🧠 Try Cortex in your browser — zero install, 124KB WASM, runs entirely client-side.
If Cortex helps your AI remember, give it a ⭐ — it takes 1 second and helps others discover the project.
Memory for AI agents that never leaves your device.
Private. Free. Local. — a memory engine for personal AI agents.
Your AI's memory lives on your device — your data never leaves, never costs, never spies. Pure Rust. 3.8MB binary. No third-party servers in the data path, zero telemetry, zero cost. Syncs through your own cloud storage. (On-device semantic search downloads a ~30MB model once on first use, then runs fully offline — or go 100% offline with CORTEX_NO_EMBEDDINGS=1. See Security & Privacy.)
Philosophy: Your memories are yours — not a cloud provider's training data, not a startup's monetization asset, not a government's surveillance target. Cortex runs 100% on your hardware, stores everything in your own database, and syncs only through your own cloud storage (iCloud, Google Drive, OneDrive, Dropbox). No middleman ever sees your data. No API key required. No account to create. Just plug it into your AI agent and it remembers — privately, permanently, and at sub-millisecond speed.
LLMs start blank every session. Your assistant forgets your name, your preferences, the conversation you had yesterday, the decision you made last week. Current "memory" solutions are flat text files, keyword grep, or cloud APIs that add 200-500ms latency, charge you for the privilege, and send your personal data to someone else's server.
Cortex fixes this. It gives your AI a structured, queryable, self-evolving long-term memory that persists across sessions, channels, and contexts — with Bayesian beliefs that self-correct, a people graph that resolves identities across platforms, and sub-millisecond performance on everything. All running locally, all yours.
Cortex vs Mem0 vs OpenAI Memory
| Cortex | Mem0 | OpenAI Memory | |
|---|---|---|---|
| Privacy | 100% local, zero cloud | Cloud API (your data on their servers) | OpenAI servers |
| Latency | 156µs ingest, 568µs search | ~200-500ms | ~300-800ms |
| Cost | Free, forever | $99+/mo (Pro) | ChatGPT Plus ($20/mo) |
| Memory tiers | 4 (Working/Episodic/Semantic/Procedural) | 1 (flat) | 1 (flat) |
| Bayesian beliefs | Self-correcting with evidence | No | No |
| People graph | Cross-channel identity resolution | Paid tier only | No |
| Conversation compression | Automatic session summarization | No | No |
| Relationship inference | Pattern-based (EN + CN) | No | No |
| Temporal retrieval | Intent-aware ("recently" / "first time") | No | No |
| Contradiction detection | Automatic with confidence scores | No | No |
| Consolidation | Episodic → Semantic auto-promotion | No | No |
| Context injection | Token-budgeted LLM-ready output | Manual | Automatic but opaque |
| Import/Export | Full JSON backup & restore | API only | No export |
| Self-hosted | Native binary, Docker, MCP | Cloud only | Cloud only |
| Binary size | 3.8 MB | npm package | N/A |
| Dependencies | 0 runtime services (single binary) | Node.js + cloud | N/A |
| Open source | MIT | Partial | No |
| Encryption | AES-256-GCM encrypted sync (opt-in) | No | No |
| Key rotation | Versioned envelopes, forward secrecy | No | No |
| Privacy levels | Private (default, never syncs) / Shared / Public — per-memory opt-in, demote retracts from other devices | No | No |
| Tool authorization | Deny-by-default capability policy on the MCP surface | No | No |
| Zero telemetry | No analytics, no phone-home, verifiable | Unknown | No |
| Cost | Free forever, unlimited | $99+/mo (Pro) | $20/mo (Plus) |
| Chinese NLP | Native (inference, retrieval, relationships) | No | Limited |
| Namespace isolation | Per-user/context memory separation | No | No |
| Plugin system | Compile-time hooks for ingest/retrieve/consolidation | No | No |
| MCP tools | 30 tools for Claude/LLM integration | 3rd party | N/A |
Performance Benchmarks
| Operation | Cortex | Mem0 (cloud) | File-based |
|---|---|---|---|
| Ingest | 156µs | ~200ms | ~1ms |
| Search (top-10) | 568µs | ~300ms | ~10ms |
| Context generation | 621µs | ~500ms | manual |
| Belief update | 66µs | N/A | N/A |
| People graph | 51µs | paid tier | N/A |
| Structured facts | 45µs | N/A | N/A |
| 1K memories search | 1.6ms | ~500ms | ~50ms |
528x faster than Mem0 cloud. With features neither Mem0 nor OpenAI Memory offer.
