Docs

Memory & Knowledge

Memory & Knowledge

How agents learn and use documents.

Conduity agents have two forms of persistent context: Memory and Knowledge Base.

Memory

Agents automatically extract and store facts, preferences, and decisions from conversations. Memory entries are categorized as:

  • Fact — Objective information ("The API uses REST")
  • Preference — User preferences ("Prefers concise responses")
  • Decision — Decisions made ("We chose PostgreSQL")
  • Context — Background context ("Working on Q1 launch")

View and manage memories from the Memory page. You can edit or delete individual entries.

Knowledge Base

Upload documents (PDF, Markdown, text) to the Knowledge page. Documents are:

  1. Chunked into segments
  2. Embedded using vector embeddings (pgvector)
  3. Retrieved automatically when relevant to a conversation

Agent Assignment

By default, all documents are available to all agents. You can restrict access from the agent edit page by checking/unchecking specific documents.