Connect the sources
Select repositories, tickets, documents and other records. Configure source identities, access scopes and history.
Find the decisions behind your code.
Muninark connects code with the records that explain it. Use those sources when investigating behavior, reviewing a change or working with an AI assistant.
From record to answer
How records become usable knowledge.
Start by importing the relevant history, then keep it updated as the sources change. When someone asks a question, retrieval selects evidence they are permitted to access.
Build the knowledge
Extract entities, business rules and decisions, and link decisions to the changes that implement them. Rules and decisions become searchable evidence, with source references and validity in time.
Review and maintain
Experts approve, correct or reject extracted rules and decisions. Operators track processing, gaps and items that need reevaluation.
Serve the evidence
Use MCP to retrieve evidence and supported analyses. Responses carry citations, limitations and confidence information.
Anatomy of an answer
Check what supports the answer.
Open the cited source to check a statement. The response also tells you what is missing, whether it needs review and how far an advisory analysis reaches.
- Evidence · source type, record identifier, date, a supporting summary and, where the text is indexed, a short verbatim excerpt of the record.
- Gaps · missing records and limits on what can be concluded.
- Review state · whether new information calls for reevaluation, and whether an expert has approved a cited rule or decision.
- Advisory scope · coverage notes for analyses that inform a human decision.
ONE PRODUCT · CONNECTED RECORDS
DECISIONKeep the rounding exception
REPOSITORY ACalculation service
REPOSITORY BReporting service
RELEASE RECORDA different rule from v6.0
Illustrative relationships on synthetic data.
History and uncertainty
Check the date, scope and evidence.
An answer about an older release needs different evidence from an answer about the current system. Missing history and uncertain matches should be visible.
Time and versions
Facts can carry validity intervals and changes remain part of the record. Release-specific answers also need a reliable mapping between releases, components and deployment data. The engine can reconstruct the recorded state of selected record types on a given date; whether that is enabled depends on the deployment's history and configuration. Supported temporal queries are established during integration.
Confidence
Confidence is derived from the evidence: how well it matches the question, the type of source and whether sources agree. The weighting is set during evaluation. The band helps you assess the evidence; it does not tell you the probability that an answer is correct.
Expert review
The system distinguishes extracted knowledge from knowledge an expert has reviewed. For the business rules and decisions a deployment designates, a query can be limited to expert-approved records, and a record an expert rejects is withdrawn from answers.
How knowledge stays currentFits your working environment
Use the engine on its own or with workflows.
Use the engine through MCP. Add selected skills for repeatable tasks, then supervised processes where coordination helps.
- Engine and MCP
Knowledge, evidence, policies, retrieval and operations. - Optional skills
Procedures with defined inputs, checks and outcomes. - Optional processes
Roles and steps that compose the selected skills.
Which part of your product takes too long to explain?
Tell us about a recurring question and the sources your team uses. We can help define a pilot around it.