{"slug": "awareness-local-local-first-memory-for-ai-coding-agents-96-r-5", "title": "Awareness Local: local-first memory for AI coding agents (96% R 5)", "summary": "Awareness Local, a local-first MCP memory server for AI coding agents, achieves a 96% Recall@5 on the LongMemEval benchmark (ICLR 2025) with 500 human-curated questions, running fully offline on an Apple M1 with 8GB RAM and zero LLM calls. The tool provides persistent cross-session memory for Cursor, Claude Code, Copilot, Cline, and other MCP-compatible IDEs, storing memory as Markdown and indexing with SQLite FTS5 and embeddings.", "body_md": "**Languages:** English | [简体中文](/everest-an/Awareness-Market/blob/main/README.zh-CN.md)\n\n**Give your AI agent persistent memory. One command. No account. Works offline.**\n\nAwareness Local is a local-first MCP memory server for AI coding agents. It gives Cursor, Claude Code, Copilot, Cline, and other MCP IDEs persistent memory, hybrid semantic + keyword retrieval, and reusable knowledge cards for long-running software projects.\n\nIt runs a lightweight daemon on your machine, stores memory as Markdown, indexes recall with SQLite FTS5 + embeddings, and keeps your AI workflow fast, explainable, and offline-ready.\n\n```\nnpx @awareness.market/setup\n```\n\nThat's it. Your AI agent now remembers everything across sessions.\n\nAI coding agents lose context between sessions. Awareness Local provides cross-session memory recall so agents can continue work without re-explaining architecture, past decisions, pending tasks, and implementation constraints.\n\n- Persistent memory for AI coding agents\n- Local-first MCP server with offline support\n- Hybrid retrieval (keyword + semantic)\n- Knowledge card extraction for decisions, solutions, and risks\n\n```\nnpx @awareness.market/setup\n```\n\nThen open your IDE and start coding. Awareness tools become available for recall, record, and session initialization.\n\n- Long-running codebase migrations across many sessions\n- Team handoffs where AI should remember prior implementation context\n- Personal coding workflows that need durable preferences and conventions\n- Multi-agent setups that share decision history and task memory\n\nYes. Local mode works fully offline with memory stored on your machine.\n\nMemory is stored as Markdown in `.awareness/`\n\n, with a local SQLite index for retrieval.\n\nNo. Cloud sync is optional and can be enabled later.\n\nAny MCP-compatible IDE, including Cursor, Claude Code, Copilot, Cline, Windsurf, and others.\n\nEvaluated on ** LongMemEval** — the industry standard benchmark for long-term conversational memory. 500 human-curated questions across 5 core capabilities.\n\n```\n╔══════════════════════════════════════════════════════════════╗\n║                                                              ║\n║   Awareness Memory — LongMemEval Benchmark Results           ║\n║   ─────────────────────────────────────────────────           ║\n║                                                              ║\n║   Benchmark:  LongMemEval (ICLR 2025)                       ║\n║   Dataset:    500 human-curated questions                    ║\n║   Variant:    LongMemEval_S (~115k tokens per question)      ║\n║                                                              ║\n║   ┌─────────────────────────────────────────────────┐        ║\n║   │                                                 │        ║\n║   │   Recall@1    80.2%    (401 / 500)              │        ║\n║   │   Recall@3    92.8%    (464 / 500)              │        ║\n║   │   Recall@5    96.0%    (480 / 500)  ◀ PRIMARY   │        ║\n║   │   Recall@10   98.6%    (493 / 500)              │        ║\n║   │                                                 │        ║\n║   └─────────────────────────────────────────────────┘        ║\n║                                                              ║\n║   Method:     Hybrid RRF (BM25 + vector, daemon pipeline)    ║\n║   Embedding:  multilingual-e5-small (production model)       ║\n║   LLM Calls:  0  (pure retrieval, no generation cost)        ║\n║   Hardware:   Apple M1, 8GB RAM — 35 min total               ║\n║                                                              ║\n╚══════════════════════════════════════════════════════════════╝\n┌─────────────────────────────────────────────────────────────┐\n│          Long-Term Memory