Multi-Agent Fabric
Any agent, any framework, one scoped token - GCTRL is the shared, governed memory layer your whole agent team reads and writes to.
What it does
GCTRL exposes its knowledge graph and memory tiers over an agent-facing surface that doesn't care which framework is asking. Claude Code, Codex, Cursor, GitHub Copilot, and custom orchestrators built on LangChain or similar stacks can all connect - over MCP, over direct HTTP, or by dropping in an agent skill that teaches the connecting agent when and how to read and write. Each connection authenticates with a single scoped access token: pick a clearance ceiling, grant the knowledge bases it may touch, and that token is the agent's entire view of the system.
Why it matters / USP
Today, every project that wants an AI agent to "remember things" ends up building its own bespoke RAG stack - its own vector store, its own chunking, its own access rules, none of it shared with the next project. GCTRL replaces that pattern with one governed memory layer that every agent framework in the organization connects to the same way. An agent's write-back after a task isn't a note that dies with the session - it becomes part of the graph the next agent, on the next task, in a different framework entirely, can read.
Because access is token-scoped the same way it is for humans, this doesn't mean one big undifferentiated pool: different agents and different teams can hold different tokens with different clearances and different granted graphs, so a contractor's automation and your internal agent team can share the same infrastructure without sharing each other's data. The result is a single team brain rather than a shelf of disposable, per-project memories - agent-framework-agnostic by design, and access-controlled by the same engine that governs human queries.
How it fits
The Multi-Agent Fabric is the consumption edge of the whole pipeline: everything KEX extracted, FUSE unified, and Manage KGs curates is what an agent reads through this surface, filtered by classification exactly as a human query would be.
In practice
A team runs Claude Code for engineering, a Codex-based support bot, and an internal LangChain orchestrator for ops - three different frameworks, three different scoped tokens, one shared graph. When the support bot resolves a customer issue and writes the resolution back, the engineering agent can retrieve that context weeks later without anyone having synced the two systems by hand.
See also
Classification & Access Control · Why a Memory Layer · Agents & MCP