🧠 Tool: @remember

The @remember tool allows the AI agent to persist critical project context into a long-term SQLite database, ensuring important decisions or patterns are not lost between sessions.

Technical Implementation

When the agent invokes @remember, the system executes an INSERT OR REPLACE SQL statement against the local memories table (located in ~/.local/share/zed/).

The implementation in crates/agent/src/tools/memory_tools.rs enforces categorization into one of five schemas:

1. Architecture: structural rules.
2. Patterns: code styles.
3. Issues: bug resolutions.
4. Procedures: build/deploy steps.
5. Notes: general facts.

Usage Examples

Example 1: Preserving Architectural Decisions

User

Note that this project uses Rust 1.76 specifically for this workspace. Never use 1.78+ features.

Zed Agent

Persisting tool-chain constraint to memory...

Tool Output

@remember({ category: "Procedures", content: "Workspace strictly uses Rust 1.76. Do not use 1.78+ features." })

Example 2: Documenting Legacy Debt

User

The `legacy_token` module in `crates/auth` is deprecated. Always use `SessionManager` instead.

Zed Agent

Recording deprecation policy...

Tool Output

@remember({ category: "Architecture", content: "`legacy_token` is deprecated. Use `SessionManager` for all new auth code." })

Workflow Impact

• Institutional Memory: The agent remembers why you chose a specific pattern, even months after the conversation.
• Rule Enforcement: Injects these memories directly into the agent's subconscious prompt for every future turn.
• Onboarding Speed: Quickly catches the agent up on obscure project constraints that aren't in the global LLM training data.
• Scalable Context: Unlike .rules files, memories are retrieved contextually, saving tokens while maintaining precision.