Connect an AI agent
Example prompts
Here is what the agent does for a few common requests, and what to check after each.
Prompts to try
Connect this repo to Malmoi
The agent lists your repositories, detects the translation files, and creates the project with its first sync, like Create a project in the browser. The result includes the workflow file and a push token. Ask the agent to commit the workflow file on the base branch you chose, after storing the push token.
If your GitHub connection or the GitHub App is missing, the agent returns a link to Account. Finish there in the browser and ask the agent to try again. If no translation files are found, the link opens project setup in the browser, where you can choose the format yourself.
Store the push token
The push token appears once in the agent's conversation, which your agent also sends to its AI provider. If that matters, rotate it later from Settings in the browser. Ask the agent to store it as the repository's PUSH_TOKEN secret by passing it on standard input:
gh secret set PUSH_TOKEN --repo OWNER/REPOReplace OWNER/REPO with the repository. Don't add --body; --body - would store a single dash. The name must stay PUSH_TOKEN, because the workflow reads that secret. See Add the workflow for the rest of the setup.
Fill the empty fr translations and publish
The agent finds keys with no French value, writes them, and saves them. It then previews the changes and publishes them as one pull request, like Publish in the browser. Logs show the saves and the publish under your name.
Review the pull request before merging it. The agent can publish only if the token allows Translate & publish.
What happens next
Check Logs to see what the agent changed. On the MCP connector page, disconnect an app or rotate or revoke the token when you no longer need it.