Local runtime for multi-agent implementation work.
It schedules tasks, isolates worktrees, talks to Codex, Claude, or a fake provider, verifies the result, and applies only when the run is ready. Lens stores and projects the evidence. You drive it with the waygent CLI.
apps/ CLI, API, console
packages/ TypeScript control plane
native/kernel/ Rust execution kernel
docs/ Current docs, plus history
Codex checkout: local setup.
bun install --frozen-lockfile
bun run agent:contract
bun run check
bun run platform:demowaygent run --latest
waygent status --last
waygent inspect --last --json
waygent explain --lastApply only when the source checkout is clean and explain says the run is ready:
waygent apply --run <run_id>Live provider checks are opt-in:
WAYGENT_LIVE_PROVIDER=codex bun run waygent:live-smoke
WAYGENT_LIVE_PROVIDER=claude bun run waygent:live-smokeAgent work starts at AGENTS.md. Plans use PLANS.md. Reviews use code_review.md.
Keep .waygent/, .agentlens/, .claude/, .codex-orchestrator/, .orchestrator/, .superpowers/, node_modules/, build outputs, and local venvs out of git.