What Is Workspace Memory?
Workspace Memory gives the AI persistent context that carries across every conversation in the workspace. Instead of starting from scratch each time, the AI remembers key facts, preferences, and insights that your team has established. There are two types of memory:Core Memory
Facts and instructions that you explicitly tell the AI to remember. Core Memory entries are always included in the AI’s context.
Memory Crystals
Patterns and insights the AI detects automatically from your conversations. Crystals are surfaced for your review before becoming active.
Core Memory
Core Memory is the workspace’s explicit knowledge base — facts, rules, and context that the AI should always keep in mind. Think of it as a persistent briefing document that the AI reads before every response.Common Uses
- Project context: “We are building a mobile app for inventory management. The target launch date is Q3 2026.”
- Terminology: “When we say ‘the platform’, we mean our internal CRM system called Apex.”
- Preferences: “Always format financial figures with two decimal places and USD currency symbols.”
- Constraints: “Never suggest solutions that require AWS. We are a GCP-only organization.”
Managing Core Memory
Open the Memory tab
Navigate to your workspace and click the Memory tab. The Core Memory section is displayed at the top.
Add or edit entries
Click Edit Core Memory to open the editor. Write or update the information you want the AI to retain. Keep entries concise and factual — the AI performs best with clear, specific context.
Memory Crystals
Memory Crystals are insights that the AI detects automatically by analyzing conversations in the workspace. When the AI notices a recurring pattern, an important decision, or a frequently referenced fact, it creates a Memory Crystal candidate for your review.How Crystals Are Detected
After each conversation, a background process analyzes the exchange for notable information. This might include:- A decision the team made (“We agreed to use PostgreSQL instead of MongoDB”)
- A recurring question or topic that suggests the team relies on certain knowledge
- A correction or clarification that indicates an important distinction
Memory Crystals are generated by an automated detection process. They are never activated without human review. The AI suggests; you decide.
Reviewing Crystals
- Approving a Crystal
- Rejecting a Crystal
- Open the Memory tab in your workspace.
- Scroll to the Memory Crystals section. Pending crystals are shown with a yellow indicator.
- Read the crystal’s content. If it is accurate and useful, click Approve.
- The crystal becomes active and the AI will reference it in future conversations, just like Core Memory.
Crystal Lifecycle
| Status | Meaning |
|---|---|
| Pending | Detected by the AI, awaiting human review. Not yet active. |
| Approved | Reviewed and accepted. Included in the AI’s context for all workspace conversations. |
| Rejected | Reviewed and discarded. Will not be used by the AI. |
Memory Stats
The Memory tab displays a summary of your workspace’s memory usage:- Core Memory size — How much text is stored in Core Memory. Keep this focused; overly large Core Memory can dilute the AI’s attention.
- Total Crystals — The number of Memory Crystals detected, broken down by status (pending, approved, rejected).
- Last updated — When the memory was last modified, so you can tell if it is current.
Back to Overview
Return to the Workspaces Overview to explore other workspace features.