agor.live - Multiplayer AI Agent Orchestration

aiagentstoolsdevelopment

What if you could run Claude Code, Codex, and Gemini sessions simultaneously on a spatial canvas, coordinating them like pieces on a game board?

That’s agor.live.

The Core Idea

Agor is a multiplayer spatial layer for AI coding agents. Think Figma, but instead of design elements, you’re orchestrating AI coding sessions. Multiple agents working in parallel, each in their own isolated git worktree, all visible and manageable from one interface.

What Makes It Interesting

Worktrees everywhere. Each agent session runs in an isolated git worktree. No branch conflicts, no stepping on each other’s toes. You can have one agent refactoring your API while another writes tests and a third updates documentation—all simultaneously.

Spatial coordination with zones. Drag a worktree into a “Ready for Review” zone and it auto-prompts for code review. Drop into “Needs Tests” and it generates test prompts. The spatial metaphor becomes a workflow automation layer.

Session forking and spawning. Fork a session to create parallel exploration paths with shared context. Spawn child sessions for subtasks with focused context. Agents can supervise other agents through an internal MCP service.

Bring your own everything. Connect VSCode, Cursor, or any IDE via SSH to Agor-managed worktrees. Use Claude, Codex, Gemini—whatever mix suits your work.

What This Enables

Run a fleet of coding agents in parallel on complex refactoring. Have specialized agents for different concerns (security review, performance optimization, test coverage) working simultaneously. Coordinate team efforts where both humans and AI agents share the same spatial workspace.

The mental model shifts from “me directing one AI assistant” to “me orchestrating a team of AI specialists.”

Early Days

It’s pre-1.0 and source-available under BSL 1.1. There’s a learning curve with the spatial workflow paradigm. And running many parallel agents burns through API credits fast.

But the direction is compelling: treating AI agents not as chat interfaces but as first-class team members in a shared workspace.

Worth exploring at agor.live.