Product · How it works

Five agents argue. One engine decides.

Sentinel's AI swarm drafts and challenges a custody policy the way a practice would. The verdict is not the swarm's: it comes from a deterministic Rust proof engine that gives the same answer to the same policy every time, and can be run without any model at all.

I. The swarm

The order in which a policy is questioned.

  1. 01Manager

    Reads the plan and the client's answers. Sets the brief for the two specialists and decides what evidence would be enough.

    1. 02aMarket

      Compares the arrangement to the vault providers and schemes actually available in the client's market.

    2. 02bCompliance

      Tests the same arrangement against the jurisdictions the client operates in: MiCA, SEC guidance, national frameworks.

  3. 03Concierge

    Writes the brief, the document itself, in the language of the person who has to act on it, from what the specialists found.

  4. 04Arbiter

    Not an agent. A Rust proof engine that checks the policy itself and returns the verdict. Nothing above it can overrule it.

A licensed person reviews and approves every plan before it is finalised. The swarm proposes; the engine proves; the person signs.

II. What deterministic means

Same policy in, same verdict out.

Deterministic means exactly this: the engine contains no model, no temperature and no judgement call. Run it today, run it in ten years, run it on any machine: a given policy produces the same result. That is what makes the verdict something a lawyer can cite.

  1. 1Same input, same verdictA given policy gives the same result every time it is checked, on any machine, in any year.ALWAYS
  2. 2No model insideThe engine contains no model, no temperature and no judgement call. Nothing above it can overrule it.ALWAYS
  3. 3Plain codeVerifying a policy is a computation, not a call to a model or a service. The engine has no network dependency of its own.ALWAYS
  4. 4A person signsA licensed person reviews and approves every plan before it is finalised.ALWAYS

III. Inside your own assistant

No new dashboard to learn.

The Sentinel Skill is being built for the MCP-compatible assistants a practice already runs. The same swarm, the same engine, the same brief, from inside the conversation where the client's question was asked. The session below is a sample of what that looks like.

Sample sessionSkill in development› validate this policy for a client in Portugal with two heirssentinel · manager: brief set · succession · EU (MiCA) · 2 heirssentinel · market: two schemes fit this market; one carries the arrangement as writtensentinel · compliance: no conflict with the national frameworksentinel · concierge: brief drafted from what the specialists foundsentinel · arbiter: verdict: needs attention; §3 of the brief says where→ Custody Policy Brief (draft) · awaiting approval

IV. What the AI does not do

The model never gets the last word.

Sources: Sentinel specification, the swarm and the verification engine. Verdicts are produced by the engine, never by a model.

Score your arrangement

Free, and it does not need an account.

Custody Agents S.L.