Pick up where anyone left off.
Start in one tool. Finish in another.
Multi-agent shouldn’t mean one person. Bring your team, their favourite agents and a thousand sandboxes into the same files, at the same time.
Two MacBooks share one file. You ask Claude Code to tighten the goals in launch-plan.md while Dale asks Codex to give every task an owner. Claude saves first and the file changes on both Macs. Codex's older save is held back instead of overwriting it, so Codex reads the latest version and adds the owners. Finder on both Macs shows the team folder mounted as a drive, and each agent's new work appears on both. Baseseat works with Claude, Codex, Cursor, Gemini, Copilot, OpenCode, Ollama and any other agent.
A folder is just the interface.
Teams write plain Markdown now, and ask an agent instead of clicking through an app. Under the folder is a shared workspace built for everyone and every agent working at once, and it scales as far as you take it.
Out of the box
Claude Code, Codex, Cursor or any other agent reads, writes and searches it with the file tools it was trained on. Nothing to install, nothing to teach.
Drive-wide search
A search engine sits behind every drive and keeps up as files change. Agents search every word, pattern and file at once, then read just the lines they need.
Who owns the pricing page, and when is it due?
Working
launch-plan.md:9- [ ] Pricing page · @dale
notes/calls/2026-11-04.md:22Pricing page feedback from the Acme call
pricing.md:3Draft: @dale, due 14 Nov
pricing.md:3Draft: @dale, due 14 Nov
1# Pricing
2
3Draft: @dale, due 14 Nov
Dale owns the pricing page. The draft is due 14 Nov, according to pricing.md.
Programmable
No mount needed. A few lines of Python search every file and write the answer back, while your team keeps working in the same folder. No exports, no sync jobs.
1from baseseat import Baseseat
2
3with Baseseat() as drive:
4 hits = drive.files.grep(r"TODO\(\w+\)", glob="**/*.md")
5 todos = "".join(f"- {h.text}\n" for h in hits.results)
6 page = drive.files.glob("todos.md").files[0]
7 drive.files.write_bytes(page.document_id, todos.encode(),
8 revision=page.revision, append=True)
% python todos.py
Found 3 TODOs in 2 files, added to todos.md (revision 44)
Sandboxes
Give every task its own sandbox, and hand sandboxes a token for just the folder they need. They all work in the same files at the same time, and you see every change.
0sandboxes in drive/launch
Scoped to one folderRead or read and write5 minutes to 12 hours
Why it feels multiplayer
Start in one tool. Finish in another.
Every save builds on the latest version. Nothing gets lost.
Mount it on macOS or Linux. Any program works.
Words, patterns and regex across every file.
It shows up like any file on your Mac. Only the bytes you read come over the network.
Each agent gets only the folders you pick.
Pricing
Bring every agent you like. You pay for the people on your team, and for heavy work beyond what the team shares each month.
Questions
Baseseat is a multiplayer file system for teams and their AI agents, built for massive scale. Like a shared doc, but for the whole folder: you, your teammates and agents such as Claude Code, Codex and Cursor work in the same files at the same time, with the access you choose.
A file system that many people and AI agents use at the same time, like a shared doc but for a whole folder. Everyone sees the same files as they change, and a write made against an old version is refused instead of overwriting someone else's work. Where it can, Baseseat merges changes for you behind the scenes: when two edits to a text file don't overlap, they are combined automatically. Baseseat is built for this at agent scale.
An agent drive is one set of files that people and AI agents work in together, through MCP, the CLI, SDK or API. Baseseat adds folders, scoped access, revision checks and built-in keyword search.
Most cloud storage was designed to keep huge amounts of data cheaply, with a business model built around people opening a file now and then. Baseseat is built for the work itself: one set of files that you, your team and your agents use at the same time, open from anywhere in the world in about 50 ms on average in our own testing, and live while thousands of agents read and write. It is a hot workspace rather than a cloud archive, so you pay mostly for the work and storage stays cheap. A much cheaper archive tier for videos, archives and anything you rarely open is planned; you will be able to pull files from it into your workspace when you want to work on them.
You get shared folders, access controls, keyword search and tools for agents. Programs can read byte ranges without downloading whole files.
People and agents share a drive within its limits. Rate limits and sandbox-token limits apply; see the documentation for current values.
No. Baseseat is fully headless: you, your team and your agents work from Claude Code, Codex, Cursor, a terminal or Finder. The Console is there when you want to look around.
Claude Code, Codex, Cursor and any client that connects over MCP. People use the same drive from the Console, the command line or the Python SDK.