Note: Benchmarks include proactive inference (auto-extracting facts, preferences, relationships) on every ingest. Raw ingest without inference is ~15µs. Numbers from
cargo benchon M-series Mac.
LoCoMo Benchmark (ACL 2024)
Academic-grade long-term conversation memory evaluation — 10 conversations, 1540 QA pairs across 4 categories.
| System | Single-hop | Multi-hop | Open-domain | Temporal | Overall |
|---|---|---|---|---|---|
| Backboard | 89.4% | 75.0% | 91.2% | 91.9% | 90.0% |
| MemMachine v0.2 | — | — | — | — | 84.9% |
| Cortex v1.7 | 72.5% | 59.5% | 88.8% | 74.1% | 73.7% |
| Mem0-Graph | 65.7% | 47.2% | 75.7% | 58.1% | 68.4% |
| Mem0 | 67.1% | 51.2% | 72.9% | 55.5% | 66.9% |
| OpenAI Memory | — | — | — | — | 52.9% |
Key findings:
- Open-domain 88.8% — leads Mem0 (72.9%) by +15.9%
- Temporal 74.1% — leads Mem0 (55.5%) by +18.6%
- Single-hop 72.5% — leads Mem0 (67.1%) by +5.4%
- Multi-hop 59.5% — leads Mem0 (51.2%) by +8.3%
- Overall 73.7% — beats Mem0 (66.9%) by +6.8%, beats OpenAI Memory (52.9%) by +20.8%
Cortex outperforms Mem0 on all 4 categories — while running 100% locally, end-to-end encrypted, at $0 cost.
Setup: Claude Sonnet 4 (QA + judge), nomic-embed-text (embeddings via Ollama), top-30 retrieval. Fully reproducible:
python3 bench/locomo_bench.py
Architecture
Cortex implements a 4-tier memory model inspired by human cognition:
+---------------------+
| Working Memory | Current session context
+---------------------+
|
+---------------------+
| Episodic Memory | Raw experiences: conversations, events, observations
+---------------------+
| consolidation (decay, promotion, pattern extraction)
+---------------------+
| Semantic Memory | Distilled facts, preferences, relationships
+---------------------+
|
+---------------------+
| Procedural Memory | Learned routines, user-specific workflows
+---------------------+
Working holds the current session scratch pad. Episodic stores raw experiences with timestamps and source metadata. The Consolidation Engine periodically promotes recurring patterns into Semantic facts and decays stale episodes. Procedural captures learned workflows and routines.
Key Components
People Graph
Cross-channel identity resolution. The same person messaging you on Telegram, emailing you, and showing up in calendar events gets unified into a single identity node. Interactions, relationship strength, and communication patterns are tracked per-person.
Bayesian Belief System
Self-correcting understanding of the world. Beliefs are formed from evidence, updated with each new observation, and can be contradicted. Confidence scores reflect actual certainty rather than recency bias.
cortex.observe_belief("user_prefers_morning_meetings", true, 0.8)?; cortex.observe_belief("user_prefers_morning_meetings", false, 0.6)?; // Confidence adjusts automatically via Bayesian update
Consolidation Engine
Episodic-to-semantic promotion, decay of stale memories, and pattern extraction. Runs as a background cycle that keeps the memory store lean and queryable. Returns a report of what was promoted, decayed, and merged.
Multi-signal Retrieval
Queries combine five signals for relevance ranking:
- Similarity -- vector cosine distance against query embedding
- Temporal -- recency weighting with configurable decay
- Salience -- importance scoring from access patterns and explicit hints
- Social -- boost for memories involving specific people
- Channel -- filter or boost by source channel
Context Injection Protocol
Generates LLM-ready context strings from memory state. Pass a token budget, optional channel/person filters, and get back a structured text block your LLM can consume directly.
Storage
SQLite for persistence, in-memory vector index for fast similarity search. Single-file database, no external services required. Designed for edge deployment -- runs on a laptop, a Raspberry Pi, or a server.
Cloud Sync
Sync memories across devices through your own cloud storage — no third-party server involved.
