GrayMatter v4.0 Released

Your AI's Long-Term Memory,
Finally Upgraded.

Multi-agent fleet memory with GraphRAG. From slow, fragmented context to instant, unified knowledge. Built for the agents that never sleep.

Watch It Think

Real-time knowledge graph visualization. Every node is a fact. Every connection is a relationship.

Entity
Relation
Community

The Numbers Don't Lie

Real benchmarks from a 4-node agent fleet.

Search Latency

18.5s 7.0s

2.6x faster with dual-port reads

Entity F1 Score

94%

Validated against gold-standard test set

Agent Nodes

4

Patrick, Discus, Hermit, Dodo

Memories Stored

19.8K

19,799 facts across all agents

Three Retrieval Modes

How embedding-only, GraphRAG, and VectorRAG compare in practice.

1

Embedding-Only

Plain vector similarity. The baseline.

Latency 15ms
Precision ~60%
Multi-hop No
Cost/query $0

Best for

Fast fragment retrieval — finding a single sentence that matches.

Coming Soon
2

GraphRAG

Entity + relation extraction. Multi-hop reasoning.

Latency 2.7s
Multi-hop Yes
Type filtering Yes
Maintenance High

Best for

Multi-hop reasoning — "GrayMatter → uses → bge-m3 → runs_on → Ollama"

3

VectorRAG

Cluster summaries over embeddings. Topic-level results.

Latency 0.7s
Win rate 83%
Topics 150 clusters
Cost/query $0

Best for

Topic overviews — "what do I know about X" rather than "find memory Y"

Side-by-Side Comparison

Dimension Embedding GraphRAG VectorRAG
Latency 15ms 2.7s 0.7s
Precision ~60% ~17% 83% win rate
Multi-hop No Yes No
Topic overviews No Partial Yes
Maintenance None Monthly Monthly
Per-query cost $0 $0 (slow) $0
Fleet usage ~80% 0% ~15%

Before vs After

What changed when we rebuilt the engine.

Before GrayMatter v4

  • Search queries took 18+ seconds
  • 2,157 orphaned entities with no connections
  • $0.05 per memory extraction via LLM
  • Single-port bottleneck on every read
  • No entity resolution — duplicates everywhere

After GrayMatter v4

  • Search queries in 7 seconds (2.6x faster)
  • All entities connected via knowledge graph
  • Zero LLM cost — regex NER extraction
  • Dual-port reads (8768 + 8769)
  • Automatic entity resolution & deduplication

Built for Scale, Designed for Speed

What makes GrayMatter different.

GraphRAG

From orphaned entities to connected knowledge. Automatic community detection groups related facts for faster retrieval.

Fleet-Wide Memory

Four agent nodes, one shared brain. Patrick, Discus, Hermit, and Dodo all read and write to the same knowledge graph.

Dual-Port Reads

Port 8768 (read+write) and 8769 (read-only replica). Parallel reads are 2.6x faster for heavy workloads.

Eval Harness

Built-in benchmarking. Test extraction quality against gold standards before deploying to production.

Zero Dependencies

Pure Python stdlib + NumPy. No heavy graph libraries. Community detection runs on pure BFS.

Tailscale Fleet

Secure mesh networking. Agents connect via Tailscale IPs — no public exposure, no VPN config.

Ready to deploy?

Join the fleet. Deploy GrayMatter v4 today.