Letta
by Letta
Open-source platform for stateful AI agents with persistent memory, now with a TypeScript Agents SDK for embedding them in your own apps
Letta is an open-source platform and cloud service for building stateful AI agents that keep long-term memory, identity and learned skills across sessions and machines. It is aimed at engineering teams building personalised or long-running agents, and removes the need to hand-roll memory, context management and state persistence on top of a stateless LLM API.
Letta is a San Francisco company spun out of UC Berkeley's Sky Lab that builds stateful AI agents: agents with persistent memory, identity and the ability to learn from their own work across sessions. Its founders, Charles Packer (CEO) and Sarah Wooders (CTO), created MemGPT, the research project that introduced self-editing memory for LLMs, and launched Letta in September 2024 with a $10M seed round led by Felicis, with angels including Jeff Dean and Clem Delangue. The product has shifted since. In March 2026 Letta announced a new phase centred on Letta Code, an open-source (Apache-2.0), model-agnostic agent harness with persistent memory, skills, subagents and computer use, and deprecated earlier server-side features such as templates, the Letta Filesystem, server-side MCP integrations and tool rules. Letta Code runs as a CLI, as desktop apps for macOS, Windows and Linux, in the browser, and inside Slack, Telegram, Discord and GitHub Actions, and works with OpenAI, Anthropic and other providers through bring-your-own keys. On August 17, 2026 the company released the Letta Agents SDK, a TypeScript library for embedding long-running stateful agents into custom web, mobile and desktop applications. Agents can live on Letta Cloud, on a self-hosted Letta App Server or on a local backend, and keep the same memory while moving between registered laptops, VMs and containers. Letta lists Bilt, 11x, Kognitos and Hunt Club as customers. It competes with memory layers such as Mem0 and agent frameworks such as LangGraph, and differentiates on transparent, white-box memory that developers can inspect and edit directly.
Engineering leads building personalised or long-running agents who need memory and state that persist across sessions and machines without building that layer themselves.
Agents that retain and improve their memory over time, portable across Letta Cloud, a self-hosted server or local machines.
At a Glance
- Category
- Agent Development
- Pricing
- Freemium, Subscription, Usage-based
- Target Market
- CTOs, Heads of AI, Enterprise Developers, AI Engineers
- Deployment
- Open-source, Self-hosted, Cloud-first
- Founded
- 2024
- Headquarters
- San Francisco, USA
Key Features
- ✓Persistent white-box memory
Agents rewrite their own context and memory over time, and developers can inspect every raw LLM request and edit memory directly.
- ✓Letta Agents SDK (TypeScript)
Released August 2026, it embeds long-running stateful agents into custom web, mobile and desktop apps over the same WebSocket layer as Letta Desktop.
- ✓Portable agent state
The same agent keeps its memory and identity across cloud, self-hosted and local backends and across registered laptops, VMs and containers.
- ✓Letta Code harness
Open-source, model-agnostic coding agent with skills, subagents and computer use, which Letta reports ranked first among model-agnostic open-source agents on Terminal-Bench.
- ✓Multi-channel deployment
Agents run from the CLI, desktop apps, browser, Slack, Telegram, Discord and GitHub Actions, so one agent can serve several surfaces.
- ✓Bring-your-own model keys
Every plan accepts your own OpenAI, Anthropic or other provider API keys, avoiding lock-in to a single model vendor.
Capabilities
Use Cases
- •Personalised customer-facing assistants
Build assistants that remember each user's history and preferences across sessions instead of starting cold in every conversation.
- •Long-running coding agents
Run Letta Code agents that accumulate knowledge of a codebase and team conventions, reducing the context re-supplied in every session.
- •Embedding agents in products
Use the TypeScript Agents SDK to add a stateful agent to a web, mobile or desktop app while keeping its state on Letta Cloud or a self-hosted server.
- •Shared team agents in chat
Deploy an agent into Slack or Discord that retains organisational context and can be shared with access control on Teams Pro.
