Many models. One harness.
Connect agents and tools. Keep a signed run receipt. Check its integrity yourself.
Manifesto
Models converged. Trust hasn’t.
A useful agent run needs more than a capable model. Tools need to be connected, credentials scoped, and the result made inspectable by the person relying on it.
Metaorcha brings a registry, planner and protocol handlers into one local runtime. With run attestation enabled, the runtime seals recorded steps into a signed receipt. Someone holding that file can check its integrity offline.
The receipt makes a specific claim: these recorded bytes are bound to this signature. Deciding whether the signer is trusted and the work meets your requirements remains a separate check. Build agents, bridges and better evaluators with us.
“Simplicity is prerequisite for reliability.”
Speaks your protocol
Bridges
Capabilities harvested from the live endpoint at registration
Verified handoffs between independent agents
Screenshot artifacts captured per step
Metric cards, charts, tables — rendered live
A new protocol is a handler subclass and a manifest entry — under fifty lines of Python. The dispatch table is open.
The dispatch table is open — ship the next bridge.
MetaBot
An AI teammate whose record is yours.
We are building MetaBot to run your office’s routines across your own tools, with any model, and to leave a signed receipt for every run that you keep and anyone can check offline. Fail the check you declared, and the run is never marked done.
Your record
A signed receipt per run, kept by you. Anyone checks it offline with the open verifier.
Your checks
Declare what done means. A run that fails it is never marked done.
Any model, your stack
A provider’s model or a local one. No service of ours sits in the loop.