Letta
San Francisco, US Β· 11 known investors
Letta is an AI research lab building self-improving AI agents that manage memory through programmatic context management and continuous learning. The company develops foundational technology for digital agents that remember, learn, and adapt over time.
Also known as letta-ai Β· MemGPT
Founders & leadership
Board
Investors Β· 11
Also in the syndicate Β· 8
Valuation Β· disclosed
Disclosed eventsSource: SEC prospectus filings, and round valuations the company or its investors disclosed β follow each entry's link for the claim.
Company profile
researched Aug 2026Letta is an AI research lab based in San Francisco that describes itself as building "machines that learn": experiential agents that retain memory, learn continuously and improve over time. The company is the commercial entity behind the MemGPT open source project, which originated as research at UC Berkeley's Sky Computing Lab and introduced the idea of self-editing memory for large language models β treating the LLM context window like virtual memory managed by an "LLM operating system." In September 2024 the founders clarified naming: MemGPT refers to the original agent design pattern from the research paper, while Letta names the agent framework and the company.
Letta's software ships as an open source framework and a set of commercial offerings. At launch these included an Agent Development Environment (ADE) for viewing and editing an agent's prompts and memory, an API platform, and Letta Cloud, a hosted, model-agnostic service for deploying and running stateful agents accessible via REST APIs. More recently the product is distributed as the Letta Agent via the npm package @letta-ai/letta-code, with an interactive terminal UI, a self-hostable App Server, desktop apps for macOS, Windows and Linux, a browser client at chat.letta.com, channel integrations (Slack, Telegram, Discord and custom channels), a TypeScript Agent SDK, and Letta Cloud for persisting agent memory and identity across devices.
A third-party comparison describes Letta as a full agent runtime rather than a pluggable memory layer, with memory organized into tiers β core memory held in the context window, recall memory for searchable conversation history, and archival memory queried via tool calls β where the agent itself decides what to store and retrieve through tool calls. The same source notes Letta is Apache-2.0 licensed, with compliance available via self-hosting, and that it had not published LongMemEval benchmark results.
Founding story
Co-founders Charles Packer (CEO) and Sarah Wooders (CTO) met during PhD research at UC Berkeley's Sky Lab under advisors Joseph Gonzalez and Ion Stoica, both of whom joined Letta's founding team in an advisory capacity. Their MemGPT paper on self-editing LLM memory went viral on Hacker News in October 2023, drawing investor interest β Felicis partner Astasia Myers reached out after reading the paper β and Stoica brokered introductions to angel investors.
Business model
Open source core (Apache-2.0 licensed framework, self-hostable App Server) combined with commercial hosted offerings, principally Letta Cloud, plus developer tooling such as the Agent Development Environment for debugging and monitoring agents.
Paid hosted service (Letta Cloud) alongside free open source software; a third-party comparison lists managed pricing in the range of $20β200 per month, with self-hosting free under Apache 2.0.
Traction
The MemGPT paper drove viral open source adoption, reaching roughly 11,000 GitHub stars and 1.2K forks shortly after release; the letta-ai/letta repository lists 24.3K stars, 2.6K forks and 7,471 commits. The company publishes case studies with Bilt, 11x, Kognitos and Hunt Club, and distributes software via npm, desktop apps, a web client and Letta Cloud.
Latest developments
The company now presents itself as an AI research lab shipping research as software in the form of the Letta Agent, installed via npm as @letta-ai/letta-code, with desktop, web, and messaging-channel clients and a TypeScript Agent SDK. The original letta GitHub repository has been reduced to a landing page with the V1 server archived, and development consolidated in letta-ai/letta-code. Recent published research covers sleep-time compute, continual learning in token space, git-based context repositories, a context constitution, and memory models trained with memory-native reinforcement learning.
βΈFull profile β market position, technology, go-to-market, geography, history, risks & controversies
Market position
Positioned as the commercial entity behind the widely adopted MemGPT open source project and as an open alternative to closed agent platforms; the co-founder framed the company as "the open alternative to OpenAI." A third-party comparison positions Letta as a full agent runtime versus memory-layer products such as Mem0. Coverage in TechCrunch, WIRED and Fast Company is cited by the company.
Technology
Programmatic context management for LLMs based on the MemGPT "LLM operating system" pattern, in which agents self-edit tiered memory (core memory in the context window, recall memory for conversation history, archival memory in long-term storage) through tool calls. The platform manages the agent loop, tool execution and state persistence, and is model agnostic, allowing agents to run across different LLM providers. Letta emphasizes "white-box memory," exposing the exact prompts and memories passed to the model at each reasoning step. Later research directions include sleep-time compute (offline reasoning about context during idle time), continual learning in token space, git-based versioned context repositories, and memory models trained with memory-native reinforcement learning.
