Skip to main content

​
Overview

Cofounder remembers what matters so each session can pick up where you left off. The memory system combines short‑term context with durable preferences and knowledge from your connected tools, so the agent can answer faster and act more reliably.

​
What Cofounder Remembers

  • Preferences: Communication style, business context, common settings, and scheduling preferences
  • Conversation summaries: Recent highlights plus a compact history to keep context across sessions
  • Knowledge from your tools: Emails, docs, issues, messages, and more from integrations you connect
Cofounder only accesses data from accounts you connect, using your existing permissions. You can review and revoke access at any time.

​
Memory Layers

Cofounder has three memory systems: Core Memory, Long‑Term Memory, and Working Memory.

​
Working Memory

Working Memory is a persistent workspace for each conversation. It keeps your active context organized while handling large or long‑running tasks gracefully.
  • Stores the active conversation and recent tool results
  • Saves large tool outputs as organized files and shows concise summaries in chat
  • Manages context limits by archiving older messages while keeping recent ones active
  • Maintains a scratchpad for notes, findings, and progress during complex work
  • Persists for the session with isolated access per conversation

​
Core Memory

Core Memory serves as short‑term memory across sessions. It periodically summarizes past interactions and keeps your preferences up to date.
  • Remembers conversation context over time without repeated reminders
  • Captures preferences like writing style, calendar details, business info, and preferred response formats
  • Maintains recent, compact summaries to keep context fresh
  • Updates automatically as you interact

​
Long‑Term Memory

Long‑Term Memory stores durable business context rather than day‑to‑day details.
  • Best for persistent knowledge: mission, roadmap, employees and roles
  • Not intended for temporary items like current tickets or one‑off discussions
  • Populated during onboarding when you connect data sources, with regular updates thereafter
  • Enables tasks like “create Linear issues from my roadmap” regardless of whether items live in Docs, Notion, or Linear
  • Consulted automatically when answering questions and executing tasks
Currently supported data sources include:
  • Gmail
  • Google Calendar
  • Google Docs, Sheets, Slides, Drive files
  • Linear (Issues, Projects)
  • Notion
  • Slack (Messages)

​
Finding and Using Memory

  • Natural‑language search across the knowledge stored from your tools
  • Hybrid ranking that considers relevance, recency, and frequency
  • Source transparency with links back to where information came from
  • Continuous access: Cofounder consults memory with each question and assigned task
  • Manual updates: Tell Cofounder what to remember or change, and it will update memory accordingly
Ask questions the way you would ask a teammate. For example: “Show projects involving Sarah” or “What changed before our last launch?”
If something should be remembered but does not appear in context, ask Cofounder to search again or remind it of the relevant topic.

​
Privacy and Control

  • Your permissions: Actions and data access use your connected accounts and scopes
  • Transparency: See which sources were used to answer questions or take actions
  • Management: View, edit, or remove stored memories and adjust preference settings

​
Best Practices

  • Connect only what you need so results stay targeted and relevant
  • Review memory periodically to keep preferences and context fresh
  • Start conservative with approvals, then automate once you trust the results
⌘I