A metadata-first document knowledge system for the agentic era. Every document is a node in a knowledge graph. Every node and edge is governed by parviGov.
Documents are the lifeblood of every organization. And yet, the way organizations manage documents is fundamentally broken.
A file's path tells you where someone put it, not what it is, not what it relates to, not what depends on it.
Search helps you find a document but does not tell you how documents connect. Relationships are buried in prose.
Existing metadata systems were built for humans clicking through desktop interfaces, not for agent fleets that need to address, traverse, and act programmatically.
Every node and edge is governed by parviGov. Content-addressed, hash-based identity. Agents address documents by stable identity, traverse typed links, and act under policy with provable actions.
Every node content-addressed. Every edge typed and witnessed. Traversable by agents.
M-Files built a Finnish company into a global enterprise by recognizing that what a document is matters more than where it sits. Graph Files takes that lesson and applies it to the agentic era.
Every document has a class, required properties, lookups, and relationships. Class drives structure. Structure drives findability.
Documents identified by hash, not by path. Content-addressed means identity is intrinsic. Change the content, change the identity. Humans see friendly names. Agents work with verifiable identifiers.
Every interaction — ingestion, extraction, decision, publication — produces an AWP-signed receipt. The knowledge graph is a governed record of who touched what, when, and why.
Designed as a plane that agent platforms use as a first-class client. Not a desktop app with an API bolted on. Agent-native from the start.
Agents traverse the knowledge graph intelligently. An agent can ask: show me everything that supports this claim. Show me the derivation chain. Show me what supersedes what. The graph answers.
This document was derived from that source.
This document provides evidence for that assertion.
This document replaces that earlier version.
This document is part of that case or project.
Every document has a content identity based on its hash. Not a path. Not a URL. A cryptographic fingerprint of the content itself.
You cannot separate what the documents say from proof that agents handled them correctly. Every node, every edge, every interaction produces cryptographic evidence.
When a document enters the graph, the act is witnessed. Content hash recorded. Provenance sealed.
When information is extracted, the extraction is witnessed. What was read, by whom, when.
When a decision is made based on documents, the decision and its basis are recorded. Traceable forever.
When a document is published or shared, the publication is governed. Who shared, what, to whom.
A piece of evidence is collected, logged, transferred, analyzed, presented, and stored. Every step must be provable. In most systems today, this chain relies on human discipline and editable logs.
In most companies, this trail is scattered across email threads, Slack messages, signed PDFs, and shared drives. Reconstructing who approved what is a forensic nightmare. Graph Files makes it automatic.
Each step is a governed node. The approval is a SUPPORTS_CLAIM typed link. The approver's identity, timestamp, and basis are sealed in an AWP-signed receipt. When a regulator asks for proof, the entire decision lifecycle is a traversable, verifiable graph — generated automatically, in real time.
A document evidence plane for multi-agent systems. If you build agent platforms, you need a place where documents live, connect, and are governed.
Chain of evidence, tamper-proof documentation, provable custody. Every step cryptographic. Every transfer witnessed.
Who approved what, when, on what basis. Provable. Generated automatically in real time. Not reconstructable after the fact.
Companies where AI labor handles document-heavy processes. These organizations need documents that agents can trust, prove, and act on.
This is stated honestly. Graph Files is a designed and approved thesis with a functional MVP, moving toward production. It is not a finished product. But the direction is locked — and when it ships, it will be the only agent-native, governed, content-addressed document knowledge system in the market.
Document management systems do not have cryptographic governance. Document AI tools do not have content-addressed identity. Knowledge graph platforms do not have agent-native APIs with cryptographic witness.
Graph Files. Document context for agent fleets, with proof in every node.
If you build agent platforms or work in justice, compliance, or regulated document workflows — let's talk.