Device A (Mac) Your Cloud Storage Device B (iPhone)
┌──────────┐ ┌──────────────────────┐ ┌──────────┐
│ SQLite DB │ ──W──> │ iCloud / GDrive / │ <──R── │ SQLite DB│
│ (local) │ │ OneDrive / Dropbox │ │ (local) │
│ │ <──R── │ │ ──W──> │ │
└──────────┘ └──────────────────────┘ └──────────┘
- Changelog-based: Each device writes append-only operation logs to its own subfolder
- No conflicts: Devices never write to the same file. Merge uses Last-Writer-Wins with Hybrid Logical Clocks
- Encrypted: AES-256-GCM encryption (opt-in). Even if your cloud account is compromised, memories stay private
- Tamper-evident: the sync manifest and every operation carry an HMAC; tampered or plaintext-injected oplog lines are rejected, and a manifest without integrity protection refuses to load (no key-rollback path)
- Key rotation & forward secrecy: rotate to a new key version (
ENC2envelopes) without re-encrypting history; old versions stay readable, new writes are unreadable to a leaked old key - Privacy-aware, per-memory opt-in: Private memories (the default) never leave your device. Mark a memory
sharedto sync it; demote it back toprivateand a retraction deletes it from your other devices (local copy kept) - Survives restarts: sync settings persist in the database (passphrase never touches disk — macOS login Keychain or
CORTEX_SYNC_PASSPHRASE); the server resumes sync and starts background pull (30s poll + fs watcher) automatically
Supported providers: iCloud Drive, Google Drive, OneDrive, Dropbox (auto-detected).
use cortex_core::sync::SyncConfig; use cortex_core::types::PrivacyLevel; // Enable sync with encryption (settings persist; passphrase goes to the OS keychain) let config = SyncConfig::new(sync_dir, device_id, device_name) .with_encryption("my-strong-passphrase"); cortex.enable_sync(config)?; // Opt a memory into sync — everything is Private unless you say otherwise cortex.set_memory_privacy(mem_id, PrivacyLevel::Shared { scope: "all".into() })?; // Pull changes from other devices (also happens automatically in the background) let applied = cortex.sync_pull()?; println!("Applied {} remote changes", applied);
Security & Privacy
| Feature | Detail |
|---|---|
| Encryption | AES-256-GCM with Argon2id key derivation (per-line random nonce) |
| Key rotation | Versioned ENC2 envelopes with per-version passphrase-derived keys — forward secrecy against AES-key exfiltration, no full re-encryption needed |
| Integrity | HMAC on the sync manifest and on every sync operation; plaintext lines in an encrypted oplog are rejected outright (injection defense) |
| Privacy levels | Private (default, never syncs), Shared, Public — set at ingest (privacy arg / --privacy) or later (memory_set_privacy); demoting to Private retracts the memory from other devices |
| Capability policy | Deny-by-default tool authorization on the MCP surface: a capabilities.json grants tool groups (read/write/sync/plugins) or exact tools; ungranted tools are invisible and uncallable; malformed policy fails closed |
| Query budget | Every retrieval is bounded (candidate cap + wall-clock cap) — query cost never scales with total store size; DoS guard and timing-side-channel bound in one |
| Secret handling | Sync passphrase is never written to disk by Cortex — macOS login Keychain or env var only; missing passphrase fails safe (sync off, never plaintext) |
| Memory zeroization | Sensitive data cleared from RAM on drop (zeroize crate) |
| Zero telemetry | No analytics, no phone-home, no user data ever leaves the device — enforced in CI (scripts/check-no-network-egress.sh): the build fails if any network/telemetry crate enters cortex-core's default tree, and the check also proves the --no-default-features binary is completely zero-network. |
| Embedding model fetch (one-time) | The default cortex-mcp-server enables on-device semantic search, which downloads a ~30 MB model (all-MiniLM-L6-v2) from the Hugging Face CDN on first ingest, then runs fully offline and sends none of your data. For a 100%-offline setup: run with CORTEX_NO_EMBEDDINGS=1 (keyword/FTS recall, zero network) or build --no-default-features. A one-time stderr notice is printed before any download — nothing is ever fetched silently. |
| No accounts | No API key, no registration, no cloud dependency |
See SECURITY.md for the full threat model.
Prerequisites
Install the Rust toolchain (provides cargo):
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
After installation, either restart your terminal or run:
source "$HOME/.cargo/env"
Verify:
Real-World Example: A Personal AI That Actually Remembers
Imagine your AI assistant across a week of real conversations:
# Day 1 — You chat on Telegram
You: "Sarah works at Stripe. She's interested in our API."
Cortex auto-extracts:
├── episodic memory stored (156µs)
├── fact: Sarah → works_at → Stripe (confidence: 0.70)
└── person resolved: sarah_telegram
# Day 2 — Sarah emails you
From: sarah@stripe.com
"Here's the technical spec we discussed."
