AI understands the domain.
Texomy makes it executable.
Semantic modelling is creative work — LLMs are great at it. Extraction is repetitive, must be auditable, and needs to be cheap. Texomy gives each job the right execution model: AI develops and verifies the specification; the runtime executes it deterministically.
Start a semantic specification with AI
Provide representative text and a short intent. Studio opens an initial project where AI proposes a model and specification that you can parse, inspect, and refine.
One specification connects AI and deterministic parsing.
Model the domain
AI identifies concepts, their fields, relationships, and the distinct event or document families in representative text.
Write the specification
Texomy Language makes that model explicit together with the rules that recognize its instances in text.
Verify before release
Run representative inputs through Texomy runtime. AI and domain experts inspect the results, then review and apply focused changes before release.
AI, where the specification lives
Start with AI
Paste representative text and state what you need to recognize. AI proposes an initial semantic model and specification in Studio, ready for parsing and review. It is a starting point, not an opaque result you must accept unchanged.
Refine with AI
Ask AI to improve the active specification or explain a parse result. It can revise the model, structure, and recognition rules; every proposal returns as a side-by-side diff that you can review, apply, or discard.
Usage is rate-limited during the public preview.
Use Texomy alongside LLMs when it makes sense.
A system can use Texomy runtime for known recurring structure and reserve LLM calls for genuinely free-form or unfamiliar material. Texomy does not prescribe that architecture; it provides the semantic specification and deterministic parser that such a system can use.
One possible evolution loop is:
- 1Parse known structure. Texomy extracts the concepts covered by the accepted specification.
- 2Route the remainder. The surrounding system decides whether unmatched or free-form text needs an LLM, a queue, or human review.
- 3Propose an improvement. AI can use representative unmatched cases to propose a semantic-model or specification change.
- 4Verify the result. Run representative inputs through the revised specification and inspect the semantic result.
- 5Review the contract. A domain expert reviews the model, specification diff, and relevant test results.
- 6Release deliberately. The accepted specification becomes the deterministic runtime behaviour.
Ready to work with Texomy AI?
Start from representative text in Studio, or inspect complete specifications first.