Drag agents, tools, and models onto a canvas. Wire them with arrows. Schedule, run, replay. Research, ops, reporting, outreach. Anything an LLM can reason about, an agent runs alongside your team.
A daily treasury briefing in your inbox at 7am. A deal-flow screener that reads pitch decks and shortlists three. A risk alert that pages on-call when a Slack channel spikes. A board update drafted from real filings, not vibes. Each one is a small team: a researcher, an analyst, a writer, a reviewer. You hire from 111 prebuilt personas, brief each in plain English, scope what they're allowed to touch, and wire their handoffs on the canvas. The org chart is the program, and the canvas runs the meeting.
Tools are the actions an agent can take in the real world: send an email, read a CRM record, fetch the weather, post to Slack. Each agent only gets the ones you hand it.
Every pipeline is four primitives. Steps, tools, subagents, edges.
The model picker sits on every step. Claude, GPT, Gemini, or a local Ollama on your own runner. Mix providers inside one pipeline.
Three channels for memory. One channel for the human.
Every pipeline lives in a project. Your team shares one canvas, one set of connectors, one run history.
Connectors are project-scoped. The owner sets a key once, every pipeline inherits it. Rotate once, every run picks up the new credential. Owners decide who joins, who edits, and who only watches. Invite by username, or by a shareable link with an expiry.
Diff two runs side by side. Feed a replay back as a regression test.
Join the community