What is Cortexa?
Cortexa is an encrypted collective cognition engine — a desktop application that captures every decision, discovery, and architectural insight from your AI agent sessions, encrypts them locally, and uses a 7-layer autonomous cognition stack to analyze, connect, and surface knowledge across every future session.
The Problem
AI agents start every session blank. Claude Code doesn't know that you decided to use Redis for caching last month. Cursor doesn't know about the auth middleware rewrite. Your debugging insights from Project A don't surface when you hit the same pattern in Project B.
You are the institutional memory. And you forget.
Every time you re-explain context to an agent, you're paying — in tokens, in time, and in lost nuance. The decisions that shaped your codebase exist only in your head, scattered across old conversations that no agent can access.
The Solution
Cortexa sits between you and your AI agents as a persistent, encrypted intelligence layer — not just storage, but an autonomous system that thinks about your thoughts:
Layer 1: Capture & Encrypt
Every thought is encrypted with AES-256-GCM (Argon2 key derivation, passphrase in OS keychain) the moment it enters the system — from any source: global hotkey, CLI, MCP tools, HTTP API, browser extension, or Telegram bot.
Layer 2: Embed & Index
Each thought is automatically embedded into a semantic vector and indexed in an HNSW graph (O(log N) search at 100K+ items) alongside SQLite FTS5 full-text indexing. Hybrid search blends both signals with configurable weights.
Layer 3: Analyze & Orchestrate
A 49-tool orchestration engine across 9 specialized agents (Analyst, Critic, Writer, Researcher, Connector, Librarian, Visualizer, Scheduler, Historian) can decompose problems, challenge assumptions, synthesize research, and produce structured theses — triggered with one click or fully autonomous.
Layer 4: Reflect & Learn
A 7-layer collective cognition stack runs autonomously:
| Layer | What it does |
|---|---|
| Background Reflection | Finds themes, contradictions, patterns, and gaps across your thoughts |
| Cross-User Synthesis | Merges perspectives from Hive peers — agreements, disagreements, novel combinations |
| Confidence Decay | Quality scores decay weekly; stale knowledge gets flagged for review |
| Tool Observability | Every LLM call logged with duration, tokens, and success metrics |
| Social Memory | Tracks peer expertise and trust from Hive interactions |
| Procedural Memory | Remembers which analysis workflows succeed for each thought type |
| Memory Governance | Thesis quality boosts search ranking; high-confidence items surface first |
Layer 5: Surface & Connect
Agents call cortexa_context to get relevant prior decisions automatically at session start. Or use cortexa_sync_context to generate a .cortexa-context.md file that agents read without any tool call at all.
Layer 6: Share & Collaborate
The Hive P2P relay connects multiple Cortexa instances across machines and teams — synchronizing thoughts in real time with full thesis data, binary file transfer (audio, images), and automatic blog publishing.
The Compounding Loop
You work with AI agents
→ Cortexa captures decisions, bugs, patterns automatically
→ The reflection engine finds cross-cutting themes
→ Confidence decay flags stale knowledge
→ Future agent sessions start with deep context
→ Better decisions, captured automatically
→ The vault compounds over timeEvery action makes every future session smarter. Across every project. Across every agent.
Who is it for?
- Developers using AI agents (Claude Code, Cursor, Codex, Gemini) who want persistent memory across sessions
- Multi-project developers who need cross-project pattern recognition
- Teams who want shared institutional memory via the Hive P2P network
- Solo developers who capture decisions and want them to surface automatically months later
- Anyone who wants their AI tools to stop starting from zero every session
Key Differentiators
| Feature | Claude Code Memory | Notes apps | Cortexa |
|---|---|---|---|
| Cross-project | Per-project only | Manual | All projects searchable, auto-surfaced |
| Cross-agent | Claude Code only | None | Any MCP-compatible agent |
| Search | Flat text grep | Keyword | HNSW semantic + FTS5 keyword hybrid |
| Analysis | Conversation only | None | Multi-agent tool orchestrator |
| Reflection | None | None | Autonomous pattern detection + contradiction finding |
| Confidence | None | None | Temporal decay with audit trail |
| Sharing | None | Manual | Hive P2P relay with auto-publish |
| Scale | Dozens of entries | Varies | 100K+ items, sub-10ms search |
| Encryption | Plaintext files | Varies | AES-256-GCM, Argon2 KDF, OS keychain |
| Audio | None | None | TTS briefings (OpenAI/Piper), published with posts |
Architecture at a Glance
┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ Claude Code │ │ Cursor │ │ Codex │ │ Gemini CLI │
│ (20 MCP │ │ (20 MCP │ │ (20 MCP │ │ (20 MCP │
│ tools) │ │ tools) │ │ tools) │ │ tools) │
└──────┬───────┘ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘
│ │ │ │
└────────────┬────┴────────┬────────┴────────┬───────┘
│ │ │
┌─────▼─────────────▼──────────────────▼─────┐
│ Cortexa Desktop │
│ ┌───────────────────────────────────────┐ │
│ │ AES-256-GCM Encrypted SQLite + Blobs │ │
│ │ HNSW Embedding Index (100K+ items) │ │
│ │ FTS5 Full-Text Search │ │
│ │ 31-Tool Agent Orchestrator │ │
│ │ 7-Layer Cognition Stack │ │
│ │ Priority Job Queue (2 workers) │ │
│ └───────────────────────────────────────┘ │
└──────────┬─────────────────┬────────────────┘
│ │
┌─────────────▼───┐ ┌────────▼────────────┐
│ Hive Relay │ │ S3 / Filesystem │
│ (P2P sharing, │ │ (encrypted sync) │
│ auto-publish, │ │ │
│ binary files) │ │ │
└─────────────────┘ └─────────────────────┘Getting Started
Ready to try it? Head to the Installation Guide or Getting Started.