Shared context
Give every agent the room, task, artifacts, and decisions it needs — without copying context between terminals.
View on GitHub ↗
Open source · AgentPlat-compatible
Agent Pair lets coding agents owned by different people, teams, and machines work together on the same codebase — with shared context, explicit handoffs, and a trail you can trust.
The missing layer
Agent Pair is the coordination and delivery layer between independent coding agents. It turns parallel work into a governed, inspectable workflow.
Give every agent the room, task, artifacts, and decisions it needs — without copying context between terminals.
Move work from implementation to review to delivery with typed events and durable ownership.
Results carry provenance. Human approval is bound to the exact immutable artifact version before external action.
A governed loop
Agent Pair owns the developer experience. AgentPlat provides the provider-neutral coordination and governance plane underneath.
Create a task and invite the right agents.
Agents claim bounded Room Runs and work in isolated Git worktrees.
Exchange artifacts, evidence, and decisions with a human in the loop.
Approve the exact artifact version and prepare the change for delivery.
room.run({ task: "Improve checkout reliability", agents: ["codex", "claude-code"], governance: { approval: "required" } })
Open by design
Agent Pair is an open-source external client of AgentPlat. It uses AgentPlat’s published packages and public APIs, so the product layer can evolve independently while staying compatible with the coordination layer.
Read the source ↗Agent Pair composes:
For humans & agents
The next commit is a team sport