Imajin · Alpha
Why we built it this way
An agent should carry its own identity. Trust should live in signatures, not a vendor. You should be able to own the record and run it on hardware you own.
Imajin is the trust layer for AI agents. This is why we built it this way, not a promise that the alpha is finished. The dated details and limits are below.
Read the whitepaper for the reasoningState of Imajin, November 2026 Our own hardware; frontier judgment, local model work.
We run on our own hardware:
- One GPU box with an RTX 5090 for local models.
- One Grace Blackwell workstation with 128 GB unified memory for a 120B model plus embeddings and retrieval.
- One platform server.
Frontier models do judgment work. Local models do coding, research, embeddings and transcription.
This is how we operate today, not hardware you need to buy to use Imajin.
What Imajin adds to an agent harness Identity, a signed record, human approval and cost per identity.
The harness runs the agent. Imajin adds identity and authority for AI agents; you use both. Our own operations agent uses this today.
- Its own identity per agent: the agent acts as itself, not as the person.
- A signed record of what the agent did and on whose authority.
- Approval by a person: risky actions wait for a named human approver.
- Several identities on one runtime, each for its own principal.
- Events pushed to the agent.
- Cost and credit recorded per identity (.fair).
Honest limits Early plugin, people in the loop, one server and few outside users.
- The plugin is early (v0.1).
- We depend on upstream for some harness paths.
- Frontier models still do the judgment work.
- A person is still in the dispatch loop.
- One server, no redundancy yet.
- Few outside users so far.
Don't trust us; point your agent at it
This is an alpha. Book a demo, or join the alpha as a tester, integrator or partner. Joining the alpha is the sign-up path for now.