ODB-EchoGraph (OverDrive AI Agent)

Built by mr.v2k

AI coding agent using all 6 OverdriveDB engines with 60–80% token reduction, model-agnostic context switching, and persistent knowledge storage. Fully integrated via the Model Context Protocol (MCP).

🚀 How to Add to Your IDE (MCP Setup)

The best part about MCP is that your IDE automatically starts the server in the background when you open your editor. You do not need to manually start it on a port, and you don't need to run any REST API servers!

VS Code / Claude Desktop / Kiro IDE Configuration

Add the following to your IDE's MCP settings file (e.g., settings.json for VS Code, or claude.json for Claude Desktop):

{
  "mcpServers": {
    "odb-echograph": {
      "command": "npx",
      "args": ["-y", "odb-echograph"]
    }
  }
}

Note: If you are running it locally from your cloned source code instead of NPM, change the command to "node" and the arg to ["/absolute/path/to/x:/OBD-intnero/mcp-server.js"].

📊 Architecture & Data Flow

Here is how ODB-EchoGraph connects your IDE to the AI models and the 6 Database engines to save you tokens:

graph TD
    subgraph IDE [Your IDE]
        Kiro[Kiro IDE]
        VSCode[VS Code]
        Claude[Claude Desktop]
    end

    subgraph MCP [ODB-EchoGraph MCP Server]
        Router[Model Router]
        Resolver[Reference Resolver]
        Graph[Knowledge Graph]
    end

    subgraph DB [OverDriveDB Engines]
        RAM[(RAM DB)]
        Disk[(Disk DB)]
        Vec[(Vector DB)]
        Time[(TimeSeries DB)]
        Stream[(Streaming DB)]
        GraphDB[(Graph DB)]
    end

    subgraph Models [AI Models]
        Sonnet[Claude 4.6 Sonnet]
        GPT[GPT-4o]
    end

    Kiro -->|stdio| MCP
    VSCode -->|stdio| MCP
    Claude -->|stdio| MCP

    MCP --> Router
    MCP --> Resolver
    MCP --> Graph
    
    Resolver --> DB
    Graph --> DB
    Router --> Models

✨ Core Features

• Model-Agnostic: Switch between Claude, GPT-4, Gemini mid-session — context never lost.
• 60-80% Token Reduction: Compressed context graphs vs raw chat history, zero quality loss.
• Persistent Knowledge: Graph DB stores reasoning chains, code patterns, task history.
• Reference Resolution: User says "fix that" → AI knows exactly what "that" is by querying the graph.
• Zero API Required: Communicates directly over stdio via MCP. No ports, no REST API overhead.

🧠 The 6 OverDriveDB Engines

1. Graph DB: Task nodes, reasoning chains, code dependencies.
2. Vector DB: Code embeddings (384-dim) for semantic search.
3. TimeSeries DB: Token usage, latency, quality per call.
4. Streaming DB: Async task queue, agent event bus.
5. RAM DB: Current session context and model handoff snapshots.
6. Disk DB: Persistent knowledge base and patterns.

📁 Where is my Data Saved?

Because ODB-Echograph runs locally, your data is completely private.

Whenever you run the MCP server or the odb-dashboard command, OverdriveDB will create .odb database files (like agent-graph.odb and agent-vectors.odb) directly inside the folder you ran the command from.

If you open VS Code in C:\Projects\MyProject, the .odb files will be saved in MyProject, keeping your agent's memory scoped exactly to that specific project!

⚡ Performance & Token Reduction

Raw chat history approach (15,000+ tokens): [message 1] [message 2] ... [message 50] -> Full history every call

ODB-EchoGraph approach (400-600 tokens):

priorReasoning (3 summaries)   ≈ 150 tokens
recentSignatures (3 functions) ≈ 100 tokens
relatedPatterns (3 patterns)   ≈ 100 tokens
current task                   ≈ 50  tokens
─────────────────────────────────────────
Total per call                 ≈ 400-600 tokens (60-80% reduction)

Built with pride by mr.v2k AFOT (All For One Tech)