Memory

Assist is new in the IDA 9.5 beta, and its documentation is subject to change. Treat it as a work in progress rather than a finished reference.

Assist keeps a persistent, per-binary knowledge base so findings survive across sessions. When you reopen a database days later, the agent can recall what was learned about it before.

How memory is stored

• One knowledge base per binary. Memory is a local SQLite database, keyed to the input file's MD5.

• Full-text searchable. The store is FTS5-backed, so the agent can search stored findings by substring and full-text query.

• Injected into context. A small, fixed number of global and local facts are added to the agent's context each turn, giving it continuity without re-reading everything.

Facts are scoped:

• Global facts apply to the whole binary (binary-wide).

• Local facts are attached to a specific address.

Open Memory with /memory to search, edit or delete stored findings, or add your own.

Check the Assist integrated guide via Settings → Guide → Memory.

What gets remembered

The agent stores findings explicitly as it works. In addition, Assist auto-captures salient conclusions from the conversation (for example, a substantive statement describing what a function does), and attaches them to the current address as a note. This keeps the knowledge base populated without requiring you to save every insight by hand.

You can also add a finding yourself, from the right-click context menu: IDA Assist → Memory → Add global/local memory.