| name | agentdb-init |
| description | Initialize an AgentDB Cognitive Container (.rvf file) in the current project. Sets up storage, embedder config, and the agentdb MCP server. Use when the user is starting a new project that needs vector memory, or asks to "set up agentdb" / "init agentdb". |
Initialize AgentDB
Sets up a fresh AgentDB instance for the current project. Creates a single-file .rvf Cognitive Container that holds vectors, indexes, learning state, and the audit log.
When to use
- User asks to "set up agentdb", "init memory", "add a vector store"
- New project that needs persistent agent memory
- Existing project moving from in-memory state to durable memory
Steps
- Confirm the storage path (default:
./memory.rvf at the project root, or ~/.agentdb/<project-name>.rvf for global memory).
- Pick the embedder (default:
Xenova/all-MiniLM-L6-v2 at 384d — fast, free, runs in-process).
- Register the agentdb MCP server in Claude Code:
claude mcp add agentdb -- npx agentdb@latest mcp start
- Initialize the file via the MCP tool
agentdb_pattern_store (the first store auto-creates the schema), or via CLI:
npx agentdb@latest init ./memory.rvf
- Add
*.rvf to .gitignore unless the user explicitly wants memory checked into source control.
- Confirm with a smoke test: store one pattern and search for it.
Notes
- The
.rvf is a single binary file. Back it up like a SQLite database.
- For multi-agent setups, share one
.rvf per coordinated namespace; use separate files for trust-boundary isolation.
- All AgentDB operations after init go through the MCP tools (
agentdb_pattern_*, agentdb_reflexion_*, etc.) or the npm library (import { SelfLearningRvfBackend } from 'agentdb').
Don't
- Don't commit the
.rvf file by default — it can hold session-specific data, including content from messages.
- Don't run
init on an existing .rvf file without confirming — it will refuse rather than overwrite, but a confused user might delete the existing file thinking it's stale.