Letta · Rate Limits

Letta Ai Rate Limits

Letta does not publish fixed numeric per-minute or per-second request rate limits for the Letta Cloud REST API. Cloud usage is governed instead by plan-level constraints - the number of agents allowed on Free and Pro, and metered active-agent and tool-execution usage on the API plan - plus whatever throughput/rate limits the underlying LLM model provider (OpenAI, Anthropic, Google, etc.) enforces on the model calls an agent makes. Each agent also processes messages sequentially, so concurrent requests to the same agent are explicitly discouraged rather than rate-limited by the server. Self-hosted deployments are bound only by their own infrastructure and by the rate limits of whichever model providers they configure.

Letta Ai Rate Limits is the machine-readable rate-limit profile for Letta on the APIs.io network, conforming to the API Commons Rate Limits specification.

It captures 5 rate-limit definitions, measuring requests, agents, and requests/tokens.

The profile also includes 2 backoff/retry policies defined and response codes documented for throttled.

Tagged areas include AI, Agents, LLM, Rate Limiting, and Quotas.

5 Limits Throttle: 429
AIAgentsLLMRate LimitingQuotas

Limits

Cloud API Requests account
requests
not published
No fixed numeric request-rate limit is documented for the Letta Cloud REST API.
Concurrent Messages Per Agent agent
requests
1
Letta's own API docs warn that each agent processes messages sequentially and that sending concurrent requests to the same agent can produce undefined behavior; callers should wait for one request to complete (or use separate agents) rather than parallelize against a single agent.
Free / Pro Agent Count account
agents
per plan (Free is limited; Pro is roughly 20)
Personal plans cap the number of concurrently active stateful agents rather than request rate.
Underlying Model Provider Limits model provider
requests/tokens
set by the configured LLM/embedding provider
Each agent's message and embedding calls are ultimately bound by the rate limits of its configured model provider (OpenAI, Anthropic, Google, self-hosted, etc.).
Self-Hosted Throughput deployment
requests
hardware-bound
Self-hosted Letta servers are limited only by their own infrastructure and by their configured model providers.

Policies

Sequential Agent Processing
Concurrent requests to the same agent are not supported; each agent processes one message at a time and later requests should wait for the prior one to complete.
Backoff Strategy
Clients should implement exponential backoff with jitter and honor Retry-After on 429 responses returned by Letta or by an upstream model provider.

Sources