Every agent.
One canvas.
Prime WorkSpace is an infinite spatial workspace where one lead Claude orchestrates a fleet of coding agents. Wire them up with flows, let your principal delegate, review, and ship — never lose context again.
Free during beta · macOS 14+ · Windows 10+ · Privacy-first, runs locally
AI agents got fast. Your workspace didn't.
Coding agents write, refactor, and test code faster than any human. The bottleneck shifted — from the code to the coder.
You can spin up three Claude Code sessions at once. But you're still alt-tabbing through terminal windows, losing context, forgetting which agent owns what.
The tools caught up. Your terminal didn't.
Without Prime WorkSpace
- Terminal tabs you can't tell apart
- Context switching kills your flow
- Notes scattered across apps
- Agents finish silently, you miss it
With Prime WorkSpace
- Every agent visible on one canvas
- Pan, zoom, focus — never lose context
- Sketches and notes next to the work
- A lead orchestrator drives every worker
Everything for multi-agent workflows.
Orchestrator-led delegation
Wire a lead Claude to its workers with flows. The principal delegates, workers report back. No middleman, no copy-paste.
The prime CLI
From your orchestrator, run prime ask "Reviewer" "check this PR" — it returns the worker's response on stdout. Works with any CLI agent.
Auto-cascade
Mark a flow as auto-cascade and the next worker kicks off automatically when the previous one finishes. Pipeline your team like Unix pipes.
Infinite canvas
Pan freely. Zoom on the agent that needs attention. Group related work spatially the way your brain already thinks about projects.
Sketch your thinking
Draw shapes, arrows, sticky notes right next to the agents doing the work. Architecture diagrams that live where your code lives.
100% local
No telemetry, no cloud, no account stored. Your agents and notes never leave your machine. Anthropic and OpenAI still bill you for their APIs — that's between you and them.
Spawn. Wire. Delegate.
Spawn your lead and its workers
Press Ctrl+K to drop a Claude Code agent anywhere. One becomes the orchestrator, the rest are specialists — coder, reviewer, tester. Each with its own terminal and working directory.
Wire them with flows
Press Ctrl+E, click the lead, click each worker. A bezier flow forms. The orchestrator automatically learns who reports to it via its system prompt.
Let your lead drive the team
From the orchestrator, run prime ask "Reviewer" "check this implementation". The lead delegates, the worker responds, you watch from the meta level.
See it in motion.
60-second walkthrough of an orchestrator delegating a build to three workers in parallel.
What people build with Prime WorkSpace.
Orchestrator + Workers
One coordinator delegates tasks to specialized workers. Coder, reviewer, tester — each in its own context, each with its own tools.
Code review pipelines
An author agent writes, a reviewer agent critiques, a tester agent validates. All wired with auto-cascade so the chain runs hands-off.
Research swarms
Three agents researching different angles of the same question, results aggregated by a fourth. No prompt engineering — just spatial composition.
Pair programming with two Claudes
Two agents on different parts of the same codebase, talking to each other when they hit a shared interface.
Prime WorkSpace vs juggling terminals
| Capability | Manual juggling | tmux panes | Prime WorkSpace |
|---|---|---|---|
| Layout | alt+tab chaos | static panes | spatial canvas |
| Orchestration | you copy-paste | none | lead-to-worker delegation |
| Auto-cascade | no | no | yes |
| Visual sketching | no | no | inline |
| Local-first | yes | yes | yes |
| Works with any agent | yes | yes | yes |
$5/month. Bring your own agents.
Start with 7 days free. Then $5/month unlocks Prime WorkSpace on all your Macs. Connect as many agents as you want — your Claude / Codex / OpenCode keys stay yours, your data never leaves your machine.
macOS 14+ · Windows 10+ · Built natively in Electron · Cancel anytime
FAQ
No. Prime WorkSpace is the orchestration layer; your installed agents (Claude Code, Codex, etc.) call their own APIs with your own keys. Prime WorkSpace is privacy-preserving by design — it doesn't see your prompts or responses.
Anything that runs in a terminal. Claude Code, Codex, OpenCode, Aider, plain bash, plain zsh. The agent-to-agent CLI is shell-based, so it's truly agent-agnostic.
macOS (Apple Silicon and Intel) and Windows builds are available today. Linux is on the roadmap.
Prime WorkSpace queues agent-to-agent calls and has a 60s hard timeout per call. Cyclic agent graphs (A asks B asks A) will time out instead of looping forever, but you should still design hierarchical graphs.
tmux gives you panes. Prime WorkSpace gives you a graph with a clear chain of command. A lead Claude orchestrates its workers, delegates with flows, and auto-cascades. tmux is for humans driving terminals; Prime WorkSpace is for one lead agent driving the rest.
A community edition is on the way.