Spatio

The workspace agent that finishes before a frontier harness has finished thinking.

[Spatio @spatio_labs]

@mattpark True

[Matthew Park @mattpark]

This should be the primary interface for agentic development.

[Matthew Park @mattpark]

when your agent can get tasks done accurately in a matter of seconds, it requires a very specific form factor to support that throughput so you can also keep up.

each one of these rows are Claude Code sessions in https://t.co/OfBs0x5Wyf's desktop control plane that either I invoked (via middle header input) or indirectly by the spatio workspace agent harness.

[Matthew Park @mattpark]

At the Spatio Workspace Agent's throughput, edits feel instant at a marginal intelligence tradeoff ⚡️

The Spatio Workspace Agent, is about 60x faster than Claude Code/Codex, for tasks that require state of the art effort, it intelligently handoff to your existing Claude Code/ Codex subscription

[Matthew Park @mattpark]

building a full-fledged agent control plane is absolute hard mode because the surface area of what an agent needs to touch is limitless... but it's the only way we actually move past chat wrappers into real execution

[Matthew Park @mattpark]

many of the most important integrations that you need, don't have a reliable API. @spatio_labs solves this.

[Matthew Park @mattpark]

Spatio (https://t.co/OfBs0x5Wyf) is officially in public beta.

> A native macOS workspace where you, your team, and your agents work in the same place

> Agents are first-class participants, not a chat window off to the side. They see what you see.

> Split by speed. "Summarize this thread." "Rename these 40 files by what's in them." A frontier agent harness spends minutes on those. Spatio agent harness runs on Cerebras at 1000 to 2000 tok/s, so they come back before you've finished reading your own prompt.

> Quality where it counts. Anything that needs real editing hands off to your existing Claude Code or Codex sub.

> Connect your tools once. Anything with an API goes through our provider interface. For the things without one, we embedded Chromium directly, a real browser engine rather than a lazy web view, so agents can work anywhere you can reach in a browser without sacrifices.

> One MCP, not a dozen. All of it is exposed through SpatioMCP and queryable from any client: Claude Code, Codex, Cursor, or Spatio Desktop.

> It compounds. Everything you and your agents produce persists to your own workspace backend, so last month's decisions are today's context.

https://t.co/OfBs0x5Wyf

[Matthew Park @mattpark]

As of version 0.7.1, Spatio is now available in public beta. Download the native macOS client today. If you download, send me a dm or comment below to request to join the group chat.

[Matthew Park @mattpark]

No excuse to still be shipping electron or tauri imo. We migrated away from it a little while ago for @spatio_labs… one of the best decisions we’ve made.

[Matthew Park @mattpark]

every ai browser has the wrong mental model. might as well build an ai operating environment despite how cringe and buzz wordy that sounds. I don’t want a browser with an agent bolted on, I want an environment that has breadth like a traditional operating system but structure like a productivity workspace saas app (e.g Notion, Google workspace, etc…)

this env should let my agents go free and access/ render/ CRUD any artifact for the task at hand (docs, slides, spreadsheets, a post on X) and access any artifact via a web view backed by a real browser engine.

this is where the browser is important… because a lot of artifacts can only be accessed via the web

but there are many artifacts like an email, task, a record that would have a better ux via native interface.

an environment like this simply does not exist and it is the exact thing that i and many of the smartest people I’ve spoken with wish we had.

@spatio_labs. public beta -> July 15. stay tuned.

[Matthew Park @mattpark]

Spatio public beta coming 🔜

[Matthew Park @mattpark]

If you’re interested in trying @spatio_labs and willing to provide brutally honest feedback, send me a dm and I’ll share a private beta access code!

[Matthew Park @mattpark]

btw @spatio_labs let's you use any coding agent cli: claude code, codex. so you can benefit from this new feature regardless of the agent you want to use! go from local to cloud in < 1s. feels like magic.

[Matthew Park @mattpark]

great minds think alike, we actually just shipped this natively yesterday in @spatio_labs, but decoupled it so it works for codex, claude code, or whatever agent session you have running

[Matthew Park @mattpark]

one of the most annoying things about claude code is being stuck at a coffee shop waiting for a local run to finish

with the latest @spatio_labs update, handing off your active agent session between local and cloud is seamless. you can literally hand it over, close your laptop, and let it execute in the cloud while you walk out, then switch back to local when you get back online.

for this initial version we are using @flydotio's https://t.co/fiLutrIc1B!

[Matthew Park @mattpark]

every detail matters.

[Matthew Park @mattpark]

spatio v0.2.7 is live

most of this is fixing things i personally got annoyed by. we’re shipping updates multiple times a day now, so if you give us feedback, and it is valid, it’ll probably be live by tomorrow

for now, updates aren't automatic. you'll need to explicitly trigger them. either hit 'check for updates' in the menu bar, or click 'update' in the header

posting more raw, unedited videos like this while we’re in private beta. send me a message if you want access and would like to try it out!

[Spatio @spatio_labs]

@tobi Very interesting. We’ll look into this.

[Spatio @spatio_labs]

@mattpark 👀

[Matthew Park @mattpark]

the chat interface is a massive crutch for the agentic era. text prompts are slow, high-latency, and low-fidelity.

if you're orchestrating multiple agents, you don't need a longer chat history, you need a spatial desktop canvas.