🤖 AI

To enable AI capabilities (the Agile Coach, for instance), create an Anthropic API key. You can get one here.

This is a paid service. You can find more information about pricing here.

How to configure

Once you have your API key, add it to your docker-compose.yml file, under the backend section.

ANTHROPIC_API_KEY: sk-ant-... # API key from https://platform.claude.com/settings/keys
ANTHROPIC_MODEL: claude-haiku-5-5 # Optional. This is the default when the variable is unset.
AI_FREE_LIMIT: 5 # Number of AI calls per rolling month, per user, for non-Pro accounts
AI_PAID_LIMIT: 1000 # Number of AI calls per rolling month, per user, for Pro accounts

ANTHROPIC_MODEL can be any Claude model id your account can call. The default, claude-haiku-5-5, is sent with thinking turned off, so the coach stays a plain text chat. claude-sonnet-5-5 is sent with thinking limited to between tool calls instead, because that model rejects a fully disabled thinking setting.

If AI_FREE_LIMIT or AI_PAID_LIMIT is unset, the server still reads OPEN_AI_FREE_LIMIT and OPEN_AI_PAID_LIMIT.

Retrospective summary

On an unencrypted board, a logged-in participant can press Summarise with AI on the summary page. The browser sends only the board id. The server loads that board, refuses encrypted and blurred boards, and asks the configured model for a short write-up. The write-up is saved on the board, so it is still there after a refresh, and it is included when the summary is copied as Markdown. If the board changes afterwards, the saved write-up stays visible and can be generated again. The write-up can also be edited as Markdown; that edit is saved on the board and does not count as a new AI call.