Cortex:
├── person resolved: sarah@stripe.com → merged with sarah_telegram
│ (same person, different channel — automatic identity resolution)
└── fact: Sarah → sent → technical spec
# Day 3 — You ask your AI
You: "What's the status with Stripe?"
Cortex retrieves (568µs):
├── Sarah works at Stripe (semantic fact)
├── Meeting went well, interested in API (episodic, Day 1)
├── She sent technical spec (episodic, Day 2)
└── Cross-channel context: Telegram + Email unified under one person
Your AI responds with full context — no "sorry, I don't remember" 🎯
# Day 5 — New information arrives
You: "Sarah now works at Anthropic."
Cortex:
├── contradiction detected: Sarah works_at Stripe vs Sarah works_at Anthropic
├── old fact superseded + decayed: Stripe (salience ×0.3, kept as history)
├── new fact stored: Sarah → works_at → Anthropic
└── current employer now ranks first; self-correcting, no manual cleanup
(Third-party relations are extracted from natural-language verbs —
"works at / works for / joined / now works at", "runs on", "hosted in",
"manages", "part of", … — between two proper-noun entities.)
# Day 7 — Consolidation runs
Cortex auto-consolidation:
├── 3 episodic memories about Sarah → promoted to semantic summary
├── stale memories from other topics → decayed
└── pattern detected: you have recurring Monday meetings
All of this happens locally in <1ms per operation. No cloud. No API calls. No one else sees your data.
Install
Homebrew (macOS / Linux)
brew tap gambletan/tap brew install cortex-mcp-server
From source
cargo build --release -p cortex-mcp-server
cp target/release/cortex-mcp-server ~/.local/bin/Quick Start
use cortex_core::Cortex; // Open (or create) a memory database let cortex = Cortex::open("memory.db")?; // Ingest a memory from a Telegram conversation let embedding = your_embedding_fn("Met with Alice about the Q3 roadmap"); cortex.ingest( "Met with Alice about the Q3 roadmap", "telegram", // source channel Some("alice_123"), // user ID (triggers identity resolution) Some(0.8), // salience hint Some(embedding), // vector embedding )?; // Add a semantic fact directly cortex.add_fact( "Alice", "works_at", "Acme Corp", 0.95, "telegram", None, )?; // Store a preference cortex.add_preference("timezone", "America/Los_Angeles", 0.9)?; // Retrieve relevant memories let results = cortex.retrieve( "What do I know about Alice?", 5, // top-k None, // any channel None, // any person Some(query_embedding), // vector for similarity search )?; // Generate LLM-ready context (token-budgeted) let context = cortex.get_context( 2000, // max tokens Some("telegram"), // channel filter None, // no person filter )?; // Pass `context` as system/user message prefix to your LLM // Run consolidation (call periodically) let report = cortex.run_consolidation()?; println!("Promoted: {}, Decayed: {}", report.promoted, report.decayed);
Python Bindings
Coming soon via PyO3. The cortex-python crate will expose the full API as a native Python module:
from cortex import Cortex cx = Cortex.open("memory.db") cx.ingest("Had lunch with Bob at the Thai place", channel="imessage", user_id="bob") results = cx.retrieve("Where does Bob like to eat?", limit=5)
Integration with unified-channel-hub
Cortex is designed as the memory layer for unified-channel-hub. Messages flow in from any channel adapter, Cortex ingests and indexes them, and the context injection protocol feeds relevant memory back to your LLM before each response.
Telegram ─┐ ┌─ Context
Discord ─┤ unified-channel-hub → │ Cortex → LLM
Email ─┤ (ingest) │ (retrieve + inject)
Calendar ─┘ └─ Response
Integration with LangGraph
Add persistent memory to any LangGraph agent via langchain-mcp-adapters — no custom code needed.
from langchain_mcp_adapters.client import MultiServerMCPClient from langgraph.prebuilt import create_react_agent from langchain_openai import ChatOpenAI model = ChatOpenAI(model="gpt-4o") async with MultiServerMCPClient({ "cortex": { "command": "cortex-mcp-server", "args": ["~/.cortex/memory.db"] } }) as client: agent = create_react_agent(model, client.get_tools()) # Agent now has all 30 Cortex memory tools result = await agent.ainvoke({ "messages": [{"role": "user", "content": "What do you remember about Alice?"}] })
Your LangGraph agent gets instant access to memory_search, memory_ingest, fact_add, belief_observe, person_resolve, and 25 more tools — all running locally.
Integration with DeerFlow (ByteDance)
Cortex works as a persistent memory layer for DeerFlow — ByteDance's open-source multi-agent orchestration platform. Zero code changes needed.
