Docs

ModelBrain exposes a small surface to your assistant over MCP — five tools, deliberately. Everything an assistant needs to remember, recall, correct and forget is here; everything that manages the installation is not, and stays on the command line.

The five MCP tools
Tool
Behavior
Returns
remember
Stores a fact, decision, or correction, tagged with where it came from and when.
memory id, timestamp
recall
Searches stored memories for what's relevant to the current conversation.
ranked memories, each dated + sourced
forget
Removes a memory or a whole document's worth of memories. Undoes a correction, or deletes outright.
confirmation, count removed
expand
Pulls more context around a memory that came back too thin — the surrounding paragraph, the rest of the document.
surrounding content, source ref
connector_sync
Triggers an on-demand re-scan of a connected folder or source, rather than waiting for the next automatic pass.
files scanned, memories added
{ "tool": "recall", "query": "what did we decide about the pricing tiers", "results": [ { "memory_id": "mem_8f2a1c", "text": "Free tier stays local-only, no device cap.", "source": "conversation · 2026-08-14", "confidence": 0.91 } ] }
Deliberately not over MCP

These manage the installation itself, not a single conversation, so they stay on the command line rather than something an assistant could be talked into calling mid-chat.

connector_add / connector_remove / connector_list

Which folders and sources feed the vault.

backup / restore

Manual snapshot and recovery of the local vault.

housekeeping

Storage-tier maintenance — compaction, tier migration.

Architecture

How the four storage tiers, the daemon and your assistant fit together.

Read the architecture →
Recall test

The evaluation these tools are held to before release.

Read the test →