Ommisa is the assistant of the OIML SMART program.
Ommisa answers questions about the publications of the International Organization of Legal Metrology (OIML) and grounds every claim in the publication text, so that each answer carries the citation chain to the clause it relies on. When the corpus does not carry an answer, the service states that plainly rather than inventing one. On the OIML SMART certification platform, the same engine assists the applicant, the issuing authority and the test laboratory in their daily work, where it prepares drafts that the operator reviews and confirms.
What the assistant answers
The public assistant reads the OIML publications corpus, which comprises approximately 900 documents and 9.3 million words in ten languages. A question receives an answer written from that corpus, with every claim accompanied by its clause-level citation, so that the reader can verify the answer against the publication itself in one step. Language models can err, which is why the citation accompanies every claim, and why the publications themselves remain the authoritative text.
The public service is described, including its machinery and its limits, on its own How it works page. Anonymous use allows 20 questions a day per address; signing in with an OIML SMART account raises the tier to 300 questions a day, a stronger model and conversation history that follows the account. Toolmakers call the same service through the documented API, which streams answers as server-sent events or returns them as complete JSON.
Where the assistant runs
ai.oimlsmart.org
The public assistant is available to every visitor, and the member tier adds the stronger model and the conversation history.
The SMART platform
On the platform, Ommisa assists the operator inside the certification workflow itself: it reads the page the operator is on, when the operator allows it, and it prepares drafts such as an application prefill that the operator confirms in the real form. The platform records the act only on that confirmation, and the audit chain marks the act as AI-assisted.
The public service and the platform's assistant are two deployments of one engine, so that an incident on the public service never touches the deployment that prepares drafts against live certification records. The integration that lets the assistant ride the platform's own sign-in session, so that the operator never meets a second login, is rolling out in September 2026; until it completes, the member tier signs in at the assistant service itself.
The discipline the assistant keeps
Ommisa operates under four standing rules, in a manner that:
- grounds every claim in a cited clause of the corpus, and states plainly when the corpus has no answer;
- proposes only the acts that the platform's state machine currently offers to the operator's role, because the platform's model, and not the model's confidence, defines what is possible;
- prepares every record-changing act as a draft that the operator reviews and confirms, and marks the confirmed act as AI-assisted on the audit chain;
- holds no write credential of its own, so that every write executes under the operator's own session and permissions.
What the assistant does not do
Ommisa does not decide conformity, does not issue certificates and does not replace the evaluation of a test laboratory or the decision of an issuing authority. Its answers are reading aids over the corpus, and its drafts are preparations that take effect only through the platform's own validation after the operator's confirmation.
Links
The public assistant is at ai.oimlsmart.org, the running platform is demonstrated at demo.oimlsmart.org, and the program that operates both is presented at www.oimlsmart.org.