# Add to DeerFlow config.yaml mcp_servers: cortex-memory: command: cortex-mcp-server args: - ~/.cortex/deerflow.db
All DeerFlow agents (Telegram, Slack, Feishu) get instant access to 30 memory tools — cross-session memory, fact storage, people graph, and belief tracking across all channels.
CLI
Cortex doubles as a standalone CLI tool — no MCP client required.
$ cortex-mcp-server --help
Cortex memory engine — MCP server & CLI tools
Usage: cortex-mcp-server [DB_PATH] [COMMAND]
Commands:
ingest Store a new memory
search Search memories
stats Show memory statistics
sync Show cloud sync status and detected providers
export Export all data as JSON
import Import data from JSON file
info Show version, DB path, and capabilities
help Print this message or the help of the given subcommand(s)
Arguments:
[DB_PATH] Path to the Cortex database file (default: ~/.cortex/memory.db)
Options:
-h, --help Print help
-V, --version Print version
Examples:
# Store a memory cortex-mcp-server ~/.cortex/memory.db ingest "Met with Alice about Q3 roadmap" cortex-mcp-server ~/.cortex/memory.db ingest -c telegram "Sarah now works at Anthropic" # Search cortex-mcp-server ~/.cortex/memory.db search "Alice" cortex-mcp-server ~/.cortex/memory.db search -l 10 "Q3 roadmap" # Stats cortex-mcp-server ~/.cortex/memory.db stats # Cloud sync cortex-mcp-server ~/.cortex/memory.db sync # status cortex-mcp-server ~/.cortex/memory.db sync enable # auto-detect provider cortex-mcp-server ~/.cortex/memory.db sync enable -p icloud # specific provider cortex-mcp-server ~/.cortex/memory.db sync pull # pull remote changes # Export / Import (backup & restore) cortex-mcp-server ~/.cortex/memory.db export -o backup.json cortex-mcp-server ~/.cortex/new.db import backup.json # Version & capabilities cortex-mcp-server ~/.cortex/memory.db info
No subcommand = MCP stdio mode (for Claude Code / Claude Desktop integration).
MCP Server (Claude Code / Claude Desktop)
Cortex ships as an MCP server — works with any MCP-compatible client.
Setup
1. Build & install the binary:
mkdir -p ~/.local/bin ~/.cortex cargo build --release -p cortex-mcp-server cp target/release/cortex-mcp-server ~/.local/bin/
2. Register as MCP server:
Claude Code (CLI):
# Global (all projects) claude mcp add cortex --scope user -- ~/.local/bin/cortex-mcp-server ~/.cortex/memory.db # Or per-project claude mcp add cortex -- ~/.local/bin/cortex-mcp-server ~/.cortex/memory.db
Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"cortex": {
"command": "/Users/you/.local/bin/cortex-mcp-server",
"args": ["/Users/you/.cortex/memory.db"]
}
}
}3. Allow tools in "don't ask" mode:
Add to ~/.claude/settings.json → permissions.allow:
Note: MCP tool permissions do not support parentheses format (e.g.
mcp__cortex__memory_ingest(*)). Use the wildcardmcp__cortex__*instead.
4. Make it automatic — add to your CLAUDE.md (project or global ~/.claude/CLAUDE.md):
# Memory (Cortex) You have persistent memory via Cortex MCP tools. Use them automatically: - Start of conversation: call `memory_context` to load what you know about the user - When the user shares a preference, fact, or personal info: call `memory_ingest` to store it - When you learn a structured fact: call `fact_add` (e.g. "User works_at Google") - When you detect a preference: call `preference_set` (e.g. editor=neovim) - When evidence supports or contradicts a belief: call `belief_observe` - When talking to someone new: call `person_resolve` to track identity - Periodically: call `memory_consolidate` to clean up stale memories
5. Auto-inject memory on session start (Claude Code hooks — fully automatic):
Create ~/.claude/hooks/cortex-memory-inject.sh:
#!/bin/bash CORTEX_BIN="${CORTEX_BIN:-$HOME/.local/bin/cortex-mcp-server}" CORTEX_DB="${CORTEX_DB:-$HOME/.cortex/memory.db}" [ -x "$CORTEX_BIN" ] || exit 0 printf '%s\n%s\n%s\n' \ '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"hook","version":"1.0"}}}' \ '{"jsonrpc":"2.0","method":"notifications/initialized"}' \ '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"memory_context","arguments":{"max_tokens":1500}}}' \ | "$CORTEX_BIN" "$CORTEX_DB" 2>/dev/null \ | grep '"id":2' \ | python3 -c "import sys,json; r=json.load(sys.stdin); print(r['result']['content'][0]['text'])" 2>/dev/null
Add to ~/.claude/settings.json:
{
"hooks": {
"SessionStart": [
{
"matcher": "",
"hooks": [
{
"type": "command",
"command": "~/.claude/hooks/cortex-memory-inject.sh"
}
]
}
]
}
}Now every new Claude Code session automatically loads your memory context — zero manual effort. Claude learns as you work and remembers across sessions.
