Previously, every Squirrel conversation started from zero. You'd correct a misinterpretation, explain a naming convention, or clarify how your team refers to a specific contract type, and next time you'd have to do it again.
Now Squirrel persists knowledge across conversations. When you correct an entity, clarify a term, or express a preference, that information gets stored and recalled in future chats.
Organization level
Knowledge that applies across your entire tenant. Examples: your company's standard contract categories, how you classify document types, naming conventions for assets.
Asset level
Knowledge specific to a particular asset. Examples: "T5 at Pauvres is actually T6" or "the O&M provider changed from Vestas to GE in 2024."
User level
Personal preferences per user. Examples: preferred language for answers, how you like compliance summaries formatted, which KPIs you care about most.
Squirrel doesn't decide what to remember on its own. A dedicated helper agent called Walnut handles memory management. Walnut evaluates each conversation, extracts only what's genuinely useful, and stores it at the right level. This keeps memory lean and relevant rather than dumping every interaction into a growing pile of noise.