Introduction
GCTRL - Ground Your AI. Command Your Data. - is on-prem knowledge-graph and memory middleware for AI and autonomous agents. It sits on top of your existing storage and turns raw, unstructured content into a queryable knowledge graph plus a set of parallel memory layers that agents can reason over.
What GCTRL is
GCTRL is a middleware layer, not another database. It runs above the storage you already operate:
| Layer | Default engine | Role | Swappable |
|---|---|---|---|
| Graph | Neo4j | Knowledge graph storage (entities, relations, compilations) | Yes |
| Vectors | Qdrant | Semantic retrieval / embeddings index | Yes |
| Relational | Postgres | Platform state, accounts, licensing, job metadata | Yes |
| Cache / queue | Redis | Coordination, job queues, ephemeral state | Yes |
All four are bundled by default so a fresh install works out of the box, and every one is swappable - point GCTRL at your own managed Neo4j, Qdrant, or Postgres when you move to production.
The ingestion + memory layer for AI
GCTRL does two things that most "AI memory" products do not:
- Ingests at scale. It pulls in documents, text, and other unstructured sources, extracts entities and relations, and resolves them into a coherent knowledge graph. This is the KEX (extraction) and FUSE (fusion / entity resolution) pipeline.
- Organises memory into parallel layers. It exposes more than one kind of memory at the same time:
- Raw deterministic storage - exact, reproducible context an agent can store and recall verbatim. No paraphrasing, no drift.
- A curated Wiki-LLM - an auto-distilled, human-readable knowledge surface over your company's data, grounded in the graph.
These layers run in parallel, so an agent can pull deterministic facts for grounding and distilled company knowledge for context in the same workflow.
How this differs from "agent memory" tools
Most agent-memory tools let an agent store and organise its own memory - notes, summaries, conversation history. That is useful, but it starts after the data is already small and structured.
GCTRL starts earlier: it ingests at scale and builds the knowledge graph that the memory sits on top of. In short:
- Agent-memory tools: organise what the agent already saw.
- GCTRL: ingests the corpus, builds the graph, and then serves memory over it.
FUSE's entity-resolution engine is what lets large, messy ingests collapse into one clean graph instead of a pile of duplicates.
On-prem and GDPR-ready by design
- Fully on-prem. Every component runs on hardware you control.
- Local inference by default. Embeddings and generation run on local Ollama - zero data leaves the machine, zero token cost.
- GDPR-ready. Because nothing is shipped to a third party unless you explicitly connect a cloud provider, GCTRL fits regulated and data-sovereign environments.
Next steps
Continue to Installation to bring up the stack with a single command.