You want answers out of the project, not out of a search box.
The chat searches for meaning rather than keywords, across tasks, comments, documents, doc pages, notes and conversations. Which model does the computing is your choice. Your software documentation stays where it belongs: as files in the repository.
What you get
One question across tasks, comments, documents, documentation pages, notes and released meetings. The answer names the sources it came from.
The source code is read as structure, not as a text file. Questions about what is actually implemented hit the repository rather than a colleague's memory.
Through the MCP server your development environment reads, creates and comments on tasks directly. The access token can be limited to individual projects.
A folder in the repository becomes the documentation in the project. Written in the pull request, read also by people without repository access.
What it does not do
Three things we deliberately are not.
- No repository hosting. Your code stays with GitHub or GitLab; we only read it.
- No pipeline. We build nothing, test nothing and ship nothing.
- The chat does not write. It creates no tasks and changes no status. What happens on the project, a person did.
What people ask at this point
- Do I have to upload my code for this?
- You connect the repository and Lalabase reads from it. Connect none and there is no code in the chat, and the rest works all the same.
- Which model answers my questions?
- The one your organisation configured. Either from the package or through your own key with OpenAI, Anthropic, Google, or at an OpenAI-compatible endpoint.
- What exactly can the MCP access do?
- Read tasks, create them, comment on them. No deleting and no status changes from outside.
Look at it
rather than take our word for it.
Productive in under 5 minutes. No sales loop.