# RavenAgent Guide

Last reviewed: 2026-04-24

Primary source surfaces:

- `/private/agent`
- RavenAgent context and semantic retrieval implementation

## What RavenAgent Uses

RavenAgent can use:

- recent thread memory
- selected document context when the user pins a file
- semantic knowledge-base matches
- semantic artifact matches
- compare history
- contacts and contact semantic retrieval
- job listener and job listing signals
- candidate profile and resume generation records

## Context Selection

Users do not always need to select a document. The context selector can be left blank so Raven can
reason about intent and retrieve relevant context.

Raven should route retrieval by intent:

- product support question: product support KB
- job-search strategy question: job-seeker support KB
- employer-specific question: employer KB plus relevant candidate artifacts
- offer or document-specific question: uploaded artifact
- networking question: contacts, job listeners, and relevant files
- job fit question: job posting, candidate profile, compare history, and company KB

## Source Discipline

RavenAgent should distinguish:

- product support KB
- job seeker support KB
- counselor-authored KB
- candidate-private artifacts
- generated artifacts
- contacts
- job listings

If source information is unavailable, Raven should say what it can answer from general product
behavior and what needs human support or additional context.