The verifiable AI surface for every company.

Every reply backed by a typed function a human signed. Every receipt verifiable without trusting us.

AI is shipping into every workplace. The current generation of tools — Copilot, ChatGPT Enterprise, internal RAG — answers questions by paraphrasing context. That works for drafts. It fails the moment a regulator, an auditor, or a procurement officer asks who is accountable for this answer.

Ctrl AI replaces the paraphrase with a typed call. Every reply our system gives is grounded in a Ctrl — a callable function with a name, an inputs schema, an outputs schema, and a signature from the human who took responsibility for it. The invocation produces a receipt — an Ed25519-signed proof that's appended to a public, chained-hash transparency log. Any third party can verify the receipt without trusting our servers.

This is the substrate the EU AI Act, DORA, SOC 2 Operating Effectiveness, and every internal audit team is moving toward. We're building the universal layer.

• Moat 1

  Open-source verifier

  Anyone can verify any receipt without running our code or trusting our servers. The verifier is open-source, the spec is frozen at v0.2.0, and the transparency-log roots are pinned. We win the standards game by being the surface that's worth standardising on — not by lock-in.

• Moat 2

  Per-customer corpus

  Every customer's Ctrl library encodes their policies, departments, roles, and accountability chains. That corpus compounds — it gets more valuable the longer they're on the platform, and it doesn't transfer to a competitor. Switching cost lives in the corpus, not in the contract.

• Moat 3

  Signed receipts on a public log

  Every successful Ctrl invocation produces a receipt anchored in a chained-hash log. The log is public. The signatures are Ed25519. Auditors, regulators, and prospects can verify the entire history of an org's AI actions in seconds. That artifact is the durable trust object — and it doesn't exist anywhere else.

The dispatcher is the single source of audit truth. Every code path that wants to invoke a Ctrl — the /ask agent, the MCP server, scheduled triggers, composed sub-calls — goes through one function. Authority for every action is recorded in the receipt's actorChain: who signed, who delegated, what role they hold, what time window they were operating in. Compliance teams stop reading logs; they read receipts.

Pick any receipt from the public Trust Portal. Verify the Ed25519 signature against the public JWKS. Re-derive the leaf hash. Walk the transparency-log chain up to the current head. If any of those checks fails, the system is lying. The steps below run against the seeded demo org acme-saas-demo — same mechanics in production.

npx @ctrl-ai/verify <receipt-id>

# Fetch the receipt + inclusion proof
curl https://ctrlai.com/api/receipts/<receipt-id>

# Fetch the public JWKS
curl https://ctrlai.com/.well-known/ctrlai/receipt-keys

# Fetch the current log head
curl https://ctrlai.com/api/transparency-log/head

# Verify Ed25519 signature locally + walk the chain
node verify.mjs <receipt-id>

• Magic-moment onboarding. Drop a folder of policy docs at /try/upload. 90 seconds later: org structure extracted, Ctrls proposed, first cited answer rendered. Conversion to a real workspace is one magic link.
• Compliance Packs. EU AI Act Starter (Articles 12/13/14) is live. DORA, SOC 2, GDPR, HIPAA, B2B SaaS Rev-Ops are next. Each Pack is a versioned, expert-co-signed bundle of Ctrl templates a customer installs with one click. Tier-priced.
• MCP-first integration. Cursor, Claude Code, Claude Desktop, and any future MCP client invoke our Ctrls through /api/mcp/[orgSlug]. The destructive ones still flow through the gate loop — every external AI client respects the same human-in-the-loop contract.
• Open-source verifier as the wedge. We publish the verifier + the spec under a permissive licence. We don't win standards by hoarding them; we win by being the surface worth standardising on.

• EU AI Act applies from August 2026; high-risk systems need documented controls today.
• DORA (financial-sector) and SEC AI risk-management guidelines force boards to ask: who is accountable for what our AI does?
• SOC 2 Type II auditors increasingly fail orgs without AI-action audit trails. Receipts are the answer.
• MCP standardised the AI-tool-call surface in 2025. The connectors are commoditising; the surface that decides who is allowed to call what is the next layer up — and it doesn't exist yet.

$1M on a SAFE at $10M post-money cap (10% dilution). YC-standard SAFE, no discount, no MFN. One round by design. The platform is built; we're funding distribution and the SF presence, not engineering. Target: default-profitable within 18 months. Any further raise is on customer revenue, not dilution.

Shape: one strategic anchor ($300–500K) plus 3–5 operator-angels who built compliance, AI-infra, or developer-tools companies. 30 days from first conversation to close.