From evidence to a decision your reviewer can inspect.
Your AI client brings the model. Ansvar supplies authoritative sources, customer evidence for Team and Company, structured workflow state, and citation checks. The run produces a reviewable assessment with each factual finding cited or left unresolved.
bring your own model · EU-hosted gateway · structured workflows for people and standing agents
One evidence chain from scope to decision.
The workflow keeps the sources, customer inputs, analysis state, review points, and resulting findings in one reviewable run. Completed runs are records; a new run can cite a prior report as evidence.
- 1Define the decision and scopeworkflow · system boundary · roles · jurisdictions
- 2Collect the evidenceofficial sources · licensed standards · selected customer documents
- 3Execute and reviewrequired stages · labelled judgments · human review gates
- 4Produce the assessment recordcited findings · unresolved gaps · deliverable · Company receipt
Legal, technical, and customer evidence in one run.
A workflow can draw from several evidence classes while preserving the attribution and access rules of each source. The selected inputs stay visible to the reviewer.
Authoritative public sources
Primary and national law, case law, preparatory works, and agency guidance arrive with publisher, licence, and provision anchors. Source tools report coverage and freshness separately.
SIS-licensed clause evidence
Entitled users can bring selected licensed clauses into an assessment with the standard designation, clause reference, and required attribution.
Technical evidence
NIST, MITRE, CWE, CAPEC, CVE, KEV, and EPSS records can support a threat model, TARA, or vulnerability decision.
Customer evidence
Policies, architecture documents, prior reports, and other registered files stay within the customer tenant and resolve to paragraph-level citations.
The method, evidence, and review state travel with the result.
Ansvar enforces the workflow around the customer's model. This keeps the model choice separate from the controls that define a complete run.
The workflow starts with a declared job
The run records its method, system boundary, roles, jurisdictions, inputs, and required output before analysis begins.
Required stages are server-enforced
The agent works through the registered stages and required review points. A client cannot skip a required state and present the result as complete.
Human conclusions stay labelled
Factual support, analytical judgment, approval, and open questions remain distinct in the result so a reviewer can inspect each one.
Missing support remains visible
A source gap, missing national layer, failed lookup, or unsupported conclusion becomes an unresolved item for review.
A missing source stays missing.
No-match results, service failures, malformed responses, and hollow responses have separate states. An empty or failed lookup cannot support a conclusion.
illustrative response · the final row needs more evidence or a reviewer decision
A reviewer can resolve the evidence behind each factual finding.
Citation resolution is a non-model check. It confirms identifiers, anchors, attribution fields, and served text. Source status tools report coverage and freshness for the run; those values are separate from the citation envelope. A new run can retrieve another evidence set when the question, scope, source availability, or customer inputs change. Team and Company can register private documents in Cite your documents.
| Field | What the reviewer can check |
|---|---|
| source_url | Canonical source link or customer-document reference for the evidence behind the finding. |
| publisher | The public publisher or the owner of the registered customer document. |
| license | The source licence or entitlement code. An item without required attribution is refused. |
| article / segment | The provision, clause, paragraph, or document segment that supports the factual statement. |
citation fields carried with served evidence
A finished assessment with its review record.
The outcome is a structured deliverable that shows what the evidence supports, which judgments need approval, and which questions remain open.
Findings, risks, and actions
The workflow returns the assessment fields required by its method, with owners and actions where the method calls for them.
Evidence, judgments, and gaps
The reviewer sees cited support, labelled judgments, source-health context for the run, customer evidence, and unresolved items behind the result.
Signed receipt
Company can add a signed receipt and cryptographic audit record for the completed run, subject to its deployment trust model.
Bring the decision due next.
Connect your AI for source research, or open the sample to inspect a complete assessment structure.