Skip to content

x-integration-guide.md

Latest commit

 

History

History
152 lines (105 loc) · 4.52 KB

x-integration-guide.md

File metadata and controls

152 lines (105 loc) · 4.52 KB

I gave Claude a brain. Here's how to make your AI actually remember you.

THE PROBLEM

Every time you open Claude Code, it starts blank. Your name, your preferences, your tech stack, the conversation you had yesterday — gone.

Claude's built-in memory? A 200-line markdown file that silently truncates your oldest memories. No ranking. No understanding. Just a notepad.

I built Cortex — a persistent memory engine that gives Claude (and any MCP-compatible AI) structured, self-evolving long-term memory. Bayesian beliefs. Cross-channel identity resolution. Sub-millisecond retrieval. 100% local.

Here's exactly how to set it up in 5 minutes.

SETUP: CLAUDE CODE (CLI)

  1. Install the binary:

git clone https://github.com/gambletan/cortex cd cortex cargo build --release -p cortex-mcp-server cp target/release/cortex-mcp-server ~/.local/bin/

  1. Register as MCP server:

claude mcp add cortex --scope user --
~/.local/bin/cortex-mcp-server ~/.cortex/memory.db

  1. Allow the tools (if using "don't ask" mode):

Add these to ~/.claude/settings.json → permissions.allow:

mcp__cortex__memory_ingest mcp__cortex__memory_search mcp__cortex__memory_context mcp__cortex__memory_consolidate mcp__cortex__belief_observe mcp__cortex__belief_list mcp__cortex__person_resolve mcp__cortex__fact_add mcp__cortex__preference_set

  1. Tell Claude to use it — add to CLAUDE.md:

Memory (Cortex)

You have persistent memory via Cortex MCP tools.

  • Start of conversation: call memory_context to load user context
  • When user shares a preference or fact: call memory_ingest
  • For structured facts: call fact_add
  • For preferences: call preference_set
  • Track beliefs with belief_observe
  1. Make it fully automatic — add a SessionStart hook:

Create ~/.claude/hooks/cortex-memory-inject.sh that calls memory_context via JSON-RPC and outputs the context. Claude Code injects the output into every new session automatically.

"hooks": { "SessionStart": [{ "matcher": "", "hooks": [{"type": "command", "command": "~/.claude/hooks/cortex-memory-inject.sh"}] }] }

Now Claude remembers you before you say a word.

SETUP: CLAUDE DESKTOP

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{ "mcpServers": { "cortex": { "command": "/.local/bin/cortex-mcp-server", "args": ["/.cortex/memory.db"] } } }

Restart Claude Desktop. 9 memory tools appear automatically.

WHAT CLAUDE GETS

9 tools that turn it from stateless to stateful:

memory_ingest — store what you learn about the user memory_search — semantic search across all memory memory_context — generate a context summary (token-budgeted) memory_consolidate — run decay + promotion + cleanup belief_observe — "user prefers Rust" → probability 0.92 belief_list — what do I believe about this user? fact_add — structured knowledge: User works_at Google preference_set — user preferences with confidence scores person_resolve — Alice on Telegram = Alice on Slack

WHAT HAPPENS IN PRACTICE

Session 1: You tell Claude you prefer Rust. → memory_ingest stores it → belief_observe("user_prefers_rust", supports=true, 0.8) → preference_set("language", "rust", 0.85)

Session 2: You open a new conversation. → SessionStart hook fires → memory_context returns:

[Cortex Memory Context]

User Profile

  • language = rust (confidence: 85%)
  • editor = neovim (confidence: 90%)

Beliefs

  • user_prefers_rust (confident, 92%)
  • user_is_developer (likely, 95%)

Claude already knows you. No "Hi, I'm Claude, how can I help you today?" It just starts helping — with context.

Session 5: You say "actually I'm switching to Go." → belief_observe("user_prefers_rust", supports=false, 0.7) → Probability drops from 0.92 to 0.68 → Not deleted — Cortex understands belief change, not binary flip

This is the difference between a notepad and a brain.

THE NUMBERS

Ingest: 7µs (vs 200ms Mem0 cloud) Search: 95µs (vs 300ms Mem0 cloud) Context gen: 58µs (vs 500ms Mem0 cloud) Binary: 27MB (with local embeddings) Privacy: 100% local, zero cloud Cost: $0

3,151x faster than Mem0 cloud. With features neither Mem0 nor file-based systems offer.

AI that knows you — not AI with a notepad.

github.com/gambletan/cortex MIT licensed. Zero cloud. Zero cost.

#ClaudeCode #ClaudeAI #MCP #Rust #OpenSource #AIMemory #DevTools #BuildInPublic #LocalFirst #RustLang