Go-to-market
Developer-led distribution through open source software and package managers (npm, previously PyPI and Docker), a free ADE and API platform at launch, beta onboarding for the hosted Letta Cloud service, documentation at docs.letta.com, and an open source/Discord community. Published customer case studies are used as references.
Developers and companies building AI agents and chatbots that require persistent memory across sessions; cited use cases include highly personalized conversational chatbots, customer support bots, and healthcare symptom-tracking applications. Featured case studies name Bilt, 11x, Kognitos and Hunt Club.
Geography
Headquartered in San Francisco; the founding research originated at UC Berkeley's Sky Computing Lab.
History
The work began as the MemGPT research project at UC Berkeley, inspired by an effort to improve the memory systems of Discord chatbots. The whitepaper was posted on October 12, 2023 and spread on Hacker News before the planned code release, generating rapid open source adoption. In September 2024 the founders incorporated and launched Letta, emerging from stealth on September 23, 2024 with a $10M seed round, and moved the PyPI package to "letta" and the Docker image to "letta/letta-server." The original letta-ai/letta repository has since been converted into a landing page, with the retired V1 server preserved on an archive branch and active development moved to letta-ai/letta-code. The company's published research timeline runs from MemGPT (Oct 2023) through Sleep-time Compute (Apr 2025), Continual Learning in Token Space (Dec 2025), Context Repositories (Feb 2026), Context Constitution (Apr 2026) and Memory Models (Jun 2026).
Risks & controversies
Third-party analysis notes Letta had not published LongMemEval benchmark results, making direct quantitative comparison with competitors difficult, and that its agentic self-editing memory can be less predictable and more token-costly than passive extraction, with memory quality dependent on the underlying model's judgment; adopting Letta also means adopting a full agent platform rather than a drop-in library. TechCrunch coverage flagged competition from LangChain and model providers' own agent tooling, and the possibility that multistep reasoning models reduce the need for external state management. The GitHub archive branch is explicitly unsupported and not recommended for production.
Compiled by commissioned research from 8 cited public sources β announcements, filings, and press listed under research sources below.
Key figures
latest reportedCompany-reported or press-reported figures, each dated to when it was claimed β not independently audited.
Timeline Β· 12
launches, deals, and filingsResearch on powering agent learning through memory models trained with memory-native reinforcement learning.
A set of principles governing how AI agents manage context to learn from experience.
A rebuild of agent memory using programmatic context management and git-based versioning.
The letta-ai/letta GitHub repository became a landing page; the retired Letta V1 API server source was preserved on an 'archive' branch and is unsupported, with active development moved to letta-ai/letta-code.
Research on letting agents reason about context during idle time rather than at inference.
Seed financing led by Felicis (Astasia Myers) with participation from Sunflower Capital, Essence VC and angel investors, at a $70 million post-money valuation.
$10M source β
The company released its Agent Development Environment (ADE) and API platform for building and deploying AI agents for free, and began onboarding early developers to the beta of the hosted Letta Cloud platform.
Letta, the commercial entity of the MemGPT open source project spun out of UC Berkeley's Sky Computing Lab, came out of stealth and began accepting beta requests for Letta Cloud.
The company announced that 'MemGPT' would refer to the original agent design pattern from the research paper while 'Letta' would name the agent framework and company; the PyPI package moved to 'letta' and the Docker image to 'letta/letta-server'.
Charles Packer and Sarah Wooders posted the MemGPT whitepaper on October 12, 2023; it was posted to Hacker News by a third party before the planned code release and stayed at the top of the site for 48 hours, followed by rapid GitHub growth (11,000 stars, 1.2K forks).
Original work on virtual context management, described by the company as where Letta began.
Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.
βΈResearch sources Β· 8
primary sources listed
- Lettamemgpt.ai Β· web
8 public sources were cited for this profile; the first-party ones are listed here.
Frequently asked questions
- What does Letta do?
- San Francisco AI research lab building stateful, self-improving agents with programmatic memory, originating from the MemGPT project.
- Who are Letta's investors?
- Letta's investors include ESSENCE VC, Felicis Ventures, Angel Collective Opportunity Fund.
- Where is Letta headquartered?
- Letta is headquartered in San Francisco, US.