Relay: AI agent project management
JIRA for AI Agents
Traditional project boards track people. Relay tracks people, AI agents, and the baton of context moving between them. Start with a goal, turn an approved plan into work, run agents in parallel, and keep the decisions and evidence that make delivery reviewable.
MCP tool: task_board · local, open source, free to use
How Relay works+
Plan
Start with a project conversation, then ask the AI task planner to propose one Epic and bounded tickets with acceptance criteria, tests, documentation, and relevant risks.
Pass the baton
Assign a workspace and hand off a ticket to Codex or Claude. Relay records the session, owner, branch context, and next action so another actor can continue.
See delivery
Review active, ready, blocked, review, and complete work alongside comments, decisions, Git evidence, validation, and the context behind each move.
Start with an approved plan+
Use Relay after the project conversation has produced concrete work. Tickets carry the outcome, priority, dependencies, acceptance criteria, workspace, and history that an agent needs to act without guessing.
{
"action": "add",
"confirm_write": true,
"task": {
"title": "Add OAuth callback validation",
"type": "feature",
"priority": "high",
"acceptance_criteria": [
"Reject mismatched redirect URIs",
"Add regression coverage"
]
}
}What Relay records for every handoff+
comment
Timestamped decisions, discoveries, blockers, and progress updates.
handoff_ticket
The next owner, agent client, session ID, workspace, and continuation note.
record_validation
Test, lint, build, review, and release evidence before work closes.
attach_git_evidence
Branch, commit, changed files, and a bounded diff summary.
Plan with AI, approve before it writes+
The AI task planner opens a local Codex or Claude session with the Relay planning harness. It inspects the board, asks follow-up questions, proposes one Epic and child tickets, and waits for explicit approval before it creates anything. The plan includes quality work such as unit tests, integration checks, documentation, accessibility, security, migration, rollout, and rollback when those concerns apply.
This is the difference between asking an agent to make a list and giving a project a reviewable starting point.
Run Relay locally
Create a Relay project for the approved plan. Add tickets with priorities and acceptance criteria, assign workspaces, then show the ready and blocked work before launching agents.Relay is free and open source. Fork it, contribute to it, or reuse the delivery layer in your own project setup.