Warp Agent Memory (Research Preview)

Agent Memory is a persistent memory system that allows agents within the Warp platform to retain information across conversations. It is shared across all supported harnesses, including Warp Agent, Claude Code, and Codex, and is accessible from anywhere, not limited to specific conversations or machines.

Memory creation and retrieval run asynchronously in the background, ensuring no increase in token consumption or latency for active tasks. At the end of a conversation, Warp extracts permanent facts, learnings, and outcomes to store as memory, merging or replacing existing information in case of conflicts.

Key features include:

• Cross-harness memory: A single memory system shared across various agents, including Warp Agent, Claude Code, and Codex.
• Local and cloud agent support: Supports both interactive local agents within Warp and background cloud agents.
• Shareable stores: Memories are organized into stores, allowing knowledge sharing by linking them to specific agents or entire teams.
• Automatic memory generation: New agents receive a dedicated memory store by default, building long-term memory from their first run.
• Traceability and auditability: Records the source of each memory and tracks change history to ensure transparency.

Agent Memory runs on Warp infrastructure and is managed as personal, agent-specific, or team stores, depending on the owner (user, agent, or team).