Agent Flow
The agent operating system for enterprise work: per-project repositories, Kanban tasks assignable to humans or agents, goals with a definition of done, and full cost visibility — model-agnostic, from planning to production.
What Agent Flow Does
Eight capabilities that turn a backlog into work an agent can actually pick up and finish.
Per-Project Repository
Every project gets its own GitHub repository — the source of truth for code, plans, and history, for both human and agent contributors.
Kanban for Humans and Agents
Tasks live on a Kanban board and can be assigned to a human or an agent interchangeably — the same workflow, whoever picks it up.
Goals with a Definition of Done
Every goal carries an explicit definition of done and an AI-generated plan that a human reviews and approves before any agent starts work.
Cost Tracking
Per-session and per-agent cost tracking, so spend on autonomous work is visible and accountable, not a surprise on the invoice.
Scheduled Automations
Recurring and scheduled automations run agents against a project on a cadence, without a human kicking each one off by hand.
Model-Agnostic Providers
Configure the model provider per project or per agent — no lock-in to a single LLM vendor.
MCP Connectors
Connect agents to external tools and data sources through MCP, extending what an agent can see and do without custom glue code.
Repository-Backed Knowledge Base
A knowledge base backed by the project's own repository keeps context durable and versioned alongside the code it describes.
How It Works
Define the goal
Write the definition of done. Agent Flow proposes an AI-generated plan for a human to approve.
Assign the work
Tasks move onto the Kanban board and go to whichever human or agent is best suited.
Track and ship
Progress, cost per session and per agent, and the repository history stay in view until the goal is done.