Specialized agents & business workflows
Evidence turns your business processes into workflows of specialized agents, designed around your rules, your flows, and your enterprise's knowledge. The agents take on the operational work; your experts keep the approvals and the decisions that matter.
Ask a question, analyze documents, or pick up a case: each piece of work keeps its context and its status.
Where it fits in the architecture
AI no longer just answers your experts. It works with them — under control.
The problem
Reviewing a case file, checking compliance, preparing a recommendation, classifying an incident, assessing a supplier: this work draws on several areas of expertise, internal rules, and approval points. A generic assistant knows neither your process, nor your rules, nor your responsibilities.
Use
Specialization does not come from a library of agents: it comes from the way your organization actually works, in five stages.
Evidence maps the steps, responsibilities, rules, and information the work requires.
Each agent is given a precise role based on the real flow: analyze, verify, research, challenge, prepare, or check.
Agents work with the knowledge, procedures, reference frameworks, and authorized data relevant to their mission.
Evidence coordinates the agents according to the workflow steps and the rules defined by the organization.
Human approvals and arbitration are placed where they truly create value or where accountability requires them.
Business outcome
How it works
Six stages, adapted to your process. Agents work in parallel when it helps; the expert steps in where they add the most value.
Step 01 / 06
Steps, responsibilities, rules, expected result
Specialization
Evidence starts from your business flows to define each agent's role: what it must understand, the information it can draw on, the actions it can prepare, the rules it must follow, and the situations it must hand over to a human.
Interface
Real screenshots from the demo environment: the home screen, the summary, the analysis, the risks, and the audit of a case.
Ask a question, analyze documents, or pick up a case: each piece of work keeps its context and its status.
Operational architecture
For technical readers: the specialized agents run the workflow, draw on Atlas for context and business understanding, and call on PRISM critical control only when the criticality of the processing requires it.
A case or request enters Evidence, where specialized agents run the workflow and draw on Atlas for context, knowledge, and business reasoning. The processing can then be assessed for criticality. The standard path continues the workflow directly; when a critical control is required, PRISM is brought in before processing returns to Evidence. Human approvals take place at the steps defined by the process, then Evidence delivers the output along with the elements intended for its review.
Flow view: ordinary flows progress directly; only processing classified as critical brings in the PRISM engine.
The agents coordinate the investigation and production steps according to the role defined by the process.
Atlas provides the understanding of enterprise context and the reasoning engine that inform each of the agents' actions.
The workflow can assess a processing task to determine whether reinforced critical control is required.
The workflow continues without systematic multi-model cross-checking. Routine searches, document analyses, and summaries progress directly in the workspace.
Brought in when the stakes call for multi-model cross-checking and a strict policy before delivery. Processing that commits the organization is thus subject to independent verification.
At the steps defined by the business
Checkpoints and approval points placed according to the workflow's requirements (before commitment, before release, or at a sensitive milestone). The decision stays within the remit of the authorized owners.
Summary, report, or expert memo that business teams can use directly.
Outputs designed to be sourced can return the references and passages used, for review.
Trace according to the workflow: agent log and approvals. An integrity artifact may be issued for critical outputs subject to the applicable policy.
Engineering & proof
Technical claims about Evidence rest on documented measurements and artifacts, presented with their methodological limits.
14
14 active profiles under agents/: compliance, contradictor, default, developer, finance, hacker, infrastructure, legal_drafting_guarded, legal_safe, marketing, medical, multitask, researcher, sales. The _example folder is excluded.
5 required checks
Production release chain for the Evidence backend: protected main branch, mandatory PR, 5 required checks, enforce_admins, build/deploy guardrails, drain before restart, /healthz, and VERSION.json.build_commit.
Offers
The journey always starts with a real business process. All offers are on quote after scoping; details are on the Deploy KOREV page.
On quote
Prove out Evidence on a real business process.
On quote
Scale one or more business workflows.
On quote
Deploy in a sovereign or regulated setting.
Architecture
Evidence is the operational layer: this is where your teams work with the specialized agents. The agents rely on Atlas to interpret enterprise context and knowledge. When processing reaches the level of criticality defined by the policy, PRISM takes over to cross-check several independent models and suspend processing when the criteria set by the critical policy are not met. Evidence delivers the result with its sources, its checks, and, for critical outputs that go through the relevant finalization pipeline, an integrity artifact.
Evidence organizes the workflow and interacts with Atlas for context and reasoning. After the criticality assessment, processing either follows the standard path or branches to PRISM when critical control is required. The result then returns to Evidence.
The workflow can classify processing by its sensitivity and the applicable policy in order to adjust the level of control.
Searches, information summaries, and routine analyses run directly, without systematically triggering multi-model control.
Result · Sources · Required approvals
Deliverables are returned to the operational workspace with their justifications and the configured human review points.
Operational Intelligence Interface
The workspace where teams use KOREV agents to handle case files, follow the workflow, and deliver the outputs.
Capabilities
A unified environment to define, run, and monitor workflows of specialized agents: analysis, research, checking, comparison, synthesis, recommendation.
Each agent's role, access, responsibilities, and limits are defined according to your flows, your rules, and your knowledge. A framework of 14 documented profiles serves as an engineering foundation, not a fixed catalog.
Internal procedures, reference frameworks, and corpora are drawn on for the agents' understanding, without blind raw injection into the weights.
The workflow assesses the criticality of a processing task to determine whether reinforced control is required before delivery.
Inconsistencies and missing information are explicitly flagged before any conclusion.
Approval points are placed at the steps where the business process requires them; the expert stays in charge of the decision.
Outputs designed to be sourced retain the passages and reference frameworks used, allowing rigorous review.
For critical outputs that go through the relevant finalization pipeline, an integrity artifact can be issued according to the applicable policy.
Connection to the sources and tools defined at deployment: DMS, ERP, CRM, internal databases, APIs.
Governance
Evidence retains the elements the organization needs to account for every processing task: sources, agents involved, checks, and approvals.
Regulatory compliance depends on the context, the system, its use, and the organization's role. KOREV provides technical governance capabilities; on its own, it does not constitute a legal certification of compliance.
Deployment
Evidence runs where the organization decides. The agents rely on Atlas to understand the business context, within the chosen isolation perimeter.
Hosting under European jurisdiction, fast setup, strict data isolation.
Data and internal knowledge stay with you; orchestration and models are distributed according to the sensitivity of the processing.
Platform, agents, Atlas reasoning engine, and data in your infrastructure. No mandatory external dependency.
Use cases
Describe it in a few lines. We assess its sensitivity, its approvals, and its deployment level, then propose an Evidence Pilot scope.
KOREV — AI that shows its work.