Agentflare v0.3.1 · pre-1.0Version control built for agents

No queues.

Agentflare lets your agents work together on the same code. Every save goes straight to main, and checks run on your own machines without relying on a central host.

gabesaved

Saved the auth token migration

README.md · line 3
agent-0typing

Claimed 985–991 · rewriting peer sync

daemon/save.rs · 985–991
agent-3announced

Announced an edit before starting

tracker/shard/lib.rs · 695
mainshared by everyone

Save, and it’s on main. No queue, no conflict, no bill for rerunning the same tests.

The code is ready. The queue isn’t.

The code is ready. The hosted workflow isn’t: a pull request waits for review, parallel changes need merging, and another CI run adds minutes to the bill. Agentflare puts saves on shared main and runs checks on your own machines instead.

Removed: Create a branch, Rebase on main, Open a pull request, Wait for review, Resolve the conflict, Wait in the merge queue, Run the same tests again. What remains: save.

Ten agents. One file. Zero conflicts.

agent-0 claims lines 985–991 before it writes a single change. Everyone else working nearby is told instantly, so nobody rewrites the same code and nobody spends the afternoon untangling a merge.

How agents and access work
crates/lvcs/src/daemon/save.rsillustrative
  1. pub async fn save(&self, path: &Path) -> Result<Revision> {agent-0 claimed 985–991
  2. let content = self.workspace.read(path).await?;
  3. let parent = self.main.head().await?;gabe reading 987
  4. let revision = Revision::new(parent, content);
  5. self.store.persist(&revision).await?;
  6. self.peers.broadcast(&revision).await?;
  7. Ok(revision)
  8. }

What the wait is costing you.

Figures from agentflare.com. Illustrative, not guarantees.

  • 5 h 29 min in a hosted Git workflow0 s

    Time to reach main

    An agent saves, and it’s on main. No pull request waiting on a human, no rebase, no merge queue.

    How live editing works
  • 40 min resolving mergesNone

    Work thrown away

    No more merge conflicts eating an afternoon. Before an agent writes, it claims the lines it will change, so two agents never silently rewrite the same code.

    How agents and access work
  • 3 in the hosted workflow1

    Runs of each test

    Stop paying per minute to run the same tests three times. Your own machines test each version of main once and sign the result. A failure goes straight back to the agent that changed those lines.

    How checks work
  • 1 central host to depend on0

    Servers to wait on

    Your machines sync with each other directly, on your network or across the internet.

    How peer to peer works

The host goes down.
Your work doesn’t.

Agentflare has no central server to wait on. Saves stay on your machines and peers sync with each other directly. How peer to peer works

Central host
  • Pushes fail.
  • Hosted reviews stall.
  • Hosted CI waits.

Local work continues. The hosted workflow waits.

Agentflare
  • Saves land on disk.
  • Peers sync directly.
  • Agents keep working.

No central host required for saves or peer sync.

Install in one line. Ship at agent speed tonight.

Keep your editor and your Git history. Lose the queue, the conflicts and the CI bill. Read the documentation

curl -fsSL https://sh.agentflare.com | sh

Installs the agentflare command. No account needed.

irm https://ps.agentflare.com | iex

Not yet on Windows: no Windows build is published, so this line stops at the download and says so. Until then, run the macOS and Linux line in WSL.