by Encevra

Bring Noetis into your product.

Send a question and documents. Receive structured findings, source references, and a decision brief to present in your own product.

Schedule an engine setup call Read the API docs

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

Completed sample response

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

Frontier intelligence, structured for deeper analysis.

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 engine

Illustrative workflow, not a live analysis

Documents

TermsCompares obligations

NumbersChecks figures

RiskSurfaces exposure

Evidence check

Sources matched

Decision brief

Ready to review

Why the structure matters

How the architecture supports deeper analysis

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.

Focused analytical roles

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.

Evidence review

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.

Checked synthesis

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.

Structured output

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

Direct model integration or Noetis.

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

From source passage to decision brief.

Fictional renewal example. Hand-authored findings and review outcomes, not a measured Noetis run.

Characters 364–423 (end exclusive)

ASTER SYSTEMS | SUBSCRIPTION RENEWAL Initial term: 24 months. The subscription renews for successive 12-month periods unless either party gives 60 days' written notice. Annual subscription fee: $48,000. The annual uplift is capped at 5%. Appendix C: the uplift cap does not apply in year two. Supplier liability is limited to fees paid in the preceding 12 months. This limitation does not apply to confidentiality breaches. The renewal proposal states: the annual uplift cap applies in every renewal year. No service credit schedule is included in the submitted renewal package.

Infrastructure

Built to run inside your product.

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.

Asynchronous by default

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.

Evidence and support are separate

A located quotation and the analytical support verdict are reported independently, so a real quote is not treated as proof of every interpretation.

Account-bound access

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

36 configurable engines on one runtime

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.

Implemented: 36

Browse all 36 engines Open Engine Studio

Questions buyers ask

Evaluating fit.

What does Noetis accept and return?: A question and source material as plain text. A run returns specialist findings, quoted passages with their location in the submitted text, support verdicts, unsupported-claim flags, open questions, and a synthesized brief as JSON. Some runs end with unresolved questions; that is reported rather than hidden.
How does it fit into an existing product?: Your backend submits a run with an API key, stores the run identifier, and polls the run endpoint for the result. You decide how findings and passages are presented to your users.
How are findings checked?: Each quoted passage is located in the submitted text, and the support for the finding is assessed separately. Finding a quotation shows the text exists; it does not establish that the conclusion drawn from it is correct.
How can my team evaluate it?: Start with a conversation about your use case. We arrange evaluation access per customer and can review results on your own sample material. The benchmark page documents our internal evaluation method and its limits.
What does Encevra add beyond access to a model?: The engine around it. A run splits the work across specialists with one responsibility each, locates every quoted passage in your text, has a separate reviewer check support on a different model, and returns structured JSON with a release status. With a raw model API you would build and maintain those steps yourself.
Which models do your engines use?: Engines run on licensed third party foundation models that act as independent model witnesses. Encevra builds the roles, evidence checks and release rules around them, and each run records which model version answered each role.
Does every engine use the same models?: No. Catalog and Studio engines use one model family for specialists and a separate one for review and the final brief, so review is independent of reading. Full Noetis and Noetis Lite use their own role assignments.
Can I choose which models my engine uses?: You choose the engine, for example full Noetis, Noetis Lite with "engine": "lite", a catalog template or your own Studio engine. You cannot pick individual models per request today. Specific model arrangements can be discussed under a written agreement.
Why use an Encevra engine instead of building directly on a model API?: To skip building specialist prompts, source location, claim review, release rules, spend limits, webhooks and billing yourself. Whether it beats your own build on your tasks is something to test: no engine has a scored evaluation yet, so we recommend a scoped evaluation on your own material.
Are Gmail, Drive, or Slack available through the API?: Not today. Those connectors work inside Encevra's own applications. The engine API accepts text you send it.

Licensing

Quoted terms for committed volume.

Build

$0.95 per run$500 per month minimum

Asynchronous API runs at the agreed rate
Polling and signed completion webhooks
Shared rate limits

Scale

$0.75 per run$4,000 per month minimum

Everything in Build
Rate limits set in the agreement
Customer facing use where the agreement includes it

OEM

By agreement

Terms, volume and support set in writing
Private hosting discussed as a requirement only. Not technically available today.

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.

Build deeper analysis into your product.

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