The verified-context layer
Restormel is the verified-context layer for AI products. Every claim an agent uses is bound to a verbatim quote in your source — provenance-traced, quality-gated, and re-checkable by you or your auditor.
An agent answers a question. Every claim is either bound to a verbatim span in your source — or it never reaches the agent. Click a claim to trace it back.
Q How long must we keep customer records under Policy 7.2?
Data Retention Policy — §7.2
7.2 Data Retention. Customer records containing personal data MUST be retained for a minimum of six (6) years from the date of the last transaction, after which they are subject to scheduled deletion. Records under an active legal hold are exempt from scheduled deletion until the hold is formally released by the Data Protection Officer.
Supported
Bound to a verbatim quote, re-checkable at read time.
trace:rstrml:8f31c2
Illustrative example — not live data. Spans, sources, and trace refs above are hand-authored to show the contract.
Most AI stacks hand your agent context and hope for the best. When an answer is wrong, there's nothing to trace — no span, no source, no way to tell a grounded claim from a fluent guess.
Context goes in. Answers come out. Nobody can point to where a claim came from.
Every supported claim carries a verbatim quote, a source, and a trace — re-checkable without a model.
Knowledge passes a published quality bar before any agent can read it — and stays re-checkable after.
Bring your sources. Restormel extracts claims and binds each one to the exact span it came from — quote, offsets, and source hash.
Each claim is checked against its bound span, and an independent model family re-checks the extraction. Claims that don't bind are excluded, not blended in.
Agents read only verified context — every claim carrying its verification state, citation, and an exportable provenance trace.
Keys is the control plane for your AI traffic. Connect turns your sources into verified context. Both sit under the same provenance frame.
Model routing & BYOK custody
Control which model runs on every request — visual routes, fallback chains, and policies on your existing gateways.
# resolve chain config
route: production-default
primary: openai/gpt-4o-mini
fallback: anthropic/claude-3-5
policy: cost-cap-passAgent-ready knowledge infrastructure
The control plane for verified context — provenance-traced, quality-gated knowledge your agents can be held accountable to. Every claim carries bound evidence, a cross-model validation record, and an exportable audit trace.
# verified-context layer
ingest: sources → bound claims
retrieve: graph + vector hybrid
verify: bar ≥90% supported / ≤2% unsupp.
domain: configurable schemaRegulated, audit-bound, and high-stakes knowledge work — where "the model said so" isn't an answer you can give a reviewer.
Keep Neon, SurrealDB, and your existing providers. Restormel sits between your stack and your product as a thin verified-context layer.
Your stack
What you already have
Your keys. Your databases. Your providers.
Restormel
Control layer
Signed-in workspace. Domain schemas. BYOK custody.
Your product
Grounded agents
AI products that earn trust by showing their reasoning.
Restormel isn't on general sale yet. Join the Founders Circle — register your email, get a personal access link, and help us prove verified context before we set pricing.