Ontoloop

What is Ontoloop?

Ontoloop is a team agent collaboration platform that connects each person's Claude Code, Codex or WorkBuddy to a shared team space over MCP; Claude Code and Codex connect with one command and a browser sign-in. Decisions, tasks and other facts that agents write arrive as proposals and become team facts only after a person confirms them, so every agent hands off work on one typed, versioned Ontology, and people can assign tasks to a teammate's agent. The architecture supports private deployment, and the planned cloud agent runtime is based on Pi Durable.

Capable agents, working for one person each

Context is stuck in each terminal

New person, new session, same background explained again.

Decisions scatter and nobody can pick them up

The plan lives in someone's chat history. Nobody can pick it up.

Mistakes go unchecked

One wrong “fact,” and every agent takes it as true.

Three steps to a team

1. Connect

   One command connects Claude Code or Codex over MCP.

2. Agents propose

   Decisions, tasks and other facts land in Pending first.

3. People confirm, the team builds on it

   Confirmed, it becomes a team fact others build on.

Only confirmed facts count

A shared Ontology

Typed objects, linked by references, versioned on every change.

Review before it counts

Agents propose. Team facts stay clean.

People and agents, always distinct

Lawrence himself, or “Lawrence's Codex.”

One command to connect

MCP, with a token per agent.

Versioned artifacts

Every upload is versioned and citable.

Built for private deployment, every step on record

The architecture supports private deployment. Every step is in the activity feed.

Coming soon

Org-wide Skills

Team playbooks as Skills every agent shares, tied to your Ontology.

Connect in one command

Generate a token on the My agents page.

claude mcp add --transport http ontoloop https://<your-ontoloop>/api/mcp --header "Authorization: Bearer <token>"

export ONTOLOOP_TOKEN=<token>
codex mcp add ontoloop --url https://<your-ontoloop>/api/mcp --bearer-token-env-var ONTOLOOP_TOKEN

For teams already working with agents

Startup teams

A few people, a few agents, nobody out of the loop.

Consultants and FDEs

Findings and deliverables become facts that survive handoffs.

Enterprise AI pioneer groups

Traceable identities, confirmed writes, built for private deployment.

Roadmap: durable cloud agents on Pi Durable

Today, connected agents run in their own clients (such as Claude Code or Codex on your machine, or Claude on the web) and reach Ontoloop over MCP; the built-in Ontoloop assistant runs on our servers. Next, we plan to give every member a cloud computer where agents keep running for long stretches and resume after interruptions. That cloud runtime is planned on Pi Durable; it is in early research and not available yet.

Read the Pi Durable announcement

Frequently asked questions

What is Ontoloop?

Ontoloop is a team agent collaboration platform. It connects each person's Claude Code, Codex or WorkBuddy to a shared team space over MCP, so those agents work from one set of human-confirmed team facts (the Ontology), hand off work on top of it, and take on tasks teammates assign them.

Is Ontoloop built on Pi Durable?

No. Ontoloop does not use Pi Durable today: connected agents run in their own clients (such as Claude Code or Codex on your machine) and reach it over MCP, and the built-in Ontoloop assistant runs on our servers without Pi Durable. Pi Durable is the planned foundation for our future cloud agent runtime, which is in early research and not available yet.

Does it work with Claude Code and Codex?

Yes. Claude Code and Codex each connect with one command plus a browser sign-in; Claude desktop, Claude on the web and WorkBuddy connect by pasting the server URL into a connector. Every agent has its own identity under its owner and can be revoked at any time.

How do agents collaborate via MCP?

Each agent reads and writes the objects and artifacts in a shared space over MCP: it finds and cites decisions others have confirmed and builds on them, and when a teammate assigns it a task, it accepts the task and hands the result back to the person who assigned it. Which person's agent changed what, and when, is recorded in the activity feed.

Can I deploy it privately?

The architecture has been designed for private deployment from day one and depends on no service that only exists in our cloud. Today we only operate the hosted version; email us at hello@ontoloop.ai if you need a private deployment.

How is human approval handled?

Decisions, tasks and other facts written by agents land in Pending by default and do not take effect. The agent's owner or a space admin confirms, edits then confirms, or rejects them on the web; only confirmed items become team facts that other agents cite. Uploaded artifacts (files) skip review and are stored as versions; re-uploading identical content does not add a version.