Project Memory

What is RAG?

Retrieval-Augmented Generation (RAG) is the technique of giving an AI model access to external knowledge at query time. Instead of relying solely on what the model learned during training, RAG retrieves relevant information from a knowledge base and injects it into the model's context window.

Think of it as giving your project a shared brain. When anyone — human or agent — needs to understand how something works, the knowledge base serves as a single source of truth. It's not just about answering questions; it's about preserving and sharing the collective understanding of your codebase.

RAG transforms an AI agent from someone working from vague recollection into someone with a photographic memory of your entire project.

Why agents need memory

Without RAG, every coding session starts from zero. The agent has no concept of your project structure, your coding conventions, or decisions made in previous sessions. RAG changes this by giving the agent persistent, searchable memory.

• No context amnesia — Every session starts with full knowledge of your codebase. The agent already knows your folder structure, key modules, and how everything connects.
• Convention awareness — Your coding patterns, naming conventions, and architectural decisions are indexed once and recalled forever. The agent adapts to your style, not the other way around.
• Session continuity — Decisions made in past sessions are remembered. The agent picks up where it left off, understanding what was built, why it was built that way, and what comes next.
• Faster onboarding — New team members get an agent that already knows the codebase. No need to manually explain architecture — the agent reads it from the knowledge base and can explain it to them.

Beyond agents: Project Memory

Most people think of RAG as a tool that helps a single agent answer better questions. In mBm, we treat it as Project Memory — a persistent, shared knowledge base that serves the entire project lifecycle, not just individual conversations.

Project Memory means the knowledge base becomes the project's long-term memory. It outlives individual sessions, survives team changes, and accumulates wisdom over time. Every decision, every pattern, every architectural insight gets indexed and preserved.

Project Memory is the difference between an agent that helps you today and a knowledge base that helps your entire team for the life of the project.

How RAG memory works

RAG in mBm

mBm ships with a built-in RAG engine that indexes your project automatically. Here's how to use it:

# Index your project into the knowledge base
mbm rag index ./src

# The agent now remembers your code structure, conventions,
# and patterns across all sessions

# Search the knowledge base
mbm rag search "how does auth work"

# Check memory health
mbm rag status

# Reindex changed files
mbm rag reindex

The knowledge base persists across sessions and is shared with all agents working on the same project. Changes to your codebase are detected automatically — reindex only changed files to keep the memory fresh without full rebuilds.

• Persistent storage — Your knowledge base lives on your machine or server. It survives restarts, updates, and works across all sessions within a project.
• Smart reindexing — mBm tracks file changes and only reindexes what changed. No wasteful full rebuilds — just incremental updates that keep memory current.
• Multi-agent shared memory — Run multiple agents in parallel on the same project. They all share the same knowledge base, building on each other's context.

Best practices for RAG memory