Profile Memory
Archer's persistent memory layer of facts, evidence spans, and evidence links across every document and interview in your account.
Profile Memory is the persistent layer Archer uses to remember everything about you: the facts it has learned, the exact evidence in your documents that supports each fact, and the links between them.
Every stage of the agentic engine reads from Profile Memory, and every stage can contribute new entries to it. It is how Admit Architect never forgets something you told it in week one when you come back in week eight.
Three primitives
Profile Memory has three primitives.
Facts are statements about you. Things like "led a team of six engineers at Stripe", "scored 740 on the GMAT", "wants to work in climate investing after MBA". Each fact has a type, a value, and a confidence, and is owned by your account.
Evidence Spans are verbatim excerpts from source documents, such as a paragraph from your resume, a sentence from an interview answer, or a line from a file you uploaded. They are stored with a pointer back to the exact location in the source.
Evidence Links connect a fact to one or more evidence spans, so every claim in Profile Memory is auditable. You can always trace back from a fact to the exact words that support it.
How facts get in
Facts appear in Profile Memory in three ways. The engine extracts them during the Ingest stage (when you upload a resume or finish an interview). You can add one directly from a Mind session. Or Archer can propose one in the middle of another action, and the addition happens immediately because creating a new fact is an additive, safe operation.
Facts are append-only in practice, but you can edit or correct one at any time. Archer will flag a fact whose evidence span no longer matches it, so stale facts are surfaced instead of silently rotting.
How evidence links make the engine honest
Because every fact is linked to evidence, every downstream stage of the engine is forced to support its claims with something concrete. Your Dean Brief cites the exact answer span behind each rubric comment. Your Essay Workbench can show you evidence gaps before you submit a draft. Your Connections outreach can reference specific shared context that the engine actually has evidence for.
How it fits in the engine
Profile Memory is the substrate every other stage reads from. Ingest writes to it. Interrogate writes to it. Synthesize reads and writes. Draft reads it heavily. Orchestrate reads it for matching in Connections. When you hear Archer "remembering something", this is literally where it remembers.