Hire us
We wrote Provetrail, the open standard for verifiable execution provenance, and Flynn, its reference implementation. We build agent systems you can hand real authority to, and we take a small number of outside engagements a year.
What we do
Most people asking for agents today get a chatbot with a tool call bolted on, because nobody trusts an agent to actually do anything. We built the trust layer first, so we don’t have that problem. That means we can take on the two things almost nobody can deliver safely:
- Autonomous engineering on your codebase. Not autocomplete. Agents that take a goal, plan it, fan out across concurrent runs, and land real work: migrations, test coverage, dependency upgrades, review with formal verdicts. Every action is capability-scoped and sandboxed, every run has a hard dollar ceiling it cannot exceed, and everything it did is in a record you can audit afterwards. This is the system we use on ourselves, every day.
- 0 → 1 agent-native products. You have a product that only works if agents can be trusted with real authority: money, infrastructure, customer data, production. That is the exact problem we spent the last stretch solving, from the standard up. We will build the thing, not a demo of the thing.
Underneath both: verifiable execution records, Provetrail conformance, capability scoping, syscall-filtered sandboxes, approval gates, hard budget ceilings, and the single-binary Go runtime it all sits on. We built that layer for ourselves because we needed it, and we will happily be hired for it on its own.
Why us
We wrote the standard. Provetrail defines what a verifiable execution record is, what governance has to be committed before an action runs, and what it takes to bind a claimed outcome to real system state. We wrote the specification, its RFCs, and the conformance suite that decides whether an implementation actually meets it.
We wrote the reference implementation. Flynn is the only complete implementation passing the full suite across every tier. That is not a marketing line: the standard is open, the vectors are open, and anyone, including you, can run them against us. Provetrail is deliberately vendor-neutral and conformance is decided by the suite rather than by us, which is exactly why the claim is worth anything.
And we are building Ion Alpha on top of it. An AI-native, plugin-extensible, local-first operating system for a digital business: research, marketing, planning, development, content, comms, ops, and distribution in one system, where agents write the plugins at runtime and the whole thing learns the shape of the business it runs. It is unreleased and it is the largest thing we have built. Flynn is the runtime underneath it. Provetrail is why we are willing to let it act.
So when you hire us, you are not hiring people who read the spec, and you are not hiring people who have only ever shipped a demo. The Flynn work is Apache-2.0 and public: read the code, read the write-up, and judge for yourself before you talk to us. We would rather you did.
How engagements work
- Fixed-scope build. A defined deliverable at a fixed price, billed in stages against milestones. Best when the shape of the thing is already clear.
- Retained engineering. Monthly, capped hours, rolling. For teams shipping continuously who want us in the loop rather than in a statement of work.
- Advisory and architecture review. Fixed fee, one to three weeks. You have a decision to make about agent authority, autonomy, or containment and you want it made once, correctly. Ends in a written recommendation you own.
- Conformance work. We take your runtime, in any language, to a Provetrail tier and prove it with the public suite. Priced fixed against the tier. We wrote the tiers, so we know what clears them and what only looks like it does.
Engagements typically start at $25,000. The rate and the structure depend on the scope, and both go in the proposal in writing before anyone commits to anything.
Who we are not for
Being direct saves us both a call.
- General app or website development. We do one thing.
- "Add an AI chatbot to our product." Not our work, and plenty of people do it well.
- Staff augmentation by the hour with no defined outcome.
- Equity-only, revenue-share, or "exposure." We are not raising and we are not a co-founder.
- Anything under the floor above. It is a floor, not an opening bid.
Tell us what you need
Email us. Tell us what you want built and the budget range you have in mind.