MCP-Native · Edge-Deployed · Zero Infrastructure
Every session starts from zero. Every lesson gets re-learned. Every client preference gets re-asked. RJ Knowledge Graph gives every agent you run one permanent, shared brain — plugged in via MCP in 60 seconds.
Free tier · No credit card · Key generated instantly on this page
The problem nobody budgets for
This is a normal Tuesday for a team running AI agents without persistent memory:
Third time this month. The client's patience — and the relationship your sales team spent 6 months building — takes another hit. The bot knew this yesterday. Today it doesn't.
45 minutes of tokens burned re-discovering the same root cause. The fix was documented — in a session that no longer exists. Multiply by every agent, every day.
Your agents can't share what they learn. Every bot is an island. The compound knowledge your business should be building? It evaporates at midnight.
of agent tokens re-establish context that was already known (typical multi-session workload)
wasted per re-solved problem your agents already solved before
for the DIY alternative: Neo4j + Pinecone + Redis + the DevOps time to babysit them
Amnesia isn't a quirk. It's a silent tax on every agent-hour you run.
The fix
Not another vector database. A real knowledge graph with multi-tier memory, built for how agents actually think:
Episodic, semantic, procedural & short-term lanes — with recency, importance and reinforcement scoring. Memories that matter float up; noise decays away.
PageRank centrality, community detection, shortest-path "how is X related to Y?" — answers vector DBs simply can't give you.
Each agent gets a private namespace and a shared lane. Sales bot learns it → support bot knows it. Compound intelligence, finally.
Cloudflare D1 + Workers AI embeddings, replicated to 330+ cities. No Docker, no clusters, no 3 AM pages. It's just… there.
How it works
Scroll down, type your name, get an API key instantly. No credit card, no sales call, no "book a demo" hostage negotiation.
Add our MCP server to Claude Desktop, Claude Code, Cursor or Windsurf. Your agent instantly grows 7 new tools: remember · recall · ask_graph and more.
Every conversation, doc and lesson flows into one graph. Tomorrow's session starts where today's ended. Next month, your agents know your business cold.
Drop into claude_desktop_config.json or .mcp.json:
{
"mcpServers": {
"rj-knowledge-graph": {
"command": "npx",
"args": ["mcp-remote",
"https://rj-knowledge-graph.pages.dev/mcp"]
}
}
}
Instantly available: remember · recall · ask_graph · search_graph · find_path · ingest_document · graph_overview
# Store a memory
curl -X POST .../api/memory/remember \
-H "x-api-key: YOUR_KEY" \
-d '{"content":"Client prefers email follow-ups",
"agent_id":"sales-bot",
"namespace":"agent:sales-bot"}'
# Recall (hybrid vector + recency + importance)
curl -X POST .../api/memory/recall \
-H "x-api-key: YOUR_KEY" \
-d '{"query":"client communication preferences"}'
Full reference at /docs · OpenAPI at /api/openapi.json
Don't take our word for it
The dashboard you'll get isn't a mockup — it's running live at /app with a real graph inside. Fly through the 3D knowledge graph, ask the Graph-RAG engine a question, watch PageRank recompute in real time.
Built for your stack
Building agents with Claude, Cursor, LangChain or raw APIs — and tired of context evaporating between sessions.
Running GoHighLevel, Make or n8n stacks for clients — and every client bot has goldfish memory.
Shipping AI features and staring down a $200+/mo infra bill plus a hiring req for someone to babysit Neo4j.
The honest comparison
Each tool below is great at its one job. The problem: agent memory needs all five jobs done at once.
| Capability | Pinecone / vector DBs |
Neo4j / graph DBs |
Mem0 / memory APIs |
RJ Knowledge Graph |
|---|---|---|---|---|
| Semantic recall (embeddings) | hybrid + recency + importance | |||
| Relationship reasoning (paths, PageRank) | + communities | |||
| Multi-tier agent memory | 4 tiers + namespaces | |||
| Native MCP server | partial | 7 tools built-in | ||
| Infrastructure you manage | managed, per-pod $ | cluster + ops | managed | zero — Cloudflare edge |
| Typical monthly cost to combine | $70 + $65 + $50 = $185+/mo across 3 vendors | $0–49/mo, one API | ||
Comparisons based on publicly listed starter/standard tiers as of 2026. Your mileage may vary — that's why the free tier exists.
Run your own numbers
Drag the sliders. The math updates live. (Assumes memory reuse eliminates ~70% of redundant context work.)
Estimated waste from re-established context
per month (23% redundant-context share)
Recoverable with persistent memory
per month vs. $49/mo Pro
Objections, handled
shared namespace crosses over). Everything runs on Cloudflare's SOC 2-audited edge network.Pricing
Start free. Upgrade when the memory pays for itself (see the calculator above — it usually does).
For kicking the tires properly
For agents doing real work
For fleets & white-label
The knowledge your business generated today could be compounding by tonight. One name. One click. One permanent brain.
No credit card. Generated on the spot — you'll be storing memories in under a minute.
Your API key is ready — copy it now, it's shown once:
Built by RJ Business Solutions — AI automation & growth systems. Edge-native on Cloudflare.