← Index

io.github.sachitrafa/cognitive-ai-memory

io.github.sachitrafa/cognitive-ai-memory·v1.0.2·Memory & RAG
Quality Score
84
/100

Ebbinghaus-based persistent memory for Claude. Memories decay with time, strengthen on recall.

§01  Install
Claude Desktop (Docker)
{
  "mcpServers": {
    "cognitive-ai-memory": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "ghcr.io/your-memory/cognitive-ai-memory:1.0.2"
      ],
      "env": {
        "DATABASE_URL": "<database_url>",
        "OLLAMA_URL": "<ollama_url>",
        "EXTRACT_MODEL": "<extract_model>"
      }
    }
  }
}
§02  Environment variables
DATABASE_URL
required

PostgreSQL connection string with pgvector. Example: postgresql://localhost:5432/yourmemory

OLLAMA_URL

Ollama server URL for local embeddings (nomic-embed-text) and classification (llama3.2:3b). Default: http://localhost:11434

EXTRACT_MODEL

Ollama model used for fact/assumption classification. Default: llama3.2:3b

§03  MCP Quality Score  ·  methodology
freshness
24
completeness
10
installability
25
documentation
15
stability
10
§04  Alternatives in Memory & RAG