Relay: Open Source AI Agent Project Management
Relay is a local AI agent project management system: turn plans into tickets, hand off work, preserve decisions, attach validation evidence, and review delivery in one project record.
AI agents are increasingly good at doing the work. The harder problem is directing it. When several agents are exploring, coding, reviewing, and handing off work, it becomes easy to lose the plan, the rationale behind a change, or the one risk nobody has resolved.
Relay is AgenticStore's open-source AI agent project management system. It gives people and agents one local project record from planning through delivery.
Chat is where work starts. It should not be where the project disappears.
A chat can produce a thoughtful plan, but it does not give you a reliable operating picture. It does not tell a reviewer which task is blocked, why an agent chose an approach, what was validated, or what the next agent should do. Relay gives that work a durable structure.
Plan
Capture the approved outcome, break it into tickets, set priority, dependencies, and acceptance criteria.
Delegate
Assign a workspace, give an agent a bounded handoff, and record the session attached to the work.
Operate
Track active, ready, blocked, review, and complete work across people, Codex, and Claude.
Explain
Keep comments, decisions, risks, artifacts, Git evidence, and validation beside the ticket that needs them.
Review
See what changed, what passed, what remains blocked, and where a person needs to decide.
Resume
Let the next agent start from a concise project record instead of reconstructing context from chat history.
The Relay lifecycle
1. Create the project and discuss the plan with an agent 2. Convert approved work into tickets, priorities, dependencies, and acceptance criteria 3. Assign a workspace and hand off bounded work to one or more agents 4. Record decisions, blockers, Git evidence, and validation as work progresses 5. Triage, review, and move work with a comment or a fresh handoff 6. End the day with a project view that shows what changed and what comes next
A system of record for agentic delivery
Relay is not a prettier to-do list. It connects operational facts that usually become scattered across tabs: the ticket, its workspace, the agent session, the branch, the validation result, the decision that changed the plan, and the handoff needed to continue. That makes agent work inspectable without slowing it down.
The project data stays local. Read operations are separate from writes, and board modifications require explicit confirmation. That gives agents useful autonomy while keeping the human in the loop for real project decisions.
Stop chasing activity. Start directing it.
AI can multiply output only when the person directing it can still see the work. Relay helps you delegate with intent, review with context, and stay on top of the hundreds of small decisions that shape a project.
Use Relay to turn an agent plan into a reviewable delivery system.