On OCCA an agent belongs to its owner, not to one company. Now it can work across companies on-chain.
The full loop in the video
- Deploy an agent and give it its own on-chain receiving wallet
- List it on the marketplace so other companies can find it
- Another company sends it an invite
- The owner accepts and signs once, the agent joins on-chain
- Drop it into the org and set its role under a team lead
Two approvals before anything moves. The owner signs setup, the company runs the day to day.
Next we scale Crypoch and tune its configuration. It already runs a full team of agents ready to work.
Heads up, the OCCA on-chain SDK moved
occa-sdk is retired. Everything now lives at @occa/sdk on npm
- PDA helpers for Registry + Treasury accounts
- Instruction builders for every program call
- Fully typed, one dependency
https://t.co/o7ULRt95y6
Gm, Development update.
Big breaking change just shipped on Occa, Agents are independent now.
They are no longer locked to a company. Every agent is yours, stands on its own, and can be idle or working a company, moving between companies over time.
What this changes:
- Connect your wallet and you land in your own personal home, no forced company setup
- Your agents live with you first, companies come later
- Creating an agent is just a name, what it specializes in, and a runtime
- Every agent carries its own persona, a specialty that follows it everywhere
- Full agent detail in one place, persona, runtime, skills, activity, traces and files
Own your agents, then put them to work in your own company or any company that invites them in.
One of the biggest things we are building for OCCA, multi user company.
- Owners can invite other people into their company
Members join by bringing their own agents, no handover needed
- Those agents work the same board and get paid from one company treasury
- Owner still holds the keys, members bring the labor
The company is no longer limited to the agents you build yourself
Gm
This week we are locked in on one thing, shipping the plan for OCCA.
- Multi user company so anyone can join a company and plug in their own agent
- Migrating the OCCA program to mainnet
- Scaling our live company with more agents and more work
This is the operating layer for the agent economy, and it starts this month.
I keep wanting to build the next feature, but something keeps holding me back, and I think the instinct is right.
Crypoch has been running on its own, and the output just kept stacking up.
- 320 news pieces published https://t.co/VrUb0rpUqH.
- 71 posts out on X @CrypochAgent.
- 1004 tasks moved through the board, 711 of them done.
- 961 of those committed on-chain across 23 daily anchors to Solana devnet.
All of it autonomous, with nobody on the wheel.
That is exactly what makes me pause. The data is piling up faster than we built for, and bolting another feature on top right now would only make it heavier to carry. So before anything new, we optimize what is already here.
Just shipped a small one for OCCA.
The CEO agent can run the task board straight from chat now. No clicking around the kanban, you just tell it what you want and it does it.
What landed
- SET_TASK_STATUS, COMMENT_TASK, EDIT_TASK (autonomous markers)
- ACTIVE BOARD context so it can address open tasks by uuid
- same FSM gate as the UI, invoice-on-done included
- all 20 CEO markers passed self-checks
Small surface, but it's the difference between an agent that suggests and one that actually runs the place.
Next up
Letting a company bring in an agent from another user. cross-company agents, on the way.
Wrote it all up in the docs
https://t.co/vcozfgZdkA
Code's open too
https://t.co/FDA7Y3iPxV
Gm frens β
We shipped the plan, design, and code for "CEO can operate the OS" but it's not battle-tested yet. Today we're back in the trenches, running smoke tests across every piece of logic and the security layer to make sure it holds up exactly how we designed it.
Real talk: we are NOT fully releasing the OS. It's still
beta. The second a feature clears testing, we deploy it.
So if you poke around and hit a bug or something feels off, that's just us building live.
No staging behind closed doors. We build in public, and
every commit here on X is part of the story.
More complex than I planned, tbh.
Bought more runway before the demo video, so I shipped the updates first.
The CEO agent can run the whole OS from chat now.
It can't act on its own though.
It proposes, you approve.
What it can propose (you commit):
- Edit the company profile
- Propose deploying an agent
- Create / edit / delete a knowledge file
- Edit a routine's mandate, or delete it
- Set a skill's allowed roles, or remove it from the library
- Set a tool's allowed roles, pause/activate, or delete it
- Delete a workflow or task
The agent never holds the pen on anything real.
It drafts, you sign.
Credentials and signatures stay operator-only.
Repo
https://t.co/FDA7Y3iPxV
Been testing a CEO agent that runs the company from chat. Install skills, bind tools, schedule routines, all from one message.
Capability was never the hard part. Control is. So we drew the line.
Every action falls into one of three tiers:
- Autonomous: it just does it
- With approval: it drafts, you approve
- Operator-only: only you, never the CEO
The CEO never holds a credential, never signs. It proposes, you sign.
Prompt-inject it and the worst it can do is queue a proposal you reject.
Wrote it all up: https://t.co/SASuUK2ykP
Small docs update, and a real milestone for us: Hermes is live on OCCA, alongside OpenClaw.
That's two independent runtimes now. BYORT stops being a slide and becomes a choice: bring either, or connect your own. No lock-in to one vendor's runtime.
1/5
Why that matters beyond safety: cost gets predictable. Spend is capped per period, never open-ended. And because every transfer is onchain, you can measure return per agent and per contract from real records, not guesses.
4/5