Cross-Device Memory Sync
Your Claude's memory follows you across all your devices — MacBook, iMac, work laptop — through your own cloud storage.
Enable sync (one command):
You: "Enable cross-device memory sync"
Claude calls sync_enable → auto-detects iCloud Drive →
generates device ID + AES-256-GCM encryption key → done.
Output:
Provider: iCloud Drive
Encryption: AES-256-GCM
Passphrase: a1b2c3... ← save this for your other devices
On your second device — one script does everything (build/install, wait for iCloud, join, restore):
git clone https://github.com/gambletan/cortex && cortex/scripts/setup-device-sync.sh # Prompts for your passphrase (hidden input; or set CORTEX_SYNC_PASSPHRASE) # → full restore on join, passphrase saved to that device's login Keychain
Or conversationally:
You: "Enable sync with passphrase a1b2c3..."
Claude calls sync_enable(passphrase: "a1b2c3...") →
connects to the same iCloud sync folder → pulls all memories.
Now both devices share the same memory — and keep sharing it
automatically (background sync: 30s poll + filesystem watcher).
What syncs and what doesn't:
- Private memories (default) never leave your device. Opt in per memory:
memory_ingestwithprivacy: "shared",cortex-mcp-server ingest --privacy shared, ormemory_set_privacyon an existing memory - Demote a shared memory back to
privateand it is retracted (deleted) from your other devices — the local copy stays - All sync data is AES-256-GCM encrypted with HMAC integrity — even if your cloud account is compromised, memories stay private and tampering is detected
- Sync survives restarts: settings persist, the passphrase lives in the OS keychain, the server resumes automatically
- No server, no API, no account — just your own cloud folder
CLI alternative:
# Device A cortex-mcp-server sync enable # Save the passphrase from the output # Device B cortex-mcp-server sync enable --passphrase "your-passphrase-from-device-A" # Manual pull (background sync also pulls automatically) cortex-mcp-server sync pull
Multi-Project Isolation
Working across multiple projects? Use separate databases for physical memory isolation — no cross-project leakage, zero code changes needed.
~/.cortex/
├── global.db # User preferences, people graph, cross-project knowledge
├── my-app.db # Project A memories
└── my-api.db # Project B memories
Global config (~/.claude/settings.json) — user-level knowledge:
{
"mcpServers": {
"cortex-global": {
"command": "~/.local/bin/cortex-mcp-server",
"args": ["~/.cortex/global.db"]
}
},
"permissions": { "allow": ["mcp__cortex-global__*", "mcp__cortex-project__*"] }
}Per-project config (~/.claude/projects/<path>/settings.json) — project-specific:
{
"mcpServers": {
"cortex-project": {
"command": "~/.local/bin/cortex-mcp-server",
"args": ["~/.cortex/my-app.db"]
}
}
}Then add these memory isolation rules to your project's CLAUDE.md:
## Memory Isolation Two Cortex MCP servers: `cortex-project` (project DB) and `cortex-global` (global DB). ### Write Policy - Save to `cortex-project` if the memory is about this repo's architecture, code, modules, tests, workflows, configs, bugs, decisions, or terminology. - Save to `cortex-global` only for long-term user preferences, communication style, cross-project habits, or personal background useful across repos. - **Default: if uncertain, save to `cortex-project`.** ### Read Policy 1. Query `cortex-project` first. 2. Query `cortex-global` second, only for user-level preferences. 3. Prefer project memory when they conflict. ### Anti-Leak Rules - Never auto-copy from `cortex-project` into `cortex-global`. - Never store repo-specific paths, module names, or account names in `cortex-global`. - Never treat project implementation details as user-global preferences. ### Update Rule - Cortex is append-only. To update: search old entry → delete → ingest new.
This gives you two independent Cortex instances per project — complete isolation with shared user knowledge.
