these two. Both are fully supported; choose the one that suits you or your organisation.

• Your Claude Code login: a Claude plan that includes Claude Code (Pro, Max, Ultra, Team or Enterprise), signed in on this computer. Runs count against your plan’s limits.
• An Anthropic API key from the Claude Console. Runs are billed at API prices to your organisation, and Anthropic’s commercial terms apply. This is the route Anthropic recommends for products built on its Agent SDK.

You do not need to install anything for Limn itself: Limn downloads the Claude programs it needs (about 110 MB) the first time it starts.

#Set it up

With your Claude Code login

1. Install Claude Code from Anthropic if you do not have it.
2. In a terminal, run claude and sign in with your Claude account.
3. Start Limn. In Settings → AI the Claude line reads Signed in to Claude Code.

With an API key

1. Create an API key in the Claude Console.
2. Start Limn with the key in the environment: ANTHROPIC_API_KEY=… npx limn-review. To make it permanent, export ANTHROPIC_API_KEY in your shell profile.
3. If you are also signed in to Claude Code and want the key to be used, tick Use my API keys first in Settings → AI (or add LIMN_AI_USE_API_KEY=true the same way).
4. The Claude line reads Using ANTHROPIC_API_KEY.

When you have both

Limn uses the login first and hides the key from those runs, so nothing is billed to the key by accident. To use the key first, tick Use my API keys first in Settings → AI. See Which account Limn uses.

#What Limn runs on Claude

Every AI feature can run on Claude:

• AI review: reviews, follow-ups, auto review, Check threads, the CI review, design diagrams and the review chat.
• Story checks against the ticket and your docs.
• AI Fix and Implement story, in the sandbox.
• The Docs library, This Sprint’s briefs and critical paths, and the summaries in Reports, Bots Monitoring and Slack.

Limn picks a Claude model for each job (from Haiku for short summaries to Opus for deep reviews). Where a screen has a model menu, you can choose. Each run is labelled with the model and effort that ran it. Set the effort on the AI tab.

#What Limn sends to Anthropic

For each run, the prompt and the material the run needs: code, diffs, pull request and ticket text, and docs sections. Every request goes from your computer straight to Anthropic. How Anthropic treats it depends on your account; see How Anthropic and OpenAI treat your work.

#Limits and spend

On a subscription, Limn’s automatic runs stop at 50% of your week unless you change it, and everything is shown as a share of your plan. On an API key, there is no cap until you set a weekly amount in US dollars. See AI subscription usage and spend.

#Turn it off

• To use Codex instead, choose Codex under Automatic runs and default on the AI tab, or start Limn with LIMN_AI_PROVIDER=codex.
• To turn every AI feature off, start Limn with LIMN_AI_DISABLED=true. The rest of Limn works as before.