myco-brain

Myco Brain: Persistent, Source-Traceable Memory for AI Agents

Myco Brain offers a unique solution for AI agents' memory management, allowing users to self-host on their own Postgres database without the need for API keys. Designed for teams needing to manage shared memory while ensuring stringent control over the data, Myco Brain enhances trust through its auditable and corroboration-driven confidence model.

Key Features

• Source-Traceable Facts: Each fact stored in Myco Brain is traceable back to its original document, ensuring complete transparency and eliminating unverified summaries. The auditing feature (brain_why) allows users to check the history of each fact and its sources.
• Compound Trust Model: The system is designed to recognize independent corroboration, which increases confidence in a fact. In cases of contradiction, the new information supersedes the previous record and is retained in an audited claims ledger, avoiding silent overwrites.
• Local-First Architecture: All functionalities, including full-text and semantic search, and the knowledge graph, are available without any external hosted dependencies. An API key for enhanced accuracy with an Anthropic model is optional, ensuring flexibility in use.
• Broad Compatibility: Myco Brain is compatible with multiple AI models and tools, including Claude, Cursor, Windsurf, and custom agents. It supports scenarios where different agents work with shared memory while maintaining isolation levels based on client requirements.

Usage and Quickstart

Users can quickly bootstrap their environment using Docker, setting it up with minimal configuration:

# Clone & navigate into the repository
 git clone https://github.com/thegoodguysla/myco-brain.git && cd myco-brain
# Start the services
 docker compose up -d

Once set up, agents can begin to utilize their memory:

# Ingest documents from a GitHub repository
 npx -y -p @mycobrain/mcp-server mycobrain-ingest github:your-org/your-repo
# Query the agent about the ingested memory
 "What decisions have we made about authentication and where are those documented?"

Designed for Teams and Agencies

Myco Brain is ideal for development teams needing comprehensive shared memory across agents or agencies requiring robust client isolation. It facilitates memory management by allowing users to onboard their past conversations from platforms like ChatGPT or Claude, ensuring continuity with their AI assistant right from day one.

Memory that Adapts and Evolves

The dynamic schema capability of Myco Brain allows the database to evolve based on incoming data. New entity types and relationships can be proposed automatically, ensuring that memory always reflects the most relevant and updated information without losing the integrity of prior evidence.

Secure and Isolated Workspaces

Myco Brain ensures strict data isolation for different clients. Each client operates within their own workspace, leveraging Postgres's row-level security to protect sensitive information while still sharing a foundational knowledge base across agents.

Comprehensive Documentation and Demos

To facilitate understanding and adoption, extensive documentation is available, including a quickstart guide, training materials for agents, and insight into the trust model guiding information integrity in Myco Brain. Users can also explore verified demos illustrating varied scenarios of cross-session and cross-agent recall, underpinning the application's unique capabilities.

For more information, visit the repository: GitHub - Myco Brain.