Runner inventory
See Codex, Claude Code, and shell availability before dispatch.
openRuntime
The local control planefor coding agents.
openRuntime gives Codex, Claude Code, and shell tasks a shared operating surface: runner availability, task state, approval gates, isolated git worktrees, event history, and diff-first review before changes are merged back.
Running task
Claude Code · 12 events
policy snapshot frozen
git worktree created
streaming runner output
capture diff for review
cargo test task_store --all
stdout · 00:02:18 · budget 30m
1.0 Control
Replace scattered terminals with visible agent sessions.
Each run becomes an auditable task with lifecycle state, runner identity, session events, stdout, stderr, input, runtime budget, and blocking context. You see what is running, what needs approval, and what is ready for review.
Task lifecycle
Queued, running, needs input, ready for review, completed, failed, or stopped.
Local event store
SQLite persists metadata and event history across backend restarts.
2.0 Guardrails
Execution boundaries before the run begins.
openRuntime applies local policy checks for workspace access, command fragments, network mode hints, file/tool allowlists, secret redaction, approval requirements, and runtime budgets. Every task freezes an effective policy snapshot at start.
Approval gates
Workspace allowlists
Blocked commands
Network hints
Secret redaction
Runtime budgets
Review the diff before work returns to the main tree.
Repo-backed tasks run inside isolated git worktrees. openRuntime captures diff stats and patches, then supports merge and cleanup actions from the UI after human review.
Review workflow ->+ persist execution workspace for nested repos
+ capture task attempt status and exit metadata
stream lifecycle events into task timeline
- write directly against the selected workspace
Ready for reviewDiff +128 -42 · isolated worktree
Run the local agent control plane.
cd backend && cargo run
cd frontend && npm install && npm run dev
open http://localhost:3000