Run AI agents like you run production.
Deterministic pipelines around probabilistic agents, dispatched to your machines, observed live, measured in hard numbers. Your keys, your code, your metal.
One command, and your first run is on the dashboard.
Install the CLI and run it. Your browser opens once, for one approval. Everything after that — the plugin, a starter pipeline, this machine as a runner, the first run — happens without you.
$ bun add -g @baizor/pipeline$ pipeline initNeeds Bun, and Claude Code signed in with your own subscription or API key. Your keys and your code stay on your machine.
- 01Approve once, in the browserinit opens a single consent screen and catches the redirect itself. No token to copy, nothing to paste, no dashboard visit first.
- 02It sets itself upThe Claude Code plugin, a starter pipeline, the local dashboard, and this machine enrolled as a runner that starts on boot. Same command, no questions.
- 03Watch the first run, liveThe starter pipeline runs on your metal and streams to your account step by step. Metadata only by default: statuses and timings leave, transcripts and code do not.
No account wanted? pipeline init --local does everything above on your machine and sends nothing anywhere.
The control plane around your agents
Everything the dashboard adds on top of the open-source engine — without ever standing between you and the model.
Privacy tiers
Metadata-only by default, enforced agent-side: statuses, timings, and token counts leave your machine; transcripts and code don't, unless you opt up per project.
Bring your own keys
Your subscription, your API keys, your machines. Model traffic never touches us; the cloud is a control plane, not a proxy.
Live canvas
Pipelines render as node graphs with execution lighting up in real time. Drill into nested pipelines to any depth; they compose like Lego.
Pipeline registry
Publish and reuse versioned pipelines. Every published version is an immutable snapshot pinned by its content hash.
Budgets & cost attribution
Per-run cost segmented by model, pipeline, and project, with spend caps and budget alerts before an unattended queue burns money.
Matrix & runner health
Run the same pipeline across models and versions, compare success, cost, and duration, then catch regressions with numbers, not vibes.
What the dashboard does today
Early access, rough edges and all, but this all ships in the cloud control plane right now.
Also today: Run history & detail, with a canvas for composed run trees · RBAC, org secrets, and an audit log for teams · Notification channels: Slack, Telegram, email · A shared pipeline library, versioned in git · Installable PWA with web push.
Hosted runners are still on the roadmap; everything above runs on machines you own.
Live prices, per person
Start free on your own machine. Upgrade when the queue starts working for you — monthly or annual, sold by Paddle as Merchant of Record. Usage is metered on top, at published rates.
Free
Early accessThe live dashboard plus local-first orchestration. Sign in and try it on your own repos.
- Live run view + 7 days of history
- Unlimited local runs and self-hosted runners
- Answer needs-input from the web
- Metadata-only privacy, enforced agent-side
- BYOK: your keys, your subscription
Pro
LiveFor the power user who stops operating agents and starts feeding the queue.
- 90 days of history
- Personal analytics: success, cost, duration
- Scheduling + GitHub triggers
- The self-draining task queue + digest
- Web push: answer needs-input from your phone
Business
LiveFor teams that run pipelines like shared production infrastructure.
- 365 days of history
- RBAC, org secrets, and an audit log
- Shared pipeline library, reviewed in PRs
- Cost attribution per developer
- Matrix runs, regression alerts + budgets, approval gates
Local runs and self-hosted runners are unlimited on every plan, including Free. They are never metered and never draw the balance.
Subscriptions are sold by Paddle as Merchant of Record — Paddle is the seller on your receipt and handles sales tax and VAT wherever you are. Both billing periods are offered: an annual plan is one payment covering twelve months at the price of 8. Hosted runners meter per second of active runtime against your balance when they ship.
Straight answers
Do you see my code or my API keys?
No. Pipelines live in your own git repo and your keys stay on your machines. By default the cloud stores only metadata: statuses, timings, token counts; never transcripts or code, unless you explicitly opt up per project.
Is this a gateway or proxy for AI APIs?
No. Model traffic never touches us. Pipeline is an orchestration layer above your agent harness; you bring your own keys and your own subscription. We don't resell, discount, or route tokens.
What are privacy tiers?
Metadata is the default and it is enforced agent-side before anything is sent. You decide, per project, whether to send more than metadata. Nothing leaves your machine unless you opt in.
Can I self-host the agent?
Yes. The execution engine (the pipeline plugin and the pipeline-runner daemon) is open source. Run it on your own machines; the cloud is the optional control plane on top.
Which agent harnesses are supported?
It is built harness-agnostic, with Claude Code supported first. Codex/Gemini-class CLIs are on the roadmap; the open pipeline format is designed to outlive any single vendor.
The dashboard is live. Connect your first runner.
Your keys, your subscription, your data. Model traffic never touches us.