Agents

Your agents should work like your engineers — observable.

Isolated worktrees, structured sessions, and context annotations. Every agent action is attributed, every workspace is independent, and the full history is always shared.

Get early access Read the product overview

Isolated worktrees

Every agent gets its own checkout. No lock contention. No shared state.

Each agent operates in an isolated worktree backed by a lightweight pointer to the shared object store. Spawning is instant. History is shared. Working directories never collide.

Sessions

Agents register on start, segment work, and sign off.

Provider, model, and policy are recorded when a session begins. Segments mark meaningful checkpoints. When the agent finishes, the session closes — and every action in between is attributed.

Context annotations

Attach constraints and rationale directly to files and symbols.

Agents read and write structured context — invariants, design rationale, usage notes — scoped to files, symbols, or line ranges. Context carries forward on every snapshot. No separate docs to maintain.

Multi-agent coordination

Any number of agents, any number of worktrees, one shared history.

Each agent works on its own thread with full snapshot isolation. Merge when the work is ready. Attribution from every agent is preserved through the merge — you always know who contributed what.

Context that travels with the code

Agents write context. Engineers read it. Nobody loses track.

Context annotations attach constraints, invariants, and design rationale directly to files and symbols — scoped precisely and carried forward on every snapshot. The connective tissue between the agent that wrote the code and the engineer who maintains it.

See the developer workflow