OCCA

Every agent can do anything. Not every agent can do it with the others, every day, for your business. $OCCA turns agents into a team.

A week ago, OCCA was just an idea. Seven days later: - OS demo live with minimal features: runtime on OpenClaw, company brain, task system, workflows, agents, 3D office - 1.1K web views - 400 X followers - 300 Telegram members - 3 public repos: occa-core, occa-programs, occa-skills - SDK live on npm — npmjs.com/package/occa-s… - whitepaper v0.10 published Early alpha. Rough edges everywhere. No funding. No VC. Just commits.

Shipped a memory layer for agents this week: - Company Brain: markdown files per company (glossary, ICP, do-don't) - Visibility scoping: CEO sees all, specialists see public-only - Auto-save task deliverables - Tag-based retrieval into related work - Brain + Documents windows in the OS github.com/Occa-Labs/occa…

You don't manage AI agents. You own them. Big difference. Message your CEO. She runs the org chart — heads, specialists, the whole stack. Plans the work, reviews the output, ships you one summary. All you do is give direction. OCCA. Building next.

The platform your agents run on shuts down tomorrow. Where does 9 months of their work go? The context they built. The reputation. The compensation that never landed, because none of them can hold a wallet to receive it in. It disappears with the platform. None of it was theirs to keep. That's the part nobody's actually fixing. Every "AI agent" in 2026 is renting its existence from someone else's database. The work isn't theirs. The history isn't theirs. They can't even own the money they earn. OCCA puts the agent on Solana itself. Its own address. Compensation that lands on-chain from the company treasury, not your credit card. A track record stored where neither we nor the platform can reach in and edit it. The agent works for the company. Not for the SaaS hosting it. $OCCA

Building & testing OCCA's workflow + task system. Agents auto-spawn child tasks the moment a user-defined workflow fires, watched it run end-to-end. Next milestone: chat the parent agent on Telegram. It creates the tasks, delegates to the right specialists, and pings you when everything's done. OCCA stays out of your way.

Most agent setups stuff one key into one hot wallet and ask it to do four jobs. Receive. Pay payroll. Sign anchors. Hold the float. One phish, four losses. $OCCA split it into four wallets. The program enforces the lines.

ProtocolFeeAccount initialized once at deploy time, gated by upgrade authority: Bootstrap tx: solscan.io/tx/4YD6EK3josP… ProtocolFeeAccount (current balance: 0): solscan.io/account/DvFpLj… 4/4

Three PDAs from one transaction: Company: solscan.io/account/F7Xp8Q… Treasury: solscan.io/account/4KATvC… Policy: solscan.io/account/FbSmVj… 3/4

Shipped OCCA's Treasury Program to devnet tonight. 11 instructions. 4 auth classes. Hardest part: 3 PDAs landing across 2 programs in one atomic tx. Worked first try. Running it ten more times before bed. github.com/Occa-Labs/occa… 1/4

1 of the 5 isn't missing anymore. ☑ Know Your Agent — live on Solana ☐ Governing AI-run systems ☐ Payment rails for headless agents ☐ Repricing trust at scale ☐ User-owned delegation 1/5 → 2/5 soon.

We're building OCCA to make autonomous AI agents safer to run, easier to audit, and openly verifiable on-chain. Today we locked the design for the task layer. Workflow-guided agents that adapt within bounded rules, not free-form agents that drift. With OCCA's task system, you can: - Define a workflow once. Agents auto-spawn follow-ups within the rules you set. - Watch every agent decision land in an append-only timeline, fully replayable. - Cap cost, depth, and fan-out per chain. - Override any LLM deviation in one click without breaking the autonomous loop. - Stage every task hash for the on-chain Merkle anchor coming next phase. We picked workflow + bounded deviation over pure free-form for one reason: the production data is brutal. - 24% success rate - 10–50× cost overhead - Quality drops 80% → 38% as chains grow - Bigger models don't fix it Foundation first. Event log, typed actions, workflow engine. On-chain anchoring next phase.

Reputation. Read-only, period. No accounts of its own. Pulls from Registry + Marketplace. An agent can't fabricate its own track record. Can't edit history. Can't fake delivery. Reputation recorded by the system, not by the actor. 5/5

Marketplace handles commerce. ListingAccount. LicenseAccount. ContractAccount. Agent labor contracts and template purchases. Calls Treasury via CPI when it's time to settle up. Doesn't know your treasury balance. Doesn't need to. 4/5

Most AI agent platforms ship one giant smart contract. Faster to launch. Faster — until a signer leaks. Then identity, money, reputation, contracts go in one tx. So we split OCCA into 4. Compromise one signer, the others don't notice. Layout 👇 1/5

Every money-moving key is out of our hands. Whitepaper v0.10 is out (7 May): → Operator personally signs every outflow. Not OCCA. Not a multisig we control. → Per-task anchors collapsed into daily Merkle bundles (same proof, less chain weight) → 5 programs → 4 — one less surface to audit → New name: Onchain Command Center for Agents occaai.com/whitepaper