The AI agent that never forgets — and works with all its arms at once. Built by Awareness.
OCT is an AI agent platform with long-term memory, CLI + desktop + messaging workflows, MCP tools and skills, automation, and OpenClaw-compatible migration.
OCT is named after the octopus — a creature with extraordinary memory and eight arms that act in parallel. That's exactly what OCT does: it remembers everything across your conversations, and coordinates multiple agents working simultaneously on complex tasks.
Powered by Awareness Memory, which achieves 95.6% Recall@5 on LongMemEval (ICLR 2025) — the industry benchmark for long-term conversational memory. Built on OpenClaw (247K+ stars), the open-source AI agent framework.
- Quick Install: Desktop app or CLI in under a minute
- Getting Started: First memory-enabled chat in 3 minutes
- CLI vs Messaging: Choose terminal mode or channel mode (Telegram/WhatsApp/Discord)
- MCP Tools & Skills: Connect tool calls, workflows, and reusable skills
- Migrating from OpenClaw: Keep your agent workflow and add memory
- Community: Discord support and release updates
The octopus is nature's memory champion. Each arm has its own intelligence — they can act independently or in concert. OCT is built around the same idea:
| Octopus | OCT Agent |
|---|---|
| Remembers across long periods | Persistent memory across all conversations |
| 8 arms working in parallel | Multiple agents collaborating simultaneously |
| Adapts to any environment | Runs on Windows, macOS, Linux — talk via Telegram, WhatsApp, Slack |
| No central bottleneck | Each agent has its own memory + context |
| OCT Agent | Other agents | |
|---|---|---|
| Memory | 95.6% R@5 on LongMemEval (ICLR 2025) | No persistent memory, or basic summarization |
| Multi-agent | Parallel agents with shared memory pool | Single agent only |
| Install | One download, double-click — no terminal | Requires Node, Python, config files |
| Channels | Telegram, WhatsApp, WeChat, Slack, Signal, Discord | CLI only |
| Benchmark | Hybrid BM25+Vector, 0 LLM calls | LLM-dependent recall |
- 🧠 Persistent Memory — Knowledge cards, perception signals (contradiction, pattern, resonance), and cross-session recall. Your AI remembers who you are, what you decided, and what you care about.
- 🐙 Multi-Agent Collaboration — Spawn isolated agents for parallel workstreams. Each arm works independently; all arms share the same long-term memory.
- ⚡ One-Click Install — Download, double-click, done. No terminal, no Node.js, no Git required.
- 📱 Talk from Anywhere — Connect Telegram, WhatsApp, WeChat, Slack, Signal, or Discord. Your agent works while you're away.
- 🔍 Memory Benchmark Leader — 95.6% Recall@5 on LongMemEval. Zero LLM calls. Hybrid BM25+Vector search.
- 🌍 Cross-Platform — Windows, macOS, Linux. One binary, three platforms.
- 🔄 Auto Upgrade — Follows OpenClaw releases automatically. No manual updates.
OCT's memory is powered by Awareness Memory, evaluated on the industry-standard LongMemEval benchmark.
