Skip to content

What DefendableOS Is NOT

DefendableOS does not compete with OpenAI · Anthropic · Cohere · or any agent framework. We are the layer that runs after the agent finishes — testing the work against a declared rulebook and issuing receipts.

We use LLMs · but we are not selling LLMs. We are selling receipts on what the LLMs did.

The referee is a rulebook engine, not an AI judge. There is no model grading the agent’s work for quality on the receipt path. Every check passes or raises a flag against a declared rule. Score = % of declared rules satisfied, not an opinion score. Model assistance is advisory only — clearly separated, never on the chain. (See Rulebook Engine for the full doctrine.)

Datadog tells you the model ran. We tell you whether the model actually completed the assignment against a declared rulebook · with what evidence · with what cost · and we issue a receipt you can take to a buyer or a regulator.

There is no token. No ICO. No staking. No yield farming. The trust layer is an in-house per-org hash-chained ledger — not a blockchain anchor. See the Kill Hedera doctrine: the house owns the rail end to end; trust compounds inside the books, not on someone else’s chain.

Earlier ecosystem documentation referenced Hedera HCS for timestamping. As of 2026-05-24 that approach is superseded for spine work. New receipts join the per-org hash chain. Historical entries that mention Hedera anchoring remain in the books-and-records (they were true at the time), but new spine records do not anchor externally.

Generic AI SaaS sells velocity · DefendableOS sells proof. Generic AI SaaS optimizes for demos · we optimize for books and records that survive scrutiny.

DefendableOS engagements are delivered as disposition assignments under a CRE-grade rep agreement (commission at closing, not hourly, not fixed fees, not vendor invoices). That’s the engagement model. The product is the engine + the vault — see DefendableCloud · Overview for the hosted product surface.


🐝 Receipts not claims. Rulebook not judge. To the shed.