AI workflows on your own API key

Build AI workflows that run like software.

Read, sort and summarize documents, watch the web, and post the result where your team works. Runs on a schedule, shows the cost of every run, and uses your own OpenAI, Anthropic or Google key.

Start building Sign in

Watch the demo

Inspection reports PDF summaries Company news Daily digest Meeting actions Alerts to Slack

Upload Inspection report

Agent Sorts the findings

Ability Reads the whole file

Output What to fix first

Upload an inspection report and get every finding with how serious it is and where it is, then what to fix first.

Costs under a cent per run on your own key. Reading the file needs an OpenAI or Anthropic key.

Start with this

{# The demo video (demo/README.md). Its files load only when it scrolls into view (landing.js puts the sources' data-src in place); it then plays muted on a loop, with Pause. With reduced motion it shows the poster and waits for Play. #}

See it work

One sentence in. A pipeline out.

Build layer

Compose agents, abilities, files, and logic on a visual canvas.

Create reusable building blocks first, then wire them into workflows that are easier to inspect than scripts and more durable than one-off prompts.

Provider-backed agents Bring your own Anthropic, OpenAI, Google, or compatible key. Give each agent its own instructions, model, abilities, and answer format.

Abilities and functions Let an agent search your files, follow an exact answer format, use provider abilities, or call ready-made and custom functions.

Control-flow nodes Branch, loop, delay, catch failures, call HTTP, send messages, query SQL, and reuse configured nodes as templates.

Run layer

See every run step by step, with its cost, and recover when something fails.

Once a pipeline works, run it from chat, schedule it, trigger it from webhooks, or chain it from another pipeline without losing observability.

Triggers that fit real work Start pipelines manually, from schedules, from signed webhooks, or when another pipeline completes.

Debug from the trace Inspect every step's input and output, retry attempts, the exact prompt sent, and which model answered.

Costs and notifications Track token usage and estimated spend per run, compare outputs, and get notified when scheduled jobs fail.