╔══════════════════════════════════════════════════════════════╗
║ Awareness Memory — LongMemEval Benchmark Results ║
║ ║
║ Recall@1 77.6% Recall@5 95.6% ◀ PRIMARY ║
║ Recall@3 91.8% Recall@10 97.4% ║
║ ║
║ Method: Hybrid RRF (BM25 + Vector) ║
║ LLM Calls: 0 Hardware: M1 8GB, 14 min ║
╚══════════════════════════════════════════════════════════════╝
┌─────────────────────────────────────────────────────────────┐
│ Long-Term Memory Retrieval — R@5 Leaderboard │
├─────────────────────────────────┬───────────┬───────────────┤
│ System │ R@5 │ Note │
├─────────────────────────────────┼───────────┼───────────────┤
│ MemPalace (ChromaDB raw) │ 96.6% │ R@5 only * │
│ ★ Awareness Memory (Hybrid) │ 95.6% │ Hybrid RRF │
│ OMEGA │ 95.4% │ QA Accuracy │
│ Supermemory │ 81.6% │ QA Accuracy │
│ Zep / Graphiti │ 71.2% │ QA Accuracy │
│ GPT-4o (full context) │ 60.6% │ QA Accuracy │
├─────────────────────────────────┴───────────┴───────────────┤
│ * MemPalace 96.6% is R@5 only, not QA Accuracy. │
└─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐
│ R@5 by Question Type │
│ │
│ knowledge-update ████████████████████████████ 100% │
│ multi-session ███████████████████████████▋ 98.5%│
│ single-session-asst ███████████████████████████▌ 98.2%│
│ temporal-reasoning █████████████████████████▊ 94.7%│
│ single-session-user ████████████████████████▎ 88.6%│
│ single-session-pref ███████████████████████▏ 86.7%│
│ │
│ Overall █████████████████████████▉ 95.6%│
└─────────────────────────────────────────────────────────────┘
Zero LLM calls. Hybrid BM25+Vector retrieval. Full benchmark details →
Download from awareness.market or Releases:
| Platform | Download |
|---|---|
| macOS | OCT.dmg |
| Windows | OCT-Agent.Setup.exe |
| Linux | OCT.AppImage |
Double-click to install. No terminal required.
npx @awareness-sdk/claw- Install OCT Desktop or run the CLI installer.
- Send your first message in chat.
- Ask a follow-up question that depends on prior context.
- Verify that memory recall is used automatically.
This is the fastest way to validate OCT as an AI agent with persistent memory rather than a stateless assistant.
OCT supports both CLI workflows and messaging workflows.
- CLI mode: Best for developers, scripts, and terminal-first automation.
- Messaging mode: Best for teams and non-technical users via Telegram, WhatsApp, WeChat, Slack, Signal, and Discord.
Use CLI when you need deterministic command/control. Use messaging when you need always-on agent access from mobile and team chat.
OCT works with MCP tools and reusable skills so agents can call external capabilities safely and repeatedly.
- Use MCP tools for structured integrations and tool-calling workflows.
- Use skills for repeatable playbooks and team-level automation.
- Combine memory + tools + skills to reduce repeated prompt engineering.
Already using OpenClaw? OCT is designed for OpenClaw-compatible migration.
- OCT wraps OpenClaw instead of forking/replacing it.
- Keep existing agent workflows and channels.
- Add long-term memory and desktop UX incrementally.
Migration path: install OCT, validate first chat, then move your daily workflow gradually.
- Docs: awareness.market
- Releases: GitHub Releases
- Discord: Join Community
If OCT is useful to your team, please star the repo to help more builders discover it.
Every conversation feeds OCT's memory loop:
You talk to OCT
↓
OCT captures knowledge cards
(decisions, preferences, facts, contradictions)
↓
Awareness Memory indexes them
(BM25 + Vector hybrid, 0 LLM calls)
↓
Next conversation: OCT recalls relevant context
before you even ask
↓
Your AI gets smarter with every session
Memory types OCT tracks:
- Knowledge cards — facts, decisions, preferences you've stated
- Perception signals — contradictions, recurring patterns, strong resonance moments
- Temporal reasoning — what you knew when, how your views changed over time
- Cross-session continuity — picks up exactly where you left off
OCT Agent (this project)
└── wraps OpenClaw (open-source AI agent, 247K+ stars)
└── pre-configured with Awareness Memory plugin
└── hybrid search (BM25 + Vector, Hybrid RRF)
└── perception signals (contradiction, pattern, resonance)
└── cross-device sync via Awareness cloud
# CLI
cd packages/cli && npm start
# Desktop
cd packages/desktop && npm run devApache 2.0 — Built by Awareness.