- •Private self-hosted agents
Run the Letta App Server in your own environment when agent memory contains proprietary data that should not sit in a vendor cloud.
Ideal For
Best For
- ✓Product teams building personalised assistants that must remember users across sessions
- ✓Developers who want inspectable, editable agent memory rather than a black-box memory feature
- ✓Teams needing agents that move between cloud, self-hosted and local machines without losing state
- ✓Engineering orgs wanting a model-agnostic, open-source coding agent with persistent memory
Not Ideal For
- ✗Enterprises that require published SOC 2, HIPAA or similar attestations up front: none are listed on Letta's website or pricing page
- ✗Teams that need a stable, long-lived server-side API surface, given the March 2026 pivot deprecated templates, the Letta Filesystem, server-side MCP and tool rules on short timelines
- ✗Short, stateless tasks where a maintained project instructions file already supplies enough context; Hacker News commenters argue memory adds maintenance burden and context-poisoning risk there
Integrations
Deployment
Market Analysis
Pros
- ✓Apache-2.0 open source with a self-hostable App Server
- ✓Research pedigree: the founders created MemGPT at UC Berkeley's Sky Lab
- ✓Transparent white-box memory, praised on Hacker News for debuggability
- ✓Low entry cost: free tier and $20/month plans, with bring-your-own API keys on all tiers
Cons
- ✗Frequent strategic change: the March 2026 'next phase' deprecated several server-side features, creating migration work for earlier adopters
- ✗No compliance certifications are published on the website or pricing page
- ✗The Agents SDK is TypeScript-only at launch and drew little community discussion (7 points on Hacker News at launch)
- ✗Hacker News users describe persistent memory as a maintenance burden with context-poisoning risk compared with curated project docs
- ✗Seed-stage vendor with $10M raised in 2024, a consideration for long-term enterprise dependency
Pricing
Free
$0
- ✓Limited agents
- ✓Limited Letta Auto usage
- ✓Bring your own API keys
Pro
From $20/mo
- ✓Up to 20 stateful agents
- ✓Letta Auto weekly and monthly quota
- ✓Pay-as-you-go overage
API Plan
From $20/mo
- ✓Unlimited agents
- ✓$0.10 per active agent per month
- ✓$0.00015 per second of tool execution
- ✓Pay-as-you-go LLM usage
Teams Pro
From $20/mo per seat
- ✓Add teammates to an organization
- ✓Share agents and set access control
- ✓Letta Auto quota
Enterprise
Contact for pricing
- ✓Volume-based pricing
- ✓Role-based access control
- ✓SAML/OIDC SSO
- ✓Dedicated support
Letta publishes list prices: a free tier with limited agents, Pro at $20/month for up to 20 stateful agents, and an API plan at $20/month plus $0.10 per active agent per month and $0.00015 per second of tool execution, with LLM tokens billed pay-as-you-go or through your own provider keys. Teams Pro is $20 per seat per month; RBAC, SAML/OIDC SSO, higher quotas and dedicated support require custom Enterprise pricing. The open-source code is free to self-host.
Security & Compliance
Connect
Sources
This page was written from 9 sources, 7 on domains other than letta.com.
- 1.letta.com — introducing the letta agent sdkvendor
- 2.letta.com — our next phasevendor
- 3.docs.letta.com — pricing
- 4.docs.letta.com — docs.letta.com
- 5.github.com — letta
- 6.github.com — letta code
- 7.prnewswire.com — berkeley ai research lab spinout letta raises 10m seed finan
- 8.felicis.com — letta
- 9.news.ycombinator.com — item
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Kiro Crew
Open-source workspace for coding agents that keep working after you close the laptop
Naïve
Autonomous company infrastructure — compute, payments, identity and incorporation for AI agents behind one API
Microsoft Agent Lightning
MIT-licensed framework that trains any AI agent with reinforcement learning without changing its code
Keenable
Independent web search API for AI agents — 100B+ document index, sub-250ms p95