Your app
POST to the crew endpoint with the signed-in user.
AI agents & crews
Agents that work on your real data, with a person in the loop.
Build multi-agent crews in a visual canvas. Each agent has its own model, prompt, memory and tools, and every tool call goes through your access policies. Pause for a person whenever the stakes are high.
Tools
Give agents real abilities.
Agents read and write your entities, call any connector, run Glyphs, search the web, extract text from documents, ask a person a question, or hand off to another crew.
- Access policy applies to every tool call
- Agents with no unmet dependency run in parallel
- Pick a model per agent from any provider your org defines
- Memory within a run and across sessions
Guardrails
Bounded by caps you set.
Set a token cap and a turn cap per run, allowed and blocked topics, PII scrubbing, and approval on every write tool. Blocked input stops before the model is called.
- Crew-level defaults, overridable per agent
- Sliding-window or summary memory keeps context small
Human in the loop
Pause for a person when it matters.
A crew can stop for approval, a question or an edit. The run waits as awaiting approval until someone answers from the console or from your own app through one endpoint.
- Approve, reject or answer with one call
- Runs pick up exactly where they paused
Observe
See every run, token and cost.
Open any run to see its response and token count, and see tokens and estimated cost per run and per customer workspace. Stream runs live over WebSocket.
- Job Monitor across crews, Logic Blocks and Glyphs
- Guardrail hits on the dashboard
Start a crew from anywhere
Six ways in.
An event
Start a crew when a record is created or updated.
A schedule
Run a crew every morning or on a cron.
A Logic Block
Use run_crew as one step in a pipeline.
A webhook
Let another system start the run.
The test tab
Try it in the console before you activate it.
Ready when you are
Describe it once. Run it forever.
Connect your AI client, build your backend, ship to production.