Put one thousand agents to work.

Long running agent workflows. Open source.

Everything a production workflow needs.

Triggered however your work happens, running on any model, at a cost you set. Ordinary code, one control plane.

Triggers
Cron 0 9 * * 1-5
Webhook
GitHub · PR opened
Slack message
Human approval
Workflow call

Triggered by anything

A schedule, a webhook, a pull request, a Slack message, or a person clicking approve.

Model
claude-opus-5
gpt-5.6-sol
claude-fable-5
kimi-k3
gemini-3.6-flash
See all models

Any model, your keys

Hundreds of models: Anthropic, OpenAI, Google, Moonshot, and your own on any OpenAI-compatible endpoint. No lock-in.

This month
Spend cap$1.47 / $5.00
Tokens1.2M in · 340K out
Compute4m 12s
within budget

A cost you control

A hard budget cap on every workflow and key, transparent metering, and your choice of runner.

Build production-ready workflows in seconds.

Describe what you want and watch it ship, straight from the coding agent you already use, Claude Code, Cursor, or Codex. It writes the code and deploys it.

Cost and compute, under your control.

Choose a model per call, cap spend per workflow and per key, and pick the runner that fits, hosted or your own hardware. No opaque credits, no surprise bills.

See pricing

A workflow is just code.

morning-digest/
// workflow.jsonc
{ "slug": "morning-digest",
  "triggers": [{ "kind": "cron", "expr": "0 9 * * 1-5" }] }

// src/index.ts
import { agent, secrets } from "@boardwalk-labs/workflow";

export default async function run(): Promise<string> {
  const token = await secrets.get("GITHUB_TOKEN");
  const issues = await fetchIssues(token);

  // the model writes; it never sees the secret
  return agent(`Summarize:\n${issues}`);
}
  1. Write it

    Plain TypeScript or Python. One agent() call grows into fan-outs, verifiers, and loops, held together by ordinary control flow.

  2. Deploy it

    boardwalk deploy ships the package. It goes live on its schedule, its webhook, or on demand.

  3. It runs on a real machine

    Its own VM with a filesystem, a shell, and a real browser, for as long as the work takes. Live-tail it, or read the full record later.

Secure by default.

Every run is isolated in its own VM, with no standing credentials, secrets scoped to your org, and all traffic leaving through an egress proxy: internal networks and cloud metadata are always blocked, and any workflow can be locked to an allowlist. Runs reach your own AWS, GCP, or Azure with per-run OIDC identity instead of stored keys. Run them on our cloud, or self-host the runner on your own hardware.

How isolation works

Open source. All of it.

The engine, SDK, and CLI are Apache-2.0 and MIT on GitHub. Read every line, run it on your own hardware, and leave anytime.

Put your team in control.

Transparent pricing. Pay for the compute and tokens each run uses.