Neural Fabric
Neural Fabric is the governed knowledge graph for multi-agent business knowledge. Connections are derived mathematically from your data. Not mapped by engineers, not guessed by models. Eight MCP tools. Any agent. Minutes, not months.
Who connects to what
Any MCP-compatible AI connects to one Neural Fabric server. All agents read and write to the same knowledge graph: Claude, ChatGPT, Gemini, or any open-source model.
Every supernode represents thousands of entities that share the same type. The edges between them are aggregate harmonic couplings derived from the entities inside.
Search for an entity and land directly on its connections. No query language. No filters. Structure you navigate, not results you search through.
Every write is governed. Every edge explains itself.
The Neural Fabric doesn't accept data blindly. Every write is validated against the live graph before it's persisted. Writes that would compromise graph stability are rejected. The system gets quieter, not wronger. Your agent can't corrupt the graph even if the LLM hallucinates. Built for industries where 'the AI said so' isn't compliance.
Every accepted write carries a cryptographic provenance chain: actor, timestamp, operation, and graph state at write time. Each entry is hash-linked to the one before it. Tamper detection runs on every read. Built for regulated industries where 'trust me' isn't compliance.
Click any connection and see exactly which properties caused it. Shared identifier. Matching reference. The forensics panel decomposes every connection to the feature level. We haven't found another system that explains why two entities are connected, and proves it hasn't been altered since.
What no one else has built.
Most graph systems differ at query time. The Neural Fabric differs at construction time: relationships are derived, governed, explained, and stable before any agent asks a question.
Alternatives depend on hand-built ontology or LLM assertion. Neural Fabric relationships are deterministic: same data, same graph, every run, without external embedding API calls.
Every write is checked before it lands. Unsafe changes are refused instead of silently corrupting the graph.
Connections decompose to the feature level, so agents can see why two entities are connected. Accepted entities carry hash-linked provenance, and the graph stays model agnostic when you switch LLM providers.
Two questions usually come next: how meaning is handled, and what happens when source data conflicts.
No LLM = no semantic understanding?
The LLM handles extraction and classification. Frequency-native coupling handles graph structure. That split is why graph construction does not require embedding-token spend.
What happens when documents conflict or go stale?
Both facts are stored with provenance. The graph traces each one to its source and lets the querying agent see the conflict without corrupting graph structure.
What happens if you change models?
Switch from Claude to ChatGPT tomorrow. Your knowledge, connections, and provenance stay because the model does not build the graph.
Eight tools. Every agent.
Connect any MCP client and these eight tools appear automatically. Same tools in Claude, Cursor, ChatGPT, OpenClaw, and any MCP-compatible client.
fabric_addAdd structured entities directly to your graph. Your agent builds the entity, the Fabric governs the write.fabric_editMake partial, targeted modifications to existing entities; the graph re-derives connections and governs every write.fabric_searchSearch by name, type, or natural language. Band-code narrowing reduces thousands of entities to a ranked shortlist.fabric_listList and filter entities across your graph. Get counts, types, and metadata without loading the full graph.fabric_graphExplore entity neighborhoods and resonance subgraphs. See how entities connect through shared frequency signatures.fabric_read_sourceRead source document content from stored files. Pull the original text behind any entity without leaving your agent.fabric_deleteRemove entities cleanly. Tombstone semantics remove derived connections automatically.fabric_provenanceInspect the cryptographic audit trail: actor, timestamp, hash-linked entries, and tamper detection.Where this deploys.
The entity types change by domain. The underlying problem does not: records reference each other across systems, but the relationships are not preserved for agents.
Governed knowledge graph for multi-agent business knowledge.
Eight tools. Every agent. Start in minutes.