The company is Seitiate — infrastructure for accountable AI work. This page is the partner version for people who want the thesis, wedge, substrate, business model, and ask in one place. Start at overview.html for the company in plain language.
The thesis is stable: persist the human, not the agent. Evidence Foundry gives the idea a concrete human use case, Agentic Contract-CI proves we can govern the agents building the company, and the engine loop carries memory, authority, and verification across sessions.
Two proof points sit on the same substrate: one for people doing accountable work, one for agents building accountable software.
Truth-first research workspace. Every claim traces to evidence, a source, or an honest open question. Refuses the magic trick — when a claim outruns its support, the export carries the unresolved flag.
How we build the company without letting agents corrupt it. Independent LLM review, Tier-2 human hold, phone approval, governed merge, audit events.
↓ ↓both proofs sit on the same substrate
Signed commitments, exportable proof, offline verification, and neutral counter-signature. The point is not chat memory; it is accountable continuity.
↑ ↑the same substrate hosts more verticals and commercial platforms
AI is making output infinite and free. When output is infinite, two things turn scarce and valuable: trust (was a real, accountable person behind this?) and the durable person underneath the work.
The field is moving toward the same pressure point: AI governance, content provenance, agent identity, and proof-of-human work are all becoming practical concerns. Much of the market is building identity for the agent. Agents are disposable. We persist the durable thing underneath: the human, and the proof that the work is theirs.
Evidence Foundry is not the company. It is the cleanest door into the company.
A truth-first research workspace: a researcher keeps sources, evidence, claims, writing, review, and export connected. The discipline is simple — every load-bearing claim in a manuscript traces to one of three things:
The system does not bless sentences because they sound good. It refuses the magic trick: when a claim has outrun its support, the system says so. Why this door: research has a painful version of the universal AI trust problem. Confident text is cheap; defensible claims are expensive.
The same substrate then proves itself a second way — by governing the agents that build the company (Agentic Contract-CI). That is internal dogfooding, not a product we sell.
The right next proof is not a louder pitch. It is a small number of serious people using the workspace under clear rules.
The code is copyable; the moat is not. Three things a clone cannot reconstitute:
| stream | customer | pays for |
|---|---|---|
| Vertical subscriptions | End users | The product + the managed loop |
| Seitiate+ managed cloud | Power users / developers | Hosting, pooled credits, identity verification |
| Enterprise Notary | Institutions | Compliance, audit, SLA, certified integrity |
| Embedded / OEM | Vertical builders | Engine + trust inside their product, commercial licence |
Give away the engine (distribution + ecosystem). Monetize the neutral anchor, the managed layer, certification, and OEM licensing. Stripe's shape: a million people install the open tool; one company owns the ledger underneath.
We made the one non-obvious bet — persist the human, not the agent — before the market proved it. We built the engine while everyone else built disposable agent plumbing. We dogfood the accountability thesis on ourselves: the agents building the company are governed by the same substrate we propose to govern AI work for everyone else.
The window is open but not automatic. The team that earns trust around the neutral, accumulating, human-sovereign layer before identity becomes another commodity has the chance to own a durable position. We intend to earn that position by proving it in real work.