01 · Instructions
Each tool has its own rulebook
Claude Code reads CLAUDE.md, Codex reads AGENTS.md, ChatGPT keeps custom instructions. Each copy drifts on its own.
Space serves one playbook to every tool.
Write how your team does the work once. Space serves it, with what your company knows, to Claude, ChatGPT, Codex and every AI tool your people use.
Space hosted stopped taking new accounts on 5 November 2026. Existing accounts keep working, and the open-source edition stays free to self-host.

The problem
Your people use Claude, ChatGPT and Codex side by side. Ask each for the same client report and you get three structures, three sets of checks and three levels of quality.
01 · Instructions
Claude Code reads CLAUDE.md, Codex reads AGENTS.md, ChatGPT keeps custom instructions. Each copy drifts on its own.
Space serves one playbook to every tool.
02 · Context
One has last year’s report, another last month’s notes, a third nothing at all. Each fills the gaps its own way.
Space gives every tool the same knowledge.
03 · Format
Without one template and one checklist, the structure, the order and the checks change from one run to the next.
Space applies your template and your checks.
See it work

01 Pick a tool
02 Give it your playbook
03 Add what your team knows
04 Change the playbook once
Audit report, Dupont Industries
No checklist applied
Brief: write the audit report for Dupont Industries4 tools · 4 structures
Illustrative example. Models still vary; the playbook removes the differences that come from instructions, context and format.

Leonardo da Vinci · Codex Leicester · c. 1510
How it works
#1 Write
One kind of work: the steps, the template and the checklist your best people use.
#2 Offer
When someone asks Claude, ChatGPT or Codex for that work, the agent offers your skill and applies it once they agree.
Your Space has client-audit for this. Use it?
#3 Update
Edit the skill in Space. Every agent that reads it uses the new version on its next run.
Systems
A system is work your team repeats, such as posting, outreach, SEO or a development lane, with a goal metric, a target and a review cadence. Its lab is where new ideas are tried.
01 · Lab
Your people and their agents add hypotheses and lessons to a system’s lab. Lab entries are never recalled, so an untested idea never reaches an agent as a rule.
Untested ideas stay in the lab.
02 · Verdict
An entry becomes a verdict, Works or Does not work, only with measured evidence or the owner’s call, and a source. Demote it and it leaves recall at once, with its history kept.
Only verdicts reach your agents.
03 · Reuse
Agents record each use of a verdict. Each system shows its health, its goal and its reuse share: how many of its verdicts were used within 30 days of promotion.
Reuse share, per system.
Illustrative example. The entries and systems shown are sample content.
What a Space holds

#1 Memory
What was decided and why, linked to where it came from. Read, correct or delete any memory from the web app.

#2 Documents and data
Briefs, reports and templates with every revision, and client records as tables your agents can read and update.

#3 Every AI tool
Claude, ChatGPT, Codex, Cursor, Gemini and more connect over MCP. Nothing to migrate, nobody to retrain.
Under your control
Questions
A playbook for one kind of work, saved in your Space: the steps to follow, plus files such as a template and a checklist. When someone asks a connected agent for that work, the agent offers the skill and applies it once the person agrees. Skills come with the Teams plan.
Claude Code, Codex, Claude, ChatGPT, Cursor, Gemini, Cline and Hermes have each been tested with Space. They connect over MCP, the standard way AI tools reach outside knowledge.
Only the Space owner. Agents read a skill and offer it; nobody’s agent can rewrite it for everyone.
You choose which tools connect and whether each can read or write. Memory text is encrypted before it is stored, and search ranks an index of it, so finding a memory never decrypts your whole Space.
You could. The hard parts are serving the same instructions to every tool, keeping each team’s knowledge separate, and searching it without decrypting it. That is the part Space runs for you.
Pilots are closed. Space hosted stopped taking new accounts on 5 November 2026, and the open-source edition is the way to run it today.

Pricing
5 November 2026
Space hosted stopped taking new accounts on 5 November 2026. Existing accounts keep working, and the open-source edition stays free to self-host.
Run it yourself from the open-source repository.