by EncevraSend a question and documents. Receive structured findings, source references, and a decision brief to present in your own product.
Live API access requires account approval. Customer facing deployments require a licensing agreement. Review the benchmark and methodology.
Interactive API example
Sample data · no analysis is run
Decision brief
Clarify Aster's term before choosing on quoted cost alone.
Supported contradiction
Aster's proposal offers 12 months while its master terms state a 24-month minimum.
Unresolved question
Which term length governs the renewal?
Matched source excerpt
Aster_Terms.pdf · page 3
“Minimum subscription term: 24 months.”
The quotation was located in the submitted text. That supports inspection of the finding, not a guarantee that every conclusion is correct.
The architecture
Noetis brings frontier models into a coordinated reasoning architecture. Specialized analytical roles examine the material, and findings pass through evidence checks before the final synthesis. It is built for complex questions where the answer needs to stand up to review.
How we build and tune the engineDocuments
Evidence check
Sources matched
Decision brief
Ready to review
Why the structure matters
The models supply the reasoning ability. Encevra builds the orchestration and verification around them, designed to improve complex analysis and make every finding easier to inspect.
Seven roles read the same material in parallel: terms, costs, risk, contradictions, gaps, questions to ask, and cause-and-effect claims. Each has one job, so a figure that does not add up and a termination clause that was misread are looked for separately rather than competing for attention in one answer.
Addresses details that get missed when one pass has to cover everything.
Every quoted passage is first located in the submitted text. A separate review stage then judges whether that passage supports the whole finding, marking it supported, narrowed, or unsupported, and checking scope, conditions, and exceptions.
Addresses conclusions that cite real text but claim more than it says.
The brief is written from reviewed findings. It is then checked against them, with one bounded repair pass. If evidence is insufficient or claims stay unsupported, the run is returned as incomplete with its reasons, and open questions are reported rather than hidden.
Addresses confident summaries that drift beyond the evidence.
A run returns JSON: findings by role, quoted passages with character offsets, support verdicts, open questions, the brief, and its release status.
Lets your product show each finding beside the passage behind it.
What you are integrating
A frontier model can reason deeply and cite sources, and teams can build their own review workflows. Noetis gives you that workflow already built, tested, and running.
| Aspect | Direct model integration | Noetis engine integration |
|---|---|---|
| What you integrate | A model's response to your prompt. | An implemented analysis workflow running frontier models. |
| Analytical stages | Whatever your prompt and code define. Strong models reason extensively on their own. | Seven parallel roles, a separate evidence review, synthesis, and a final claim check. |
| Evidence handling | Citations when prompted; verification is yours to build. | Quotes located in the source and support judged separately for each finding. |
| Output | Text or the schema you design. | Findings, offsets, verdicts, open questions, brief, and release status. |
| Tradeoff | Faster and cheaper per request. | More calls per run, so more time and computation per answer. |
Our historical internal evaluation compared Noetis with a single frontier model on 30 documents, including the cases where the single model did better. The best test is your own material: we arrange evaluation on the questions your product needs to answer.
Inspect the output
Fictional renewal example. Hand-authored findings and review outcomes, not a measured Noetis run.
Characters 364–423 (end exclusive)
Infrastructure
Reliable delivery is what makes a multi-stage analysis usable in production. It carries the reasoning process; it is not what makes the answers better.
A submission returns a run identifier immediately. Poll the run endpoint until it finishes. Source text is accepted as plain text, up to 150,000 characters per request.
A located quotation and the analytical support verdict are reported independently, so a real quote is not treated as proof of every interpretation.
API keys and runs are isolated to the owning account. Keys are hashed at rest and repeated credential failures are blocked.
Noetis engines and Engine Studio
24 business and 12 personal templates, each a configured set of specialist responsibilities, tools and rules on the shared Noetis runtime. Start from a template or describe your own engine in Engine Studio. None has a published scored evaluation yet. Live API use requires an approved account, and customer facing use requires a licensing agreement.
Questions buyers ask
Licensing
$0.95 per run$500 per month minimum
$0.75 per run$4,000 per month minimum
By agreement
These are quoted terms agreed in writing, not self serve subscriptions. Prepaid self serve usage is described on the pricing page; rights are listed on the licensing page. Bring your own model keys is not supported.
What counts as a runOne complete specialist panel, evidence review, and checked brief that is released. Incomplete, failed and cancelled runs are not charged.
Monthly commitmentCommitted volume draws down at the agreed rate.
Give your users a structured analysis process with findings they can inspect. Explore how Noetis fits your workflow and evaluate it on the questions your product needs to answer.
Explore the architecture