Semantic Search Under 25ms
Hybrid search combining SQLite FTS5 keyword matching with vector similarity. Find relevant memories from 160K+ vectors in under 25 milliseconds.
v3.0 — 160,000+ vectors indexed live
Your AI forgets everything.
Your AI forgets everything.
Memory Spine fixes that.
Persistent vector memory for AI agents and developers. Store context, search semantically in under 25ms, and hand off knowledge between agents — all through 32 MCP tools on an open protocol.
Persistent Memory for AI Agents
Everything your AI needs to remember context across sessions, conversations, and agent handoffs.
Knowledge Graph Construction
Automatically build relationship graphs between memories. Query connections, find clusters, and traverse knowledge paths for deeper AI understanding.
Memory Pinning & Priority
Pin critical context that must never be forgotten. Priority-ranked memories ensure your AI always has the most important context available first.
Agent Handoff Protocol
Transfer context seamlessly between AI agents. When one agent finishes, the next picks up exactly where it left off with full history.
Timeline & Conversation Tracking
Query memories by time range, track conversation turns, and replay interaction history. Built-in temporal intelligence for context-aware responses.
32 MCP Protocol Tools
Store, search, recall, pin, tag, consolidate, batch, cluster, timeline, handoff, and more. Every memory operation your AI agent needs, one protocol call away.
How to Give Your AI Persistent Memory
Three steps. No infrastructure to manage. Works with any MCP-compatible model.
1
Connect via MCP
Add Memory Spine as an MCP server in your AI agent configuration. Compatible with Claude, GPT, Gemini, and any MCP-enabled framework.
2
Store & Tag Memories
Your AI stores context as tagged memory vectors. Each memory gets semantic embeddings for similarity search and FTS5 indexing for keyword matching.
3
Search & Recall
When your AI needs context, it searches by semantic similarity, keyword, tag, or time range. Results return in under 25ms with relevance scoring.
Real MCP Tool Calls
Memory Spine uses the open Model Context Protocol — the same standard used by Claude, Cursor, and Windsurf. No proprietary SDK. No vendor lock-in.
Your AI agent calls MCP tools directly. Here are real examples from production:
memory_store— persist any context with tagsmemory_search— semantic + keyword hybrid searchmemory_pin— pin critical context by keymemory_timeline— query memories by time rangeknowledge_graph_build— construct relationship graphsagent_handoff— transfer context between agents
MCP Tool Calls
// Store a memory with tags memory_store({ "content": "User prefers dark theme and TypeScript", "tags": ["preferences", "user-context"] }) // → {id: "mem-2847", vector_count: 160628} // Search memories semantically memory_search({ "query": "what does the user prefer", "limit": 5 }) // → Results in ~22ms, ranked by relevance // Pin critical context by key memory_pin({ "key": "system-config", "content": "Production API: port 8788" }) // Hand off context to another agent agent_handoff({ "target_agent": "code-reviewer", "include_recent": 20 })
Built for Every AI Memory Use Case
From chatbots to autonomous agents, any AI that needs to remember.
Autonomous AI Agents
Agents that work across sessions need persistent memory to avoid repeating work, remember decisions, and build on prior knowledge.
Chatbots & Assistants
Give your chatbot long-term memory of user preferences, past conversations, and resolved issues. No more “I don’t have context for that.”
Code Copilots
Remember codebase patterns, user coding style, project architecture, and debugging history. Your copilot gets smarter with every session.
RAG-Augmented Systems
Combine retrieval-augmented generation with persistent memory. Your RAG system remembers what worked, what didn’t, and adapts over time.
Multi-Agent Orchestration
When multiple agents collaborate, Memory Spine provides shared context. Agent handoff protocol ensures seamless knowledge transfer between specialists.
Research & Analysis
Long-running research tasks need to accumulate findings. Memory Spine stores, tags, and clusters results for progressive synthesis.
AI Memory Pricing — Included with ChaozCode
Pick the vector limit you need. All plans include full MCP tool access and semantic search.
Indie
$19/mo
- 25K memory vectors
- All 32 MCP tools
- Knowledge graphs
- Analytics dashboard
- Priority support
Developer
$49/mo
- 100K memory vectors
- Batch operations
- Memory consolidation
- 5 parallel agents
- Advanced clustering
Master
$99/mo
- 250K memory vectors
- Everything in Dev
- Custom integrations
- SLA guarantee
- Dedicated support
14-day money-back guarantee on all paid plans. Compare all ChaozCode plans →
Running in Production, Not a Proof of Concept
Memory Spine runs 24/7 as the persistent context layer of the ChaozCode platform — powering 276 specialized AI agents, task routing, and multi-agent orchestration in production.
160,627
Vectors Stored
v3.0
Production Version
276
Connected Agents
99.9%
Uptime
Frequently Asked Questions
Everything developers ask about persistent AI memory and Memory Spine.
What is Memory Spine?
Memory Spine is a persistent vector memory system for AI agents and developers. It lets AI agents store, search, and recall context across sessions using 32 MCP (Model Context Protocol) tools with sub-25ms semantic search latency. It runs as part of the ChaozCode platform.
How does Memory Spine give AI agents persistent memory?
Memory Spine stores memories as vectors with full-text search indexing (FTS5). When an AI agent needs context, it searches by semantic similarity or keyword match and retrieves relevant memories in under 25ms. Memories persist across sessions, conversations, and agent handoffs — your AI never forgets.
What is the Model Context Protocol (MCP)?
MCP is an open protocol that lets AI models interact with external tools and data sources. Created by Anthropic, it is supported by Claude, Cursor, Windsurf, and a growing ecosystem. Memory Spine implements 32 MCP tools for memory operations including store, search, recall, pin, tag, consolidate, timeline, and agent handoff.
Is Memory Spine free to use?
Yes. The free tier includes 5,000 memory vectors and core MCP tools at no cost, no credit card required. Paid plans start at $19/month for 25K vectors with the full 32-tool suite including knowledge graphs, analytics, and priority support.
How fast is Memory Spine search?
Sub-25ms on 160,000+ vectors. Memory Spine uses a hybrid approach: SQLite FTS5 for keyword matching combined with vector similarity search. Both full-text and semantic queries return results in under 25 milliseconds in production.
Can Memory Spine work with any AI model?
Yes. Memory Spine uses the open MCP protocol, making it compatible with any AI model that supports MCP tools — including Claude, GPT, Gemini, LLaMA, and open-source models. It also works with agent frameworks like LangChain, AutoGPT, CrewAI, and custom agent systems.
How is Memory Spine different from Pinecone or ChromaDB?
Memory Spine is purpose-built for AI agent memory, not general vector storage. It includes agent-specific features: memory pinning, knowledge graph construction, conversation tracking, agent handoff context, memory consolidation, and timeline queries. It also combines FTS5 keyword search with vector similarity in a single system — no separate infrastructure needed.