Letta Code

Build digital people whose memory, identity, and capabilities improve with experience.

PERSISTENT AGENTS }}}

Custom agents that live forever

Build your own deeply personalized agents, each with their own unique identity and expertise. Supports custom skills, harness modifications, and tools.

CONTINUAL LEARNING }}}

Best-in-class memory & learning

Agents self-improve through sleep-time compute, learning without needing model updates. All context (skills, prompts, etc.) is git-tracked through MemFS.

CHANNELS }}}

Works where you work

Communicate with agents through Slack, Discord, Telegram, or custom channels. Collaboratively teach and direct agents where your team already works.

RUN ON ANY DEVICE }}}

Run on remote devices

Connect your own device, sandbox, or VM as a remote environment. Agents can work across multiple devices simultaneously.

SCHEDULED TASKS }}}

Proactive and understands time

Ask agents to work proactively, monitor results, or run tasks on a schedule. Create schedules yourself or let agents set their own, and review past runs anytime.

MULTI-MODEL MEMORY }}}

Bring your own model and subscription plans

Works with OpenAI Codex, Z.AI, and other subscription plans, or with your own API keys. Agents transfer across models without memory loss.

AI Research

Building machines that learn

Letta is an AI lab building machines that learn: persistent agents with the ability to continuously learn and adapt from their own experience. Born from MemGPT at UC Berkeley's Sky Computing Lab, and backed by leaders in AI research and infrastructure.

Recent work

Continual Learning in Token Space } Sleep-time Compute: Scaling Inference-Time Compute for LLM Agents }

Explore our research

Research blog }