Stop re-explaining your codebase
to every new agent.
Run Claude Code, Codex, OpenCode, and other CLI agents side by side — with shared memory that makes sessions feel infinite. The lightning fast, keyboard-first Mac app for people already working agent-first.
Claude Code 
Codex 
OpenCode 
Gemini 
Cursor 
Copilot 
Amp 
Factory Droid 🦞 OpenClaw 
Kiro A workspace built for the way you actually work with agents now.
Not a chat tool. Not a terminal. Not an IDE. The one app where agent-first builders start, organize, continue, and complete work — and where it stays coherent across sessions, agents, machines, and people.
Every agent. Every model.
Claude Code, Codex, OpenCode, Gemini, Cursor, Copilot, Amp — running side by side in the same workspace. Use the best agent for the work in front of you. Switch the day the leaderboard does.
Multi-Agent 02Memory that compounds.
Background observations, status updates, decision logs. Captured automatically, replayed into every new session, shared across every agent — Claude Code, Codex, OpenCode. The longer you use Subspace, the smarter your agents get.
Memory 03A workspace, not a chat box.
Agents, terminals, docs, files, browser, and git — grouped into workspaces, persisted across restarts. Sub-100ms switches. Keyboard-first. Built for people running 3–10 agents at once.
WorkspacesUp and running in three steps.
Add a project
Point Subspace at a folder. It creates workspaces for your project and any worktrees, and backfills memory from your recent agent sessions.
Work the way you already do
Launch Claude Code, Codex, or any agent from the command palette. Agents run in real terminals. Docs, files, the browser, and git all live in the same window.
Every session starts informed
Memory builds automatically. Tomorrow’s first session already knows what today accomplished, decided, and left open — across every agent you ran.
What is Subspace?
Your agents already know.
Hit /clear and everything’s gone. The decisions. The dead ends. The plan. You’re back to re-explaining it all.
Subspace builds memory in the background. Every conversation turn is compressed into a token-efficient observation with structured tags — decisions, blockers, progress. Every few turns, a status update synthesizes recent work into a clear picture of where things stand. Your agents can search this memory mid-session — by text, tag, workspace, or timeline.
When a new session starts, that context is already there. Claude Code context flows to Codex and OpenCode. The memory belongs to the workspace, not any single tool. Everything Subspace remembers is visible in the app — no black box.
Any agent. Any model. All at once.
You shouldn’t have to pick one agent and hope it’s best at everything. Run any CLI agent side by side — Claude Code, Codex, OpenCode, Gemini, Cursor, Copilot, Amp, OpenClaw, Kiro. Use the best tool for each task.
When one gets stuck, tag in another — it already knows what’s been tried. Memory is shared across every agent in the workspace (Claude Code, Codex, and OpenCode to start). The sidebar shows who’s working, who needs input, who’s done.
A single-vendor workspace will never seriously support its competitors. We can. We have to.
Organize the chaos.
Agents made you faster. Your tools didn’t keep up. You’re Alt-Tabbing between terminals, editors, browsers, git clients, and notes apps — more windows than ever, less visibility than ever.
Subspace puts all of it in one app. Agents, terminals, docs, files, browser, and git — grouped into workspaces, persisted across restarts. Switch between projects in under 100ms.
Cmd+K. That’s it.
One input for everything. Launch an agent with a typed prompt. Switch workspaces. Open files. Run shell commands. Search your browser history. Between the palette and keyboard shortcuts, you barely touch the mouse — Subspace was built keyboard-first from the ground up.
Reply to your agents like a colleague.
Select text in a terminal, doc, or browser and leave a comment — exactly like reviewing a colleague’s work. Queue feedback from multiple spots and fire it as one batch.
Browser comments trace to the source component, file, and line via the React fiber tree. No grepping. No “I think it’s in Header.tsx.” Direct from what you see, the agent works from where it lives.
When prose isn’t enough, ship a page.
Some work doesn’t fit in a chat reply. A prototype. A dashboard. A side-by-side comparison. A PR explainer a reviewer can actually click through. Asking an agent for those usually ends in copy-pasted markdown and a Notion doc nobody opens.
In Subspace, agents author HTML artifacts directly in the workspace — self-contained, browser-viewable pages you preview inline, iterate on, and validate before anyone else sees them. One command publishes the artifact to a trusted `subspace.build/s/<slug>` URL you can drop into Slack or a PR.
Mockups, reports, specs, demos, decision aids — the things you used to describe, your agents can now hand you a page for. Teammates open the link. Reviewers click through. Nothing leaves your machine until you publish.
Hire your team. Run them like an org.
Today, standing up a personal agent means spelunking docs, Reddit, and X to wire up skills, MCPs, identity, and memory — then redoing it all on the next major version. We’re fixing that.
Personal Agents in Subspace: one click, a short wizard, and you have a teammate. A PR reviewer. A daily-research agent. A design critic. An analyst, a marketer, a project manager — running on desktop, callable from your phone. Hire, assign work, retire underperformers. Build an agent org chart that compounds.
Ask your AI if Subspace fits you.
Click to open a balanced prompt for an AI you already trust. It will read the site, weigh tradeoffs, and tell you where Subspace fits and where it doesn’t.
Start free. Upgrade when memory becomes your secret weapon.
No projects cap. No workspaces cap. Every agent CLI. The whole keyboard-first workspace. Pro adds the layer that makes Subspace stop being a launcher and start feeling like a teammate.
Everything you need to run agents day-to-day. Unlimited.
- Unlimited projects & workspaces
- Every agent CLI, side by side
- Command palette + agent launcher
- Browser pane + comments
- Ship Loop guardrails
- Workspace persistence across restarts
Memory across sessions. Auto-drafted commits & PRs. Every Pro feature we ship next.
- Everything in Free
- Memory — background capture, observations, status updates
- Cross-session replay across every agent
- Memory search across workspaces & projects
- Auto-drafted commit messages
- Auto-drafted PR descriptions
- Every Pro feature we ship next
Shared memory across teammates. Workspace sync. Admin, billing, SSO.
- Everything in Pro
- Shared memory across teammates
- Workspace + memory sync across machines
- Admin, billing, SSO
- Priority support
On Windows? Join the waitlist →
Common questions.
What’s actually free?
How is this different from Cursor?
Is my code sent to your servers?
How is memory different from Claude or Codex Memory?
Does it work with my existing setup?
When does Windows ship?
Where agents & humans work together.
It’s time to build.
Free forever · Pro 14-day trial · No credit card · macOS · Apple Silicon