letta-ai/letta-code is a stateful agent harness written in TypeScript that gives agents human-like memory and identity so they can change themselves. The latest release is v0.34.18.
The README describes the project as a harness for building agents that are closer to people than to tools. A harness is the frame that wraps and drives an agent. The agent uses memory, identity, and experience accumulated over time to directly modify its own memory blocks, skills, and system prompt, learning and evolving over the long term. That change history is managed with MemFS. It is released under the Apache-2.0 license, the repository was created on October 25, 2025, and it has 3,663 GitHub stars. Conversations can be moved across channels such as the CLI, browser and mobile, Slack, Telegram, and Discord, and it can also serve as an always-on agent that works proactively.
That said, there are no metrics or benchmarks showing how much memory actually accumulates and gets reused. The momentum score on a 10-point scale tops out at 4.2. A single score cannot stand in for performance. Since no changelog is given for v0.34.18, you have to check the release notes directly to see what this release changed. The structure and usage paths are confirmed, but performance and real-world value remain unproven.