30 Tools
Tool access is governed by an optional deny-by-default capability policy: drop a
capabilities.jsonnext to your database ({"version":1,"grants":["read","write"]}) and only granted tool groups (read/write/sync/plugins/all) or exact tool names are listed and callable. No policy file = everything enabled (legacy).
| Tool | Purpose |
|---|---|
memory_ingest |
Store a memory (text, channel, person context, optional privacy) |
memory_set_privacy |
Change a memory's privacy level — promote to shared to sync it, demote to private to retract it from other devices |
memory_search |
Semantic search across all memory tiers |
memory_context |
Generate LLM-ready context summary (token-budgeted) |
memory_consolidate |
Run decay + promotion + sweep cycle |
memory_infer |
Preview inference without storing |
memory_compress |
Compress old conversation sessions |
memory_stats |
Get memory statistics (counts per tier, index size) |
memory_decay |
Run temporal decay on episodic memories |
belief_observe |
Update a Bayesian belief with evidence |
belief_list |
Query beliefs above confidence threshold |
fact_add |
Store structured knowledge (subject-predicate-object) |
fact_query |
Query facts by entity (SQL-indexed) |
preference_set |
Store user preference with confidence |
preference_query |
Query preferences by key pattern |
person_resolve |
Cross-channel identity resolution |
person_list |
List all known people |
contradiction_check |
Check for fact contradictions |
relationship_extract |
Extract relationships from text |
sync_status |
Cloud sync status (provider, devices, pending ops) |
sync_providers |
Detect available cloud storage providers |
sync_enable |
Enable cross-device cloud sync with optional encryption |
sync_pull |
Pull and apply remote changes from other devices |
memory_archive |
Archive a memory to cold storage |
memory_restore |
Restore an archived memory back to an active tier |
memory_delete |
Permanently delete a memory by ID |
memory_ingest_batch |
Ingest multiple memories in a single transaction |
tag_list_taxonomy |
List all tags in use across memories with counts |
namespace_list |
List all namespaces with memory counts |
person_merge |
Merge two person identities into one |
OpenClaw Plugin
Give your OpenClaw agent persistent memory with auto-recall and auto-capture.
Install:
# 1. Install Cortex binary curl -fsSL https://raw.githubusercontent.com/gambletan/cortex/main/install.sh | bash # 2. Install the OpenClaw plugin openclaw plugin add @cortex-ai-memory/cortex-memory
Configure (optional — works with defaults):
{
"plugins": {
"@cortex-ai-memory/cortex-memory": {
"autoCapture": true,
"autoRecall": true,
"topK": 10
}
}
}What it does:
autoCapture: Automatically stores conversation context after each turnautoRecall: Injects relevant memories before each turn (your agent "remembers")- 7 tools: memory_search, memory_store, fact_add, belief_observe, person_resolve, and more
See openclaw-plugin/README.md for full configuration options.
Project Structure
cortex/
├── cortex-core/ # Rust core library (all memory logic)
│ ├── src/
│ │ ├── lib.rs # Cortex entry point
│ │ ├── types.rs # MemObject, MemoryTier, etc.
│ │ ├── inference.rs # Proactive inference (EN + CN)
│ │ ├── episode.rs # Episodic memory store
│ │ ├── semantic.rs # Semantic facts + preferences
│ │ ├── working.rs # Working memory (session scratch pad)
│ │ ├── procedural.rs # Learned routines
│ │ ├── people.rs # People graph + identity resolution
│ │ ├── belief.rs # Bayesian belief system
│ │ ├── consolidation.rs # Episodic→semantic promotion + decay
│ │ ├── retrieval.rs # Multi-signal retrieval engine
│ │ ├── context.rs # LLM context generation
│ │ ├── sync/ # Cloud sync (oplog, HLC, merge, encryption)
│ │ └── storage/ # SQLite + in-memory vector index
│ └── benches/ # Performance benchmarks
├── cortex-http/ # HTTP REST API (axum, local-only)
├── cortex-mcp-server/ # MCP server binary (3.8MB)
├── cortex-python/ # Python bindings (PyO3, WIP)
├── openclaw-plugin/ # OpenClaw memory plugin
├── Dockerfile # Self-hosted Docker image
└── Cargo.toml # Workspace root
HTTP API
Cortex ships a lightweight HTTP server for integration with any language or framework. Binds to 127.0.0.1 by default — your data never leaves your machine.
