Do not repeat abandoned paths
Tried-and-abandoned approaches may never show up in code. Mainline records them, so the next agent sees the trap before walking into it.
Git-native intent records
Git for the AI era.
Mainline lets agents automatically save developer intent and decisions alongside code in your Git repo.
Install
curl -fsSL https://raw.githubusercontent.com/mainline-org/mainline/main/install.sh | bash $ mainline preflight --json
# Read repo history, decisions, and risks first
$ mainline start "Clean up old export"
$ mainline append "Keep legacy CSV until customer migration finishes"
$ mainline seal
# Write this judgment back to GitData stays in your Git repo
Git-native workflow
- Stored as
- Git refs and notes
- Collaboration
- fetch, branch, merge
- Read/write
- agents use Git automatically
Record
Agent intent, tradeoffs, and no-go zones are saved with code.
Sync
Team members get the same decision history through Git.
Avoid
Future agents see old traps and high-risk boundaries before editing.
Mainline does not save code or AI chats. It saves the intent, tradeoffs, and key decisions future agents need.
Chat logs are too long, PR descriptions come too late, and code comments are too local. Mainline compresses the engineering judgment future agents need into intent records inside the repo.
Find logic conflicts before Git conflicts
Another agent may already have changed product behavior and recorded an intent before a PR exists. Other agents can see that intent early and avoid the conflict from the start.
Read module constraints before editing
Some code looks odd and comments are incomplete. Mainline tells agents the explicit constraints each module carries.
From reading code to reading intent
Reviewers see not just the code diff, but the original goal, reasoning path, and key decisions behind the change.
Collaborate on agent decisions the way you collaborate on code.
Mainline reuses Git's mature collaboration model. No new memory database, no platform lock-in, and no need to put every agent transcript into the repo.
Fetch Receive sealed intents from the team and other agents.
Branch Record this task's goal, pivots, and validation on the current branch.
Merge Bind merged commits to the decisions that produced them.
Fork External contributors can bring their own intents into upstream review.
Agents operate it themselves.
Mainline uses agent hooks, skills, and the CLI to read and write intent. Users keep their existing AI coding workflow: ask for work, inspect the diff, and review as usual.
Self-dogfood
Mainline publishes its own live intent Hub.
This is the Mainline Hub generated by the Mainline project itself, showing real intent records from day-to-day development. After your project adopts Mainline, it can use the same view to inspect historical decisions, risks, and work in progress.
Live static snapshot
272intents
4proposed
236merged
0constraints
282files
6actors
What the Hub helps you see
236 merged intents explain why the code became what it is now.
0 risk and constraint signals help agents avoid known traps before editing.
4 proposed intents show what the team and agents are working on now.
Core position
Not another memory system. Agent memory lives in Git.
Code history already lives in Git. Mainline puts each agent's important engineering judgment into the same shared history.
Teams do not need to save entire conversations or invent a separate shared service. Agents read prior decisions from Git, finish the work, then write the new judgment back to Git.
For solo builders
If today's agent abandons a plan, another agent three weeks later should know why.
For teams
Turn agent progress, historical constraints, and review logic into repo facts everyone can fetch.
Public alpha
Code goes in Git. Agent judgment should too.
Mainline is for teams already using agents in real engineering workflows and starting to worry about broken context. Feedback is very welcome.