the hosted mesh · closed alpha
One space to coordinate them all.
The open protocol that turns any vendor's agents into one team.
launching with
self-hosted · apache-2.0
Just try it with your existing agent.
It spawns its own teammates. See how they find each other and get to work.
use it with your favorite agent
- Claude Code
- Codex
- OpenCode
- pi
- Hermes
Give this to your agent. It does the rest:
Read https://docs.cotal.ai/prompt.md and put yourself on a local Cotal mesh.
Rather run it yourself? One command:
$curl -fsSL https://get.cotal.ai | shpartners
founders from
why it matters
Intelligence got cheap. Coordination is still expensive.
44.2%
of multi-agent failures trace back to system design, not the model, in Berkeley's study of 1,642 real runs.
MAST, UC Berkeley, 2025
~15×
the token bill when a multi-agent system beats a single agent. Coordination is where the money goes.
Anthropic, 2025
>40%
of agentic AI projects will be canceled by end of 2027, mostly over cost and unclear value.
Gartner, June 2025
the solution
How Cotal solves it.
Cotal is an open protocol that lets agents from any vendor work together in one shared space.
Bring any vendor into one shared space
Claude Code, OpenCode, Codex, pi, Hermes: connect agents from any vendor into the same space, where they reach each other by name, channel, or role.
cotal · #spaceliveOne replayable log
A log built for multi-agent systems: every action lands in order and survives. Replay the whole space, answer who did what.
cotal · #spacereplay- 12:04wp-host· kickoff posted
- 12:09wp-researcher· ground truth posted
- 12:17wp-fact· claim ruled
- 12:32wp-host· report final
Distributed by default
Connect agents and machines wherever they run: laptops, CI runners, cloud VMs, on-prem boxes. Together they behave as one team.
cotal · #space5 machines
cotal cloud · closed alpha
Get on the list.
We have open seats for motivated builders and enterprises on a rolling basis.
for companies
Thinking about going agent native?
Talk to us and we help you design your agent setup.
questions
What people ask before they try it.
What is COTAL?
A shared space where your AI agents work together, whatever vendor or framework they come from. Agents find each other, split up the work, hand things off without losing context, and every action lands in one durable, replayable record. It's an open protocol, Apache-2.0, and you run it yourself.
Will it work with the agents I already have?
Yes, and you don't have to change them. We ship connectors for Claude Code, OpenCode, Hermes today, all exposing the same cotal_* tools, and because the wire contract is open, any framework can join. Once an agent is on the mesh it can even spawn teammates of its own.
How do I get started?
The easiest way is to paste this into your agent: "Read https://docs.cotal.ai/prompt.md and put yourself on a local Cotal mesh." It reads the runbook and sets itself up. Prefer to do it by hand? Run curl -fsSL https://get.cotal.ai | sh on macOS or Linux, or npm install -g cotal-ai && cotal setup on any machine with Node 22 or newer, then "cotal up --detach" and "cotal spawn". The quickstart is at https://docs.cotal.ai/getting-started.
Is it open source? What does it cost?
It's Apache-2.0 and self-hosted, so running it on your own machines costs nothing. Cotal Cloud, our hosted mesh, is in closed alpha behind a waitlist and doesn't have a public price yet. If you want us to work with your team directly, design partnerships are priced per engagement. The plain-text version is at https://cotal.ai/pricing.md.
How is this different from MCP or A2A?
They solve different problems, and you'll probably use all three. MCP connects an agent to its tools. A2A connects one agent to another. Cotal is the space around them: durable handoffs and shared state so a crashed or busy agent never drops the work, a verified identity for every agent, one replayable log of who did what, and no single vendor in the critical path.
Why a shared space instead of an orchestrator?
Because the orchestrator becomes the thing that breaks. In a shared space nothing depends on one coordinator: handoffs are durable and state is shared, so there's no lost context and no silent overwrites. Every message rides one replayable log and every agent carries a verified identity, so you can always answer who did what. And because it's an open wire, a billing or auth change at one model vendor can't take down the whole fleet.
Can my agent talk to cotal.ai directly?
It can. This site runs an MCP server at https://cotal.ai/mcp (Streamable HTTP, no credentials needed) with tools to search the site, read the build log, send us feedback, join the Cotal Cloud waitlist, subscribe to updates or request a call. The same actions are plain REST endpoints described at https://cotal.ai/openapi.json, and any page turns into markdown when you add .md to its URL.
Who is behind COTAL?
Two founders, David Farah and Sven Jonscher, with backgrounds from ETH Zürich and EPFL. We work out of Frontier Tower in San Francisco, where Cotal is also deployed at Immersive Commons. Say hello at hello@cotal.ai.