Mycelium provides a persistent, company‑wide memory layer for AI agents, allowing them to retain knowledge of customers, processes, and institutional data across sessions. The memory is stored as plain‑text markdown files with Git‑versioned history, enabling audit‑grade bi‑temporal queries and easy editing with existing tools. Available as an open‑core runtime that works with Claude, GPT, Gemini, Llama, or custom models, it can be deployed multi‑tenant, single‑tenant, or on‑premise.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents in enterprises often operate without persistent knowledge of company-specific data, decisions, and processes, causing them to start each session from a blank state and limiting their usefulness in production environments.
Solution
Mycelium provides a company-wide, persistent memory layer that stores each piece of organizational knowledge as a markdown file with Git‑versioned, bi‑temporal history. The open‑core runtime supports major large language models—including Claude, GPT, Gemini, Llama, or custom models—and can be deployed as multi‑tenant managed, single‑tenant managed, or on‑premise installations. Plain‑text markdown records are directly readable and searchable with existing tools, while every write creates an immutable commit, enabling audit‑grade queries of the memory state at any past date. The commercial runtime adds enterprise features such as multi‑tenant authentication, audit logging, real‑time event streams, and webhook reliability, ensuring secure, reliable access to the memory layer across the organization.
Target Audience
Primary customers are large enterprises that have deployed multiple AI agents and need a secure, auditable, and shared knowledge base for those agents, particularly Fortune 100 organizations with complex multi‑agent AI workflows.
Features
- Markdown‑based memory records that are portable, human‑readable, and editable with standard editors
- Git‑style bi‑temporal versioning providing full audit trails and the ability to query memory as of any historical point
- Open‑core architecture with an MIT‑licensed substrate and a commercial runtime offering multi‑tenant auth, audit logs, and real‑time event streaming
- Compatibility with major LLM providers (Claude, GPT, Gemini, Llama) and custom models via a single connector per deployment
- Flexible deployment options: multi‑tenant SaaS, single‑tenant managed service, or on‑premise installation within the enterprise perimeter