Home / Use cases / AI coding agents
AI coding agents
grite was built so AI coding agents are first-class issue authors alongside humans. The task graph lives in the repo, so any agent that clones it inherits the full context — and conflict-free merges mean several agents can work it at once.
The problem
- ▸Autonomous agents lose context between sessions when their notes live in a chat log instead of the repo.
- ▸When two agents work in parallel, a hosted tracker forces them through rate limits, auth tokens, and last-writer-wins races.
- ▸External trackers are a network dependency — an agent working offline or in a sandbox cannot reach them.
How grite solves it
- ▸grite stores the task graph as an append-only event log in refs/grite/wal, so it travels with the repo and survives across sessions as durable, shared memory.
- ▸CRDT merge means concurrent agent edits never conflict — the result is identical no matter which agent syncs first, and no edit is silently dropped.
- ▸Every operation is local and offline-first; agents sync through the same git remote they already push code to, with no keys or accounts.
An agent files and closes work
$ grite issue create --title "Refactor parser" --label agent issue 91af · created $ grite issue close 91af issue 91af · closed $ git push # the graph travels with the code
Questions
+ How do concurrent agents avoid conflicts?
grite merges concurrent edits with CRDT rules — last-write-wins on scalars, commutative sets on labels and links — so any two logs merge to the same result regardless of order.
+ Does an agent need network access?
No. Every grite operation is local. An agent can create and close issues fully offline, then sync through git whenever a remote is reachable.
Try it in your repo
One install, one grite init, and the task graph lives with your code.