Credits are a standard cost unit for resources in Letta, such as LLM inference and CPU cycles. When agents run on Letta, they make LLM model requests and execute tools. Model requests consume credits at a rate depending on the model tier (standard vs. premium) and whether Max Mode is enabled for longer context sizes. Tool executions that run in Letta are charged at a flat rate per second of execution. See details on credit pricing here.

Sandbox code execution and execution of custom tools run inside of Letta, so incur a credit cost for CPU time. Remote MCP tools are executed by the MCP tool provider, so do not have a credit cost. Letta built-in tools are executed for free.

Your Letta agents use large language models (LLMs) to reason and take actions. These model requests consume credits from your monthly balance (or additional purchased credits). Your balance of monthly credits refreshes every month.

Certain models have the ability to run with extended context windows. Turning on Max Mode extends the context window of the model driving your Letta agent beyond the 100k default, which may help when working with large files or codebases, but will increase cost (credit use) and latency.

The Letta API Platform is our fully-managed service for stateful agents, handling all agent infrastructure and state management to create scalable agent services. The Letta API Platform also has additional features beyond the open source such as durable execution for long-running agents, built-in sandboxing, agent templates, optimized vector search, message indexing, and observability.

Yes, the Letta API Platform supports agent file, which allows you to move your agents freely between self-hosted instances of the Letta open source and the Letta platform.

Yes, we offer discounts for qualifying startups. Submit a request using this form.

Reach out to support@letta.com, or join our active community on Discord to chat with the Letta developer team and other Letta users!