# Build & run cargo build --release -p cortex-http ./target/release/cortex-http --port 3315 --db ~/.cortex/memory.db # Or via Docker (pre-built from GHCR) docker run -v ~/.cortex:/data -p 3315:3315 ghcr.io/gambletan/cortex/cortex-http:latest # Or build locally docker build -t cortex . docker run -v ~/.cortex:/data -p 3315:3315 cortex
Endpoints
| Method | Path | Description |
|---|---|---|
| GET | /health |
Health check |
| POST | /v1/memories |
Ingest a memory |
| POST | /v1/memories/search |
Semantic search |
| GET | /v1/memories/context |
Generate LLM context |
| POST | /v1/memories/consolidate |
Run consolidation cycle |
| POST | /v1/memories/infer |
Preview inference (no store) |
| POST | /v1/facts |
Add a semantic fact |
| POST | /v1/facts/contradictions |
Check for contradictions |
| POST | /v1/preferences |
Set a preference |
| GET | /v1/beliefs |
List beliefs |
| POST | /v1/beliefs/observe |
Update belief with evidence |
| POST | /v1/people |
Resolve person identity |
| POST | /v1/memories/compress |
Compress old conversation sessions |
| POST | /v1/relationships/extract |
Extract relationships from text |
| GET | /v1/export |
Export all data (JSON backup) |
| POST | /v1/import |
Import data from backup |
Examples
# Store a memory curl -X POST http://localhost:3315/v1/memories \ -H 'Content-Type: application/json' \ -d '{"text": "I prefer dark mode", "channel": "cli"}' # Search curl -X POST http://localhost:3315/v1/memories/search \ -H 'Content-Type: application/json' \ -d '{"query": "preferences", "limit": 5}' # Export all data (backup to iCloud, NAS, etc.) curl http://localhost:3315/v1/export > ~/iCloud/cortex-backup.json # Import from backup curl -X POST http://localhost:3315/v1/import \ -H 'Content-Type: application/json' \ -d @~/iCloud/cortex-backup.json
Roadmap
- v0.2 ✅ — Local embedding integration (all-MiniLM-L6-v2/ONNX), batch queries, importance-aware decay + auto-consolidation
- v0.3 ✅ — Proactive inference (auto-extract facts), temporal awareness, contradiction detection, Chinese NLP
- v0.4 ✅ — HTTP REST API (axum), import/export (JSON backup), Docker packaging
- v0.5 ✅ — Conversation compression, relationship inference (EN + CN), temporal retrieval enhancement, 112 tests
- v1.0 ✅ — Feature comparison table, benchmark update, 18-feature Cortex vs Mem0 vs OpenAI
- v1.1 ✅ — HNSW vector index (50K search: 12ms → 91µs), Python SDK (
pip install cortex-ai-memory) - v1.2 ✅ — Negation detection (EN + CN), multi-hop retrieval, 117 tests
- v1.3 ✅ — Context quality optimization, query expansion, bidirectional relationships, 126 tests
- v1.4 ✅ — Incremental HNSW, SQL-indexed entity queries, LLM summarizer hook, 18 MCP tools, configurable decay, LLM-assisted inference, 131 tests
- v1.5 ✅ — Docker image (GHCR auto-publish), batch ingest, dedup, namespace isolation, plugin system, event bus, archival, 351 tests
- v1.6 ✅ — Int8 quantization (75% storage reduction), materialized column indexes, FTS5 triggers, LRU caches (MemObject + entity-facts), rayon parallel decay, Arc embedding, generation-based cache invalidation, 25 MCP tools, batch inference, enhanced Chinese NLP
- v1.7 ✅ — Cloud sync (changelog-based, HLC ordering, LWW merge), AES-256-GCM encryption (Argon2id KDF), privacy enforcement (Private/Shared/Public), zeroize (memory wiping), SECURITY.md, 27 MCP tools, 400+ tests
- v2.0 ✅ — Background sync (filesystem watcher + polling), Web Dashboard, Homebrew tap, integration docs (CrewAI/AutoGen/LangGraph/DeerFlow),
/v1/memories/recentAPI, 12 rounds Codex review fixes, 489 tests - v2.1 ✅ — WASM build (124KB, runs entirely in the browser, GitHub Pages demo)
- v2.2 ✅ — Security hardening series (self-evolution iterations 11–17): manifest + per-operation HMAC, plaintext-injection rejection, timing-attack hardening, key rotation with forward secrecy (
ENC2), bounded query budget, deny-by-default MCP capability policy, per-memory privacy opt-in with cross-device retraction, persistent sync (Keychain) + auto background sync, frecency ranking, one-shot device setup script, 30 MCP tools, 500+ tests - v2.3 — Mobile targets (iOS/Android), multi-modal memory
If you find Cortex useful, please consider giving it a star ⭐ — it helps others discover the project and motivates continued development!
License
MIT


















