The archive · Developer & Business Tools · Strategic decision · 2023–2026
Letta's stateful-memory bet: MemGPT research becomes a memory-first coding agent
UC Berkeley's MemGPT paper became