Guide: Your data (RAG)

There are three ways to put your own data in front of an agent, from available-now to on-the-roadmap.

1. Data connectors (available now)

The most direct route is an MCP connector that exposes your data as tools — the filesystem connector over a docs folder, a database connector, GitHub, and so on. Add one from the UI Connectors page (see the Connectors guide), attach it to an agent under its tool access, and the agent can search and read your data through those tools.

This needs no embeddings or indexing — the agent queries the source live. It’s the right fit for structured or searchable sources (a repo, a database, an issue tracker).

2. Per-agent memory (available now)

For durable, curated facts — conventions, glossary, “where things live” — use the agent’s memory notebook. It’s a small Markdown file injected into the system prompt at the start of every session. Edit it on the Agents page, or let the agent maintain it with the memory tool. See the coding-agent guide for how it’s used in practice. Memory is for a page of stable notes, not a document corpus.

3. Knowledge bases / RAG (roadmap)

muaz ships the plumbing for embeddings-backed retrieval — a KnowledgeStore seam and a knowledge_search tool — but no embedding backend is bundled yet. Plugins already declare their knowledge: sources in the manifest, and those sources surface on the plugin card; once a backend lands, an agent’s own plugin sources are indexed and the retrieval tool is registered automatically. Until then, reach for a data connector (option 1) for search-heavy corpora.

Recommended pattern

• Structured / searchable data → an MCP data connector.
• A page of durable facts → the agent’s memory notebook.
• A large document corpus you want semantic search over → connector today; native knowledge base when the backend ships.