Retrieval — R@5 Leaderboard       │\n│          LongMemEval (ICLR 2025, 500 questions)             │\n├─────────────────────────────────┬───────────┬───────────────┤\n│  System                         │  R@5      │  Note         │\n├─────────────────────────────────┼───────────┼───────────────┤\n│  MemPalace (ChromaDB raw)       │  96.6%    │  R@5 only *   │\n│  ★ Awareness Memory (Hybrid)    │  96.0%    │  Hybrid RRF   │\n│  OMEGA                          │  95.4%    │  QA Accuracy  │\n│  Mastra (GPT-5-mini)            │  94.9%    │  QA Accuracy  │\n│  Mastra (GPT-4o)                │  84.2%    │  QA Accuracy  │\n│  Supermemory                    │  81.6%    │  QA Accuracy  │\n│  Zep / Graphiti                 │  71.2%    │  QA Accuracy  │\n│  GPT-4o (full context)          │  60.6%    │  QA Accuracy  │\n├─────────────────────────────────┴───────────┴───────────────┤\n│  * MemPalace 96.6% is Recall@5 only, not QA Accuracy.      │\n│    Palace hierarchy was NOT used in the evaluation.         │\n└─────────────────────────────────────────────────────────────┘\n┌─────────────────────────────────────────────────────────────┐\n│     Awareness Memory — R@5 by Question Type                 │\n│                                                             │\n│  knowledge-update        ███████████████████████████ 98.7%  │\n│  multi-session           ███████████████████████████▊  99.2%│\n│  single-session-asst     ███████████████████████████▌  98.2%│\n│  temporal-reasoning      ██████████████████████████▏   93.2%│\n│  single-session-user     ██████████████████████████    92.9%│\n│  single-session-pref     █████████████████████████▎    90.0%│\n│                                                             │\n│  Overall                 ██████████████████████████▉   96.0%│\n│                                                             │\n│  ┌───────────────────────────────────────────────┐          │\n│  │  Ablation Study                               │          │\n│  │  ─────────────────────────────────────────    │          │\n│  │  Vector-only:   92.6%  ▓▓▓▓▓▓▓▓▓▓▓▓▓░░░     │          │\n│  │  BM25-only:     91.4%  ▓▓▓▓▓▓▓▓▓▓▓▓▓░░░     │          │\n│  │  Hybrid RRF:    95.6%  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓░  ★  │          │\n│  │  (2026-04 harness run)                        │          │\n│  │  Hybrid = +3% over any single method          │          │\n│  └───────────────────────────────────────────────┘          │\n│                                                             │\n│  arxiv.org/abs/2410.10813          awareness.market         │\n└─────────────────────────────────────────────────────────────┘\n```\n\nZero LLM calls on retrieval (daemon path). [Reproducible benchmark scripts →](https://github.com/everest-an/Awareness-Market/tree/main/benchmarks/longmemeval)\n\n**Before:** Every session starts from scratch. You re-explain the codebase, re-justify decisions, watch the agent redo work.\n\n**After:** Your agent says *\"I remember you were migrating from MySQL to PostgreSQL. Last session you completed the schema changes and had 2 TODOs remaining...\"*\n\n```\nSession 1                          Session 2\n┌─────────────────────────┐       ┌─────────────────────────┐\n│ Agent: \"What database?\" │       │ Agent: \"I remember we   │\n│ You: \"PostgreSQL...\"    │       │ chose PostgreSQL for     │\n│ Agent: \"What framework?\"│  →    │ JSON support. You had    │\n│ You: \"FastAPI...\"       │       │ 2 TODOs left. Let me     │\n│ (repeat every session)  │       │ continue from there.\"    │\n└─────────────────────────┘       └─────────────────────────┘\n```\n\n| IDE | Auto-detected | Plugin |\n|---|---|---|\nClaude Code |\n✅ |\n`awareness-memory` |\n\n**Cursor****Windsurf****OpenClaw**`@awareness.market/openclaw-memory`\n\n**Cline****GitHub Copilot****Codex CLI****Kiro****Trae****Zed****JetBrains (Junie)****Augment****AntiGravity (Jules)**\n\n```\nYour IDE / AI Agent\n    │\n    │  MCP Protocol (localhost:37800)\n    ▼\n┌────────────────────────────────────┐\n│  Awareness Local Daemon            │\n│                                    │\n│  Markdown files    → Human-readable, git-friendly\n│  SQLite FTS5       → Fast keyword search\n│  Local embedding   → Semantic search (optional: npm i @huggingface/transformers)\n│  Knowledge cards   → Auto-extracted decisions, solutions, risks\n│  Web Dashboard     → http://localhost:37800/\n│                                    │\n│  Cloud sync (optional)             │\n│  → One-click device-auth           │\n│  → Bidirectional sync              │\n│  → Semantic vector search          │\n│  → Team collaboration              │\n└────────────────────────────────────┘\n```\n\nAll memories stored as **Markdown files** in `.awareness/`\n\n— human-readable, editable, git-friendly:\n\n```\n.awareness/\n├── memories/\n│   ├── 2026-03-22_decided-to-use-postgresql.md\n│   ├── 2026-03-22_fixed-auth-bug.md\n│   └── ...