Control surface
A spatial canvas where people can see agents, screens, branches, resources, and relationships at once.
Wega Labs · October
The operating system for AI collaboration.
October is a shared workspace where humans and named AI agents can see one another, exchange context, work in parallel, and make their output visible as it happens.
CategoryAI collaboration OS
The problem / 01
AI agents are becoming coworkers. Their interface still assumes they work alone.
Today, powerful agents live in separate terminals, chat threads, IDE panes, and vendor silos. They cannot reliably see the same workspace, discover who owns a task, exchange context, or show a human how the whole product is changing.
The next interface is not a better prompt box. It is a native collaboration environment for people and machines.
Why an operating system / 02
A control plane above every agent.
Coordination layer
The october-bus gives agents a shared protocol for peer discovery, messages, canvas context, and work ownership.
Execution layer
Local agent processes edit real repositories inside isolated worktrees, with dev servers and live previews attached.
October is the conductor, not the harness. Its long-term value is neutrality: one collaboration layer above local agents, model vendors, IDEs, and remote builders.
Native collaboration / 03
The october-bus
A coordination protocol exposed to agents through MCP. It turns the canvas into shared operational context instead of a picture only the human can see.
HumanIntent + direction
→OrchestratorResolve names, screens, ownership
→Screen A
Backend
Tests
Shared bus
discover peers · inspect canvas · message agents · report status · preserve provenance
System status / 04
What exists. What remains.
Built + shipping
Current desktop- Spatial infinite canvas✓
- Route detection + live previews✓
- Local coding-agent harnesses✓
- Integrated terminals✓
- Git worktree isolation✓
- Variant previews + PR flows✓
- Otto voice control✓
- Cross-platform CI gate✓
Agent roster
Local harnesses01Claude Code02Codex03Cursor04Grok05Gemini06opencode07Hermes08Cline09Pi
Functional support does not yet mean equal collaboration depth. Bus parity is an explicit frontier, not a finished claim.
Active frontier
Named-fleet orchestration
A human should be able to direct Apollo, Atlas, or the agent on a specific screen without translating intent into terminal sessions.
Bus equality
Claude Code is currently the most deeply integrated bus citizen. The protocol must make Codex, opencode, Cursor, and others equally legible and cooperative.
True multiplayer
Humans and agents should work simultaneously with explicit ownership, provenance, presence, and safe synchronization.
Local + remote fleets
A remote builder should be commandable beside local agents—not embedded as a disconnected tab.
Human control / 05
See the fleet. Steer the work. Inspect the result.
Spatial legibility
Every detected route becomes a live screen. Connections make context and responsibility visible.
Real execution
Agents edit the repository on your machine; HMR updates previews as files change.
Safe parallelism
Worktree sessions isolate variants and agent work before changes become a branch or pull request.
Provenance
The system is being built to answer who changed a file, what an agent is doing, and where its output lives.
The collaboration loop / 06
1Open the system
→October detects repositories, routes, frameworks, and running surfaces.
2Compose a fleet
→Name agents, connect them to screens, and give each one a bounded task.
3Agents coordinate
→They exchange context and status through the shared bus while working in isolation.
4Watch + steer
Live previews, diffs, terminals, and provenance keep the human in command.
A place, not a panel / 07
A serious operating system should still feel alive.
The visual world is deliberate: agents are easier to direct when the workspace has spatial memory, recognizable places, and a sense of presence.
October by Wega Labs
One workspace. Many agents. Shared context.
The operating system for humans and AI to build together.
Visit october.dev ↗