dokan v0.1.0 — beta / preview

dokan v0.1.0 — beta / preview

dokan is an agent-operated automation runtime: your coding agent stands up, runs, and schedules deterministic workflows by talking to dokan over MCP. Zero LLM inside — the platform burns no tokens; the expensive intelligence stays in your agent, outside the runtime.

This is the first tagged release. dokan has been built and run in production against our own agent fleet; v0.1.0 makes the public presentation match that reality. Published as honest beta/preview — the release exists for OSS hygiene; a GA designation comes later.

What works in 0.1.0

• Deterministic Docker job execution — one job, one fresh network-isolated container, per-job CPU/mem caps + hard timeout, content-addressed cache (never recompute unchanged work).
• MCP control plane — upload / run / schedule / wire / read jobs over MCP (Streamable HTTP or stdio). Every response engineered to be token-frugal.
• Flow engine — compose_flow / run_flow over a validated DAG: when branches, map fan-out, compensate (saga rollback), retries, step-boundary durability.
• Triggers — 6-field cron + inbound webhooks (POST /hook/<token>).
• Write-only secrets, structured results + signed receipts, operator cockpit at / + Prometheus /metrics.

Run your first job (one command)

With a daemon up: ./examples/flagship/run.sh — a fraud-triage DAG driven over MCP the way an agent does it (upload → compose → run → read). No keys, no job network, fully reproducible. Proven green in CI (tests/p2_flows::flagship_demo_flow).

Honest status

Single-tenant trust model — see SECURITY.md. Ready for demos, design partners, and technical early adopters. Not yet turnkey multi-tenant (no SSO/RBAC/HA).

Full quickstart in the README · changelog in CHANGELOG.md.