Letta builds stateful agents that remember across sessions, with tools, scheduled runs and self-hosting or cloud options. Pick it when an agent genuinely needs to remember things over time, not just within one conversation. The free plan gives limited agents with your own API keys; Pro is $20/month for up to 20 agents plus a usage quota.
Other good fits
Teams that want scheduled or tool-using agents with access controls, not a single stateless chat call.
+ Pros
Letta builds stateful agents: agents that keep memory, tools, permissions, schedules and even subagents persisting across sessions rather than starting fresh every conversation. It offers cloud-hosted agents, a CLI, and a self-hosting path, so you can pick how much infrastructure you want to own.
Pick it when…
You need an agent that remembers something specific about a user or project across separate conversations.
You want scheduled or tool-using agent behaviour, not just a one-off chat response.
You want the option to self-host rather than depend entirely on a hosted platform.
− Cons
Read cons
State is the point, but it's also the risk: an agent that remembers things and has tool access can produce side effects a stateless call never would, and its behaviour over time is harder to reason about than a single call-and-response. The documentation doesn't hand you a safe default for memory retention, tool permissions or production security for your specific project; decide those explicitly, and don't connect real credentials or personal data until you have.
Pick something else when…
You just need a stateless feature and can manage state in your own application. A direct model SDK call is simpler and easier to audit.
You want application-level composition and a wide integration library rather than a full agent platform.LangChain fits that better.
The actual need is a memory service to plug into an agent you're already building, not a whole platform.Mem0 is scoped to just that.
Avoid if
You mainly need a memory layer to bolt onto an existing agent, rather than a full agent platform.
Workflow
Define a narrow agent role → choose its model and what it's allowed to remember → give it only the tools or skills it actually needs → run a test conversation → check what state and actions actually persisted afterward. You still decide permissions, secrets handling and what data is allowed to be retained.
Limited agents, limited Letta Auto usage, bring your own API keys
Pro
$20/month
Up to 20 stateful agents, weekly/monthly Auto quota, pay-as-you-go overage
Teams Pro
$20/seat/month
Adds teammates, org sharing with access controls
API
$20/month base
Unlimited agents, $0.10 per active agent/month, tool execution at $0.00015/second
Enterprise
Custom
Volume rates, RBAC, SAML/OIDC SSO, dedicated support
Model and infrastructure costs are separate from the platform price. Running effort is higher than a stateless call: memory, retained data and scheduled or tool-triggered actions all need review, since state makes behaviour harder to fully predict than a one-shot request.
Restrictions
Acceptable memory retention, tool permissions, cloud configuration, export terms and production security posture aren't settled by the documentation; they're your decisions to make deliberately before connecting anything beyond a test setup.
First project
Documentation-based suggestion — not run here.
Build a campus-help agent with one remembered fact.
Create an agent that can remember only a learner's chosen topic of interest.
Give it one tool: looking up a link from a fixed, public list of courses.
Ask it a question that sets the topic, then start a new conversation and ask a follow-up that relies on it remembering that topic.
Check that it recalls the topic correctly and doesn't leak or invent anything about unrelated topics.
Done when a second, separate conversation correctly uses the remembered topic, and the agent's tool use stayed within the fixed course list you gave it.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on Letta's documentation (quickstart, stateful agents concept) and its pricing page, checked 23 September 2026. No agent was created and no conversation was run for this review.
Install Letta Code globally and run an autonomous agent.
Maintain agent memory and identity across sessions.
Version context repositories with Git-based memory management.
Letta, Inc. · free-tier · cloud · self-host
agent memory
stateful agents
memory blocks
continual learning
memgpt
write-code
open-source
View sourced imageCurrent Letta app conversation interface from its official documentation; the legacy ADE has been deprecated.
Letta official product documentation
Owner-requested editorial use of public official interface screenshots, 15 September 2026. Copyright retained by the original publisher; attribution preserved. No open licence or hands-on course execution is asserted. Source asset: https://docs.letta.com/images/desktop/chat-agent.png. PNG conversion preserves the complete frame. Converted on 15 October 2026 to a 1600px-wide WebP (quality 80) for faster page loads; the original image remains in the media library.