A repo you've never seen, read by three agents at once.
In a codebase you don't know, the first day goes on finding where things live. Point a lead session at the repo and it splits the reading: one helper takes the data layer, one the API, one the frontend and the build. Each sends back a short summary with file paths, and the lead writes one overview from all of them.
Free and open sourceRuns on Mac and LinuxYour own Claude and Codex logins
Recorded on a demo machine: the lead sums up what its helpers sent back.
Ask one agent to explain a big repo and you get a tour of the first folder
One session reads everything itself, and a big repo doesn't fit. The parts it read first come out detailed, and the rest gets a sentence each.
One agent on its own
One reader working through the whole repo
Detail where it started, guesses where it didn't
Every follow-up starts the reading again
Agents side by side in Overvibe
A helper per area, all reading at once
One overview written from every report
Follow-ups go to the helper that read that part
Three readersand one overview
You type
❯I'm new to this repo. Have helpers map the main parts in parallel and write me one overview in ARCHITECTURE.md.
1The lead splits the repo
It looks at the top-level folders and gives each helper an area and the same three questions: what's here, what calls it, what's risky to change.
2Helpers read, nobody edits
Each one in its own terminal under the lead, reporting what it found with file paths.
3The overview comes back
The lead writes ARCHITECTURE.md from the reports and ends it with what it still doesn't know.
Behind it, one helper per area. The second line is for later: the helper that read the API comes back with its conversation and answers from what it already read. Read the CLI reference
$ overvibe new --attach --tool codex --cwd . --name "repo: map api" \
--prompt "Read src/api. What is here, what calls it, what is risky to change? Give paths. Do not edit. When done, run: overvibe send \"$OVERVIBE_SESSION\" \"<map>\""
$ overvibe resume "repo: map api" --prompt "Where does the API check auth?"
One agent will do when
the repo is small enough to read in one go, or you only care about one corner of it. Ask Claude directly. Helpers earn their keep once the repo has several parts that each take a while to read.
Questions
Will the helpers change anything?
The prompt tells them not to. For a repo you don't trust yet, start them with --mode ask, so each tool asks before it runs anything and you answer in Overvibe.
How is this different from a research brief?
Same pattern, pointed at code. The research brief covers sources outside the repo, like docs, the web and exports.
Can I ask questions a day later?
Yes. Stop the helpers once the overview is in, and overvibe resume brings one back with its conversation when you have a question about its part.