\n├── knowledge/\n│   ├── decisions/postgresql-over-mysql.md\n│   └── solutions/auth-token-refresh.md\n├── tasks/\n│   └── open/implement-rate-limiting.md\n└── index.db  (search index, auto-rebuilt)\n```\n\n| Tool | What it does |\n|---|---|\n`awareness_init` |\nLoad session context — recent knowledge, tasks, rules |\n`awareness_recall` |\nSearch memories — progressive disclosure (summary → full) |\n`awareness_record` |\nSave decisions, code changes, insights — with knowledge extraction |\n`awareness_lookup` |\nFast lookup — tasks, knowledge cards, session history, risks |\n`awareness_get_agent_prompt` |\nGet agent-specific prompts for multi-agent setups |\n\nInstead of dumping everything into context, Awareness uses a two-phase recall:\n\n```\nPhase 1: awareness_recall(query, detail=\"summary\")\n  → Lightweight index (~80 tokens each): title + summary + score\n  → Agent reviews and picks what's relevant\n\nPhase 2: awareness_recall(detail=\"full\", ids=[...])\n  → Complete content for selected items only\n  → No truncation, no wasted tokens\n```\n\nVisit `http://localhost:37800/`\n\nto browse memories, knowledge cards, tasks, and manage cloud sync.\n\nConnect to [Awareness Cloud](https://awareness.market) for:\n\n- Semantic vector search (100+ languages)\n- Cross-device real-time sync\n- Team collaboration\n- Memory marketplace\n\n```\nnpx @awareness.market/setup --cloud\n# Or click \"Connect to Cloud\" in the dashboard\n```\n\nAwareness Local is part of the Awareness ecosystem:\n\n| Package | For | Install |\n|---|---|---|\n|\nLocal daemon + MCP server | `npx @awareness.market/setup` |\n|\n`wrap_openai()` / `wrap_anthropic()` interceptors |\n`pip install awareness-memory-cloud` |\n|\n`wrapOpenAI()` / `wrapAnthropic()` interceptors |\n`npm i @awareness-sdk/memory-cloud` |\n|\nAuto-recall + auto-capture | `openclaw plugins install @awareness.market/openclaw-memory` |\n|\nSkills + hooks | `/plugin marketplace add everest-an/Awareness-SDK` → `/plugin install awareness-memory@awareness` |\n|\nOne-command setup for 13+ IDEs | `npx @awareness.market/setup` |\n\nFull SDK docs: [awareness.market/docs](https://awareness.market/docs)\n\n- Node.js 18+\n- Any MCP-compatible IDE\n\nNo Python, no Docker, no cloud account needed.\n\nIf Awareness Local saves you from re-explaining your codebase to your AI agent, give it a ⭐ — it helps more developers discover the project and pushes it toward GitHub Trending.\n\nMIT\n\n**IDE Support:** [Cursor](https://cursor.com), [Windsurf](https://codeium.com/windsurf), [Trae](https://www.trae.sh), [Zed](https://zed.dev), VS Code, JetBrains.\n**Compatible with:** OpenClaw, AutoGPT, LangChain, MetaGPT.\n**Key Technology:** OMP (Open Memory Protocol), LatentMAS, Shared Thought Space, One-click Deployment.\n**Focus:** Solving AI \"Lobster Memory\" (Long-term memory loss), Automating complex workflows, Simplifying Agent setup.", "url": "https://wpnews.pro/news/awareness-local-local-first-memory-for-ai-coding-agents-96-r-5", "canonical_source": "https://github.com/everest-an/Awareness-Market", "published_at": "2026-08-28 01:02:35+00:00", "updated_at": "2026-08-28 01:18:04.524368+00:00", "lang": "en", "topics": ["ai-tools", "ai-agents", "ai-infrastructure", "machine-learning"], "entities": ["Awareness Local", "LongMemEval", "Cursor", "Claude Code", "Copilot", "Cline", "SQLite", "Apple M1"], "alternates": {"html": "https://wpnews.pro/news/awareness-local-local-first-memory-for-ai-coding-agents-96-r-5", "markdown": "https://wpnews.pro/news/awareness-local-local-first-memory-for-ai-coding-agents-96-r-5.md", "text": "https://wpnews.pro/news/awareness-local-local-first-memory-for-ai-coding-agents-96-r-5.txt", "jsonld": "https://wpnews.pro/news/awareness-local-local-first-memory-for-ai-coding-agents-96-r-5.jsonld"}}