🧠 AI agents forget. AgentOS remembers.
Works across Claude Code, Gemini CLI, Codex, and Kimi CLI. One shared memory. All agents. All sessions.
🚀 140+ developers already tried AgentOS in the first days after launch
AI coding tools are powerful — but stateless.
You fix a bug today. Tomorrow, it's gone.
Worse:
- Long sessions get compacted — context disappears
- Switching tools means starting over
- Each CLI has its own isolated memory
AgentOS solves this by making experience persistent. Not prompts. Not docs. Actual outcomes: what worked, what failed, what decisions were made.
SESSION 1 — Monday morning
> Fix CORS error in Flask API
Agent fixes it: pip install flask-cors && CORS(app)
✅ Solution saved to memory.
... 3 months later ...
SESSION 847 — Thursday evening
> I'm getting a CORS error again
🧠 EAN AgentOS: You solved this before.
$ mem suggest 'CORS error'
💡 SOLUTIONS FOR: CORS error
1. [51pts] ✅ CORS error: blocked by CORS policy
Solution: Add flask-cors middleware: CORS(app)
Code: pip install flask-cors && CORS(app)
# Never solve the same bug twice.
Works with Claude Code, Gemini CLI, Codex CLI, and Kimi CLI. All agents share the same memory.
- Developers using AI coding tools daily
- People working across multiple CLI agents
- Anyone tired of repeating the same fixes
- Builders working on long-running projects
If you've ever said: "Didn't I already fix this before?" — this is for you.
| Feature | Description |
|---|---|
| 🔁 Persistent Memory | Decisions, facts, goals, tasks — persisted across sessions |
💡 mem suggest |
Find past solutions: "Never solve the same bug twice" |
| 🧬 Knowledge Extraction | Auto-extract patterns, scoring, deduplication |
| 🔍 Cognitive Search | Search across resolutions, decisions, facts, messages |
| 🌳 Memory Branches | Branch-aware memory per git branch |
| 📊 Experience Graph | Problem → solution → outcome graph |
| 🔄 Cross-Agent Learning | Agents learn from each other's experience |
| 💾 Backup & Recovery | Auto backup, restore, integrity verification |
| 🖥️ Web Dashboard | Visualize decisions, facts, timeline, health |
| 🔌 MCP Server | Native integration with Claude Code + other CLIs |
You need at least one AI coding CLI installed first:
| CLI | Install | Required? |
|---|---|---|
| Claude Code | npm install -g @anthropic-ai/claude-code |
At least one |
| Gemini CLI | npm install -g @google/gemini-cli |
At least one |
| Codex CLI | npm install -g @openai/codex |
At least one |
| Kimi CLI | pip install kimi-cli |
At least one |
Without a CLI, EAN AgentOS has nothing to capture. Install at least one, then proceed:
git clone https://github.com/eanai-ro/ean-agentos.git
cd ean-agentos
./install.shThe installer auto-detects installed CLIs and lets you choose which to integrate.
pip install flask flask-cors
python3 scripts/init_db.py
python3 scripts/ean_memory.py install claude # or gemini, codexmem suggest "CORS error"💡 SOLUTIONS FOR: CORS error
══════════════════════════════════════════════════════════════
1. [92pts] ✅ error_resolutions#5
Problem: CORS error: blocked by CORS policy
Solution: Add flask-cors middleware: CORS(app)
Match: 100% | Confidence: 92% | Agent: claude-code
mem suggest "error message" # Find past solutions
mem search "keyword" # Search all memory
mem decisions # View active decisions
mem status # Memory status
mem graph stats # Experience graph statspython3 scripts/web_server.py
# Open: http://localhost:19876Configured automatically during install. Your AI agent receives context from permanent memory at every session.
| CLI | Integration | Install Command |
|---|---|---|
| Claude Code | Hooks + MCP Server | python3 scripts/ean_memory.py install claude |
| Gemini CLI | Hooks | python3 scripts/ean_memory.py install gemini |
| Codex CLI | Hooks | python3 scripts/ean_memory.py install codex |
| Kimi CLI | MCP Server | Manual config |
All CLIs read and write to the same database. What one agent learns is available to all.
Claude Code Gemini CLI Codex CLI Kimi CLI
│ │ │ │
│ auto-capture (hooks) │
└──────────────┬─────────────┬────────────┘
│ │
▼ ▼
┌──────────────────────────┐
│ global.db │
│ │
│ decisions │
│ learned_facts │
│ error_resolutions │
│ experience_graph │
│ solution_index │
│ ...52 tables │
└──────────────────────────┘
│
┌───────────┼───────────┐
▼ ▼ ▼
REST API MCP Server CLI (mem)
Dashboard (Claude) (terminal)
- Capture: Hooks auto-capture decisions, errors, solutions from AI sessions
- Structure: Knowledge Extractor classifies, scores, and deduplicates
- Retrieve: At each new session, the agent receives relevant context
- Learn: Solution Index + Experience Graph = memory gets smarter
ean-agentos/
├── scripts/
│ ├── mem # Main CLI
│ ├── v2_common.py # Core DB + utilities
│ ├── init_db.py # DB initialization
│ ├── solution_index.py # 💡 mem suggest
│ ├── knowledge_extractor.py # Auto-extraction
│ ├── context_builder_v2.py # LLM context builder
│ ├── experience_graph.py # Experience graph
│ ├── search_memory.py # Unified search
│ ├── backup_manager.py # Backup & restore
│ ├── web_server.py # Web dashboard
│ ├── ean_memory.py # Installer
│ └── ...
├── web/ # Dashboard HTML/JS/CSS
├── mcp_server/ # MCP for Claude Code
├── mcp-server/ # MCP for Kimi CLI
├── migrations/ # DB schema
├── install.sh # Interactive installer
├── test_full.sh # Test suite (48 tests)
└── Dockerfile # Test container
./test_full.sh48 tests covering: structure, database, imports, mem suggest, experience graph, context builder, search, web server, MCP, DB integrity, backup.
| Feature | Included |
|---|---|
| ✅ Persistent memory (decisions, facts, goals, tasks, errors) | Yes |
✅ mem suggest — find past solutions instantly |
Yes |
| ✅ Knowledge extraction with pattern detection | Yes |
| ✅ Experience graph (problem → solution → outcome) | Yes |
| ✅ Context builder (compact/full/survival) | Yes |
| ✅ Cognitive search across all memory | Yes |
| ✅ Cross-agent learning | Yes |
| ✅ Branch-aware memory | Yes |
| ✅ Backup & recovery | Yes |
| ✅ Web dashboard | Yes |
| ✅ MCP server (Claude Code integration) | Yes |
✅ CLI tool (mem) with 30+ commands |
Yes |
| ✅ Hooks for Claude Code, Gemini CLI, Codex CLI, Kimi CLI | Yes |
| ✅ 52 database tables, 48 tests | Yes |
Pro includes the full Free edition (persistent memory, mem suggest, knowledge extraction, dashboard, all CLI integrations) plus multi-agent orchestration capabilities:
| Feature | Free | Pro |
|---|---|---|
Persistent memory, mem suggest, knowledge extraction |
✅ | ✅ |
| Web dashboard, MCP server, 30+ CLI commands | ✅ | ✅ |
| All 4 CLI integrations (Claude, Gemini, Codex, Kimi) | ✅ | ✅ |
| Multi-agent orchestration (projects, tasks, leases) | — | ✅ |
| AI deliberation (multi-round, voting, synthesis) | — | ✅ |
| CLI Launcher (launch Claude/Gemini/Codex/Kimi programmatically) | — | ✅ |
| Auto-pipeline (task chaining, auto-review, conflict resolution) | — | ✅ |
| Intelligence layer (capability scoring, weighted voting) | — | ✅ |
| Skill learning (learns from reviews) | — | ✅ |
| Replay system (project + deliberation timelines) | — | ✅ |
| Peer review workflow (formal verdicts, auto-fix) | — | ✅ |
3 AI agents propose, critique each other, and reach consensus on a 16-point VPS hardening checklist. No single agent covered all points — the mutual critique filled the gaps. Full demo →
Contact for Pro: ean@eanai.ro
Real output from multi-agent orchestration sessions — see all demos:
| Demo | What it shows |
|---|---|
| Architecture | 3 agents design a REST API (testing + security + backend) |
| SQL vs NoSQL | Agents debate, disagree, and reach consensus |
| Code Review | 3 agents find the same vulnerability independently |
| Security Audit | SQL injection + weak JWT found by all 3 agents |
| VPS Hardening | Full deliberation: propose → critique → 16-point checklist |
| Debugging | 3 agents diagnose a production memory leak |
| CI/CD Pipeline | Agents design deployment pipeline together |
| Performance | 3 agents optimize a slow API endpoint |
Got questions, ideas, or feedback?
Share your use case, report bugs, or suggest improvements.
If you try this, I'd love to hear what worked — and especially what didn't.
This project is still evolving based on real workflows, so your feedback matters a lot.
If you find it useful, consider giving it a star ⭐ It helps more people discover it.
MIT — see LICENSE
Built by EAN (Encean Alexandru Nicolae) 🇷🇴
Stop restarting. Start building on what you already solved.