IDEs & dev
Editors, CI, issue trackers, and code hosts.
CortexSemantic Model
The model behind high-yield inference
AI is metered in tokens. Humans care about outcomes.
Cortex links supported AI activity with the records, conversations, artifacts, and decisions around the work. The Semantic Engine reconstructs Work Units and Work Streams so teams can value, govern, and improve outcomes.
Connected sources
Connect the systems that explain the work.
AI activity is one input. A model call can be represented as an observed record on a supported capture path. Permission-scoped linked records can add the objective, ownership, history, collaboration, knowledge, and business context around that activity.
Chat & collab
Where conversations and decisions happen.
Model providers
Frontier, open, and local models alike.
Data stores
Warehouses, lakes, and document stores.
These logos are ecosystem examples, not a promise that every connector is available. Availability is deployment-specific: configuration, source permissions, and known coverage determine what is available. A connector is not automatically a capture or enforcement path.
The Semantic Engine keeps those sources distinct as it reconstructs coherent units of work.
The Semantic Engine
Signals become attributed work.
Cortex groups supported activity, links the surrounding source context, and proposes Work Units that retain their people, agents, applications, models, cost, evidence, and outcome.
Observed record
Linked record
Inferred relationship
Observed activity from supported tools and linked source records moves through the Semantic Engine and becomes attributed Work Units inside a Customer escalation responses Work Stream.
Work StreamCustomer escalation responses
A Work Unit captures one bounded attempt. Related attempts form a Work Stream.
The work model
The model follows the work.
A prompt is often too small to explain the objective. A session is often too broad to measure the outcome. Cortex reconstructs a bounded Work Unit, then groups related units into the Work Streams the organization already recognizes.
Semantic Model
The ontology for organizational work. It defines the entities, relationships, constraints, evidence, and outcomes Cortex uses to reason about activity as work.
Semantic Engine
The system that maps supported activity and linked records into the model, proposes Work Units, groups them into Work Streams, and updates relationships as evidence changes.
Work Unit
The smallest coherent attempt whose outcome a business can value, govern, and verify. A Work Unit can span prompts, sessions, applications, agents, tools, and human decisions.
Work Stream
An evolving family of related Work Units that share an objective, owner, operating pattern, policy, or baseline.
Work Stream topology
Keep the evidence attached to the work.
Each Work Unit associates the people, agents, models, applications, source records, cost, requirements, and outcomes available for that work. It labels a direct record separately from a proposed relationship, so a reviewer can distinguish evidence from inference.
Observed recordLinked recordInferred relationship
Work Units
Selected Work UnitsWU-A
Respond to an escalation
Resolved · Trace the available evidence and verify the response.
- People
- AB
- Models
- AB
- Cost
- In scope
WU-B
Review a billing exception
Verified · Connect the objective, approval, and outcome.
- People
- C
- Models
- AC
- Cost
- In scope
Related Work Units remain available.
Work Stream
Customer escalation responsesResolve urgent customer issues
Related Work UnitsEvidence available for review
- People
- ContributorsABC
- Models
- Activity in scopeABC
- Cost
- In scope
Linked records
Work recordEscalation objectiveApproved guidanceResponse runbookConversation recordCustomer context
Work Units
Selected Work UnitsWU-E
Update the support runbook
Approved · Turn accepted work into reusable guidance.
- People
- AD
- Models
- AB
- Cost
- In scope
WU-F
Review the escalation playbook
Verified · Check the guidance against the available outcome.
- People
- BC
- Models
- BC
- Cost
- In scope
Related Work Units remain available.
Work Stream
Support knowledge maintenanceKeep support guidance current
Related Work UnitsRelationships remain reviewable
- People
- ContributorsADB
- Models
- Activity in scopeABC
- Cost
- In scope
Linked records
Work recordGuidance objectiveDocument recordKnowledge recordApproved guidanceApproved runbook
When evidence is ambiguous, Cortex marks the relationship as inferred rather than presenting it as a direct record.
That model gives Atlas a stable unit for learning.
Product principles
Quiet by default. Explicit when it matters.
Cortex can gather evidence quietly on supported paths. It becomes explicit when an operator reviews a result or a configured path asks for a decision.
- 01Observe quietlyUse supported capture paths and linked records only within their stated coverage and permission scope.
- 02Respect the configured boundaryA decision applies only where a configured path gives Cortex a role. Architecture defines that control point.
- 03Explain the decisionKeep the policy, evidence, reason, confidence, and outcome available for review.
Atlas
Atlas maps work, learns its baseline, and ranks what matters.
Repeated Work Units reveal the expected pattern, meaningful variance, and the highest-value place to investigate next. Downstream outcomes show which work was accepted, revised, shipped, reverted, escalated, or resolved.
Selected deployment · Work Stream
Customer escalation responses
Evidence available to inspect the expected pattern and meaningful variance.
CO
Review owner
Configured operator
Source evidence
- Conversation recordLinked record
- Work recordLinked record
- Document recordLinked record
- Supported activityObserved record
Expected baseline
Evidence-linked response
Expected pattern
- Work records and approved guidance are linked before drafting.
- The owner verifies the available evidence before the response.
Observed variance
Reviewable- Context was rebuilt across available source records.
- The Work Unit contains evidence that can inform the next review.
Outcome feedback
Available for review- Positive: owner-approved, linked response accepted.
- Negative: reopened, edited, or missing-evidence response.
AI Opportunity Map
Prioritized from available evidence
- Reduce context reconstructionPrioritized
- Automate evidence verification
- Tune the escalation route
AI Opportunity Map
A quantified view of where time, cost, risk, and potential value concentrate across Work Streams.
Intelligence Roadmap
A prioritized set of interventions, success measures, and safety boundaries for the Work Streams that matter most.
Atlas ranks opportunities from the evidence available to the selected deployment. Operators can inspect the source records, labels, and inferred relationships behind the ranking.
Foundry receives the constraint, success measure, and safety boundary for each prioritized Work Stream.
Foundry
Foundry selects the intervention it can defend.
Atlas supplies the constraint, success measure, and safety boundary. Foundry chooses an intervention for a supported path and records why it made that choice.
Atlas context · Customer escalation responses
Keep the constraint, evidence, and policy with the action.
A decision receipt records the Work Unit, selected intervention, supporting evidence, applicable policy, expected outcome, approval state, and measured result.
Decision record
Constraints travel with the work- Work Unit
- Customer escalation response
- Selected intervention
- Semantic routing
- Supporting evidence
- Available source records
- Applicable policy
- Configured policy
- Expected outcome
- Outcome in scope
- Approval state
- Reviewable
- Measured result
- Recorded when available
Candidate routes
Configured path requirements- AEligible routeEligible model ASelected
- Policy
- Eligible
- Quality
- Within requirement
- Cost
- Within boundary
- Latency
- Within boundary
- BEligible routeEligible model B
- Policy
- Eligible
- Quality
- Within requirement
- Cost
- Within boundary
- Latency
- Within boundary
- CEligible routeEligible model C
- Policy
- Eligible
- Quality
- Within requirement
- Cost
- Within boundary
- Latency
- Within boundary
Route selected
Semantic routing
Eligible model A
Why this route
- ✓ Linked evidence makes the route policy-eligible.
- ✓ The route meets the configured requirements.
- ✓ The approval boundary remains reviewable.
Receipt status
Evidence available for review
The receipt retains the policy, evidence, approval state, and measured result when available.
Intervention catalog
Use when
Model cost and quality vary by task
Semantic routing
Choose among eligible models using the Work Unit requirements.
Use when
The request is missing approved context
Just-in-time enrichment
Add linked evidence at runtime on the configured path.
Use when
Repeated work has private verifier data
Model distillation
Optimize against accepted private outcomes and evaluations.
Use when
An agent exceeds appropriate autonomy
Agentic governance
Apply the configured policy and approval boundary before the supported action.
Use when
Sensitive context approaches an unsafe destination
Semantic security
Evaluate the destination using the Work Unit context at the configured control point.
Map your first Work Stream.
Bring one workflow with a clear owner and outcome. We will identify the available activity, source records, success measures, and policy boundaries needed to model it.