# cognitive-ai-memory [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/sachitrafa/cognitive-ai-memory  
**GitHub Stars:** 270  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/cognitive-ai-memory

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

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "cognitive-ai-memory": {
    "command": "uvx",
    "args": ["yourmemory"]
  }
}
```

## Documentation

## What the cognitive-ai-memory MCP server does

The cognitive-ai-memory MCP server adds persistent memory to MCP-compatible AI clients. It extracts durable facts from conversations, stores them locally, and makes relevant information available in later sessions. The system is designed to avoid treating memory as an ever-growing collection of undifferentiated records.

Related facts can be consolidated into a shorter summary while the original records are archived. Memories also lose strength according to an Ebbinghaus-style forgetting curve. Frequently recalled or important information is more likely to remain available, while stale and unused facts can fade and be pruned.

Entity relationships connect memories through shared people, places, and concepts. This allows retrieval to follow related facts even when a second fact has weak direct similarity to the user’s query. The project also describes support for context resets, role-based team memory pools, export, right-to-forget operations, and a hash-chained audit record for reads, writes, and deletes.

## How it works

The system extracts candidate facts, removes duplicates, creates embeddings, and links entities before saving the results in its memory store. Retrieval combines vector search, BM25 text search, and graph relationships. Related records may later be compressed into one summary, with source records retained in an archive rather than immediately deleted.

Memory strength changes with elapsed time and recall activity. This approach is intended for agents that need long-term continuity instead of only storing the current context window.

## Setup and configuration

Install the Python package with Python 3.11 through 3.14, then register a token and run the setup command:

```bash
pip install yourmemory
yourmemory-register <your-token>
yourmemory-setup
```

The token is obtained from the project’s website through email verification. Setup can detect and configure Claude Code, Claude Desktop, Cursor, Windsurf, and Cline. Memory is stored under `~/.yourmemory/`.

DuckDB is the default backend and requires no separate database service. Postgres is available for shared or production use and requires a `DATABASE_URL` plus the pgvector extension. Built-in heuristics support local extraction by default. Ollama can provide higher-quality local extraction, while `YOURMEMORY_EXTRACT_BACKEND=anthropic` selects the documented cloud extraction option.

A standalone binary is also available for macOS Apple Silicon, macOS Intel, Linux x86-64, and Windows x86-64. The binary includes its Python runtime, dependencies, embedding model, and spaCy, but is approximately 2 GB.

## Tools and capabilities

- Extract and persist durable facts from agent interactions.
- Recall memories using vector, BM25, and entity-graph retrieval.
- Consolidate related memories into summaries.
- Apply time-based decay and prune stale information.
- Export stored data and purge memories.
- Record memory reads, writes, and deletes in a hash-chained audit trail.

## Limitations and notes

The default DuckDB configuration is local and suited to a single installation; shared use requires Postgres and pgvector. Cloud extraction is optional, while local extraction can use heuristics or Ollama. The standalone binary avoids first-run downloads but has a large download size. The project’s license is CC BY-NC 4.0, which should be reviewed before commercial use.

_Full upstream README: https://allmcps.com/mcp/cognitive-ai-memory/readme_