Humans own the what.
AI handles the how.
Fixed Entity Architecture eliminates structural hallucination.
Business concepts are human-locked anchors in a conscious model of your organization.
Every connection is mathematical.
Every AI output is grounded to the organizational graph.
The Engineering of Trust
Four layers. Risk increases upward.
Damage decreases. Each layer is grounded by the one below.
Concepts
Human-curated
AI proposes, humans approve
ZeroLinking
Mathematics
Cosine similarity + HyDE
ZeroEnrichment
Confidence-scored
AI enriches, humans review
MediumReasoning
Graph-grounded
Agents cite evidence
High but traceableThe Bridge
Bidirectional HyDE
“Customer Onboarding” and process_kyc_check() share zero common language.
Standard keyword or similarity search fails completely.
HyDE bridges the gap.
Mathematically.
Forward HyDE
Business concept → LLM generates 5 hypothetical technical documents → compared against real ingested chunks via cosine similarity.
Synthetic code vs. real code.
Reverse HyDE
Real code chunks → LLM generates hypothetical business descriptions → compared against actual concept definitions.
Code describes itself upward.
Both directions run. Results combined. Every link confidence-scored. Cost: ~$0.50 per 15 concepts (one-time).
Security
The MCP Boundary
Direct AI access to enterprise databases is an existential risk.
The MCP governance layer is a physical boundary separating autonomous agents from the database.
DGF Clearance
Per-user Red/Green clearance levels. Restricted data silently removed from agent results before reasoning.
Content Integrity
SHA-256 content-addressed IDs. Writes are idempotent and duplicate-free during CI/CD pushes.
Audit Logging
Every tool call generates structured JSON logs (Trace ID, User ID, Query Content) for regulators.
Rate Limiting
Per-user token bucket (default 60 RPM) prevents runaway agent swarms.
Multimodal Ingestion
Ingest anything.
Code, schemas, APIs, docs.
Coherence doesn't just read metadata. It parses everything your enterprise produces at AST depth, building a self-organizing knowledge fabric. 9+ programming languages. Function-level granularity. Every asset indexed, scored, and linked.
Code (Tree-Sitter AST)
Full abstract syntax tree parsing across Python, Java, TypeScript, Go, Rust, C#, Ruby, Kotlin, and more. Function-level, not file-level.
DB Schemas & APIs
PostgreSQL, MySQL, MongoDB, Snowflake, BigQuery, OpenAPI/Swagger, GraphQL, gRPC/Protobuf. auto-discovered and linked.
Infrastructure as Code
Terraform, Kubernetes, Docker Compose, CloudFormation, Azure ARM. your entire deployment topology ingested.
Project & Pipeline
Jira, Confluence, Linear, dbt, Airflow, Dagster. connect project context and data pipeline lineage to the graph.
Fast Mode
$0 / chunk
Local math only. No LLM calls. Instant.
Enriched Mode
~$0.008 / chunk
Full LLM pipeline. HyDE linking. Quality scoring.
Why This Is Different
Not a data catalog. Not a BI tool.
| Dimension | Data Catalogs | Coherence |
|---|---|---|
| Starting point | Tables & columns | Human-defined business concepts |
| Code depth | Metadata only | Full AST, 9+ languages |
| AI safety | None | FEA + MCP boundary |
| Data lineage | Table-level | Concept → API → Schema → Source Code |
| Community detection | None | AI-discovered concept clusters |
| TCO | $100K+/year | $5/month at 50K chunks |
Full 7-dimension comparison available on the Ontology page.
This site uses cookies
We use essential cookies for the site to function and analytics cookies (Google Analytics) to understand how you use it. Analytics cookies are only activated with your consent. We do not track you across other websites. Your data is stored in the EU and processed in accordance with GDPR. Read our Privacy Policy
