Reviewed by Harsh Desai · Last reviewed:
Letta
Machines that learn
Best for
Letta is an open-source AI agent framework from the Letta team that solves the single biggest weakness of standard LLM agents: forgetting everything between sessions. By providing a tiered memory architecture, a hosted control plane, and a fully open-source codebase under Apache 2.0, Letta lets builders ship stateful agents without rebuilding memory infrastructure from scratch.
What Letta does:
- •Persistent long-term memory agents store information in a tiered memory architecture (core, recall, archival) that mimics human recall, retaining context and personal facts across thousands of sessions.
- •Visual Agent Development Environment the ADE lets builders inspect, debug, version, and edit agent personas through a visual interface rather than reading raw JSON state.
- •Open-source under Apache 2.0 the full server, SDKs, and CLI are open-source on GitHub, so teams can self-host the entire stack with no vendor lock-in.
- •MCP support native Model Context Protocol integration lets Letta agents call tools across any MCP-compliant server, simplifying connections to file systems, databases, and external APIs.
- •Multi-step task orchestration agents autonomously manage background memory updates, re-summarisation, and multi-step workflows without manual prompting.
- •Model-agnostic API swap between OpenAI, Anthropic, local Ollama, or any OpenAI-compatible endpoint with a single config change while keeping memory and persona intact.
- •REST API and CLI production-grade REST API with Python and TypeScript SDKs, plus a CLI for local development, scripting, and CI workflows.
- •Sub-agent and multi-agent architectures compose hierarchical agent systems where specialised sub-agents share a common memory store for collaborative reasoning.
Pricing:
- •Free $0/month: up to 3 agents on Letta Cloud for personal exploration. Self-hosted deployments are completely free under Apache 2.0.
- •Pro $20/month: up to 20 agents, increased memory capacity, and standard support.
- •Max $200/month: up to 50+ agents, priority queues, increased quotas, and dedicated support.
- •API plan $20/month base fee plus $0.10 per active agent per month and $0.00015 per second of tool execution. You bring your own LLM provider and pay those costs separately.
Limitations:
- •Steep learning curve for non-technical users Letta requires familiarity with API keys, agent configuration, and some Python or TypeScript. Vibe Builders comfortable with code will adapt; pure no-code users will struggle.
- •Self-hosting demands real engineering the open-source stack runs Postgres plus pgvector and a stateful server. Self-hosting at scale requires database tuning, backups, and monitoring discipline that most small teams underestimate.
- •Cost complexity for high-volume API usage the API plan stacks a base fee, a per-agent fee, and a per-second tool execution fee on top of your underlying LLM costs. Bills can climb fast without careful instrumentation.
- •Hosted Cloud regional coverage is limited Letta Cloud runs in a small number of regions, so latency-sensitive deployments outside North America and Europe may need self-hosted clusters closer to users.
Our Verdict
Letta scores 9.2/10 because it is the most credible open-source answer to the stateful-agent problem in 2026. The combination of a tiered memory architecture, native MCP support, an open-source Apache 2.0 codebase, and a polished visual ADE is rare. Most agent frameworks give you an in-memory dictionary called memory; Letta actually models recall the way humans do.
For the Vibe Builder, Letta is the practical bridge between a basic chatbot and a real long-running collaborator. The visual Agent Development Environment lets you debug, version, and tune agent personas without reading raw JSON state, so you can ship a stateful assistant for a small business or internal team without first hiring a backend engineer.
For the Developer, Letta gives unparalleled granular control over agentic state. The REST API, Python and TypeScript SDKs, and CLI make it natural to slot into existing CI and orchestration workflows. Native MCP support means tools and integrations port across the wider AI ecosystem rather than locking you to one vendor's plugin format.
Skip it if you only need a simple stateless chatbot (consider OpenAI Assistants API for a managed thread model) or a lightweight workflow runner without persistent agent memory (try n8n or Lindy for visual automations where memory is optional). Letta shines when memory is the point; for anything else, the operational cost is hard to justify.
Related Tools
View allCompare Letta With
Also Useful For
Frequently Asked Questions
How much does Letta cost in 2026?
Letta is free to self-host under Apache 2.0. On Letta Cloud, the free tier includes up to 3 agents, the Pro plan is $20/month for up to 20 agents, and the Max plan is $200/month for high-volume deployments. A separate API plan adds usage-based fees on top of your own LLM provider costs.
Letta vs LangChain: which should I pick?
Choose Letta when persistent memory across sessions is the core requirement and you want a managed memory architecture out of the box. Choose LangChain when you want maximum flexibility for chaining LLM calls and you are willing to build memory plumbing yourself. Letta is opinionated about state; LangChain is unopinionated about everything.
Can I self-host Letta?
Yes. The full Letta stack is open-source under Apache 2.0 on GitHub. You run a Postgres database with pgvector, the Letta server, and your preferred LLM provider. Self-hosting is free of Letta licence fees, but expect real engineering effort to manage the database, backups, and stateful server in production.
Does Letta support MCP and local models?
Yes to both. Letta has native Model Context Protocol support so agents can call tools across any MCP-compliant server. The framework is also model-agnostic and works with Ollama, vLLM, and any OpenAI-compatible endpoint, alongside frontier providers like OpenAI and Anthropic.
Is Letta suitable for non-technical users?
Letta is best for Vibe Builders and Developers who are comfortable with API keys and basic agent configuration. The visual Agent Development Environment lowers the barrier compared to raw code, but pure no-code users will still struggle. For a fully no-code stateful assistant, the Letta team recommends starting with their hosted Pro plan and the ADE.
What is Letta?
Letta is Machines that learn.
Is Letta free?
Yes, Letta offers a free version. Paid plans start at $20/month.
Who should use Letta?
Letta is built for vibe builders who want AI to handle the technical work and developers looking to accelerate their workflow. Common use cases include stateful-agent-development, long-term-memory, task-orchestration, codebase-context, multi-agent-systems.
What are the best alternatives to Letta?
Popular alternatives to Letta include Relevance Ai, Lindy, N8n. Compare features and pricing in our AI Agents directory to compare options.
Affiliate link: we may earn a commission. How this works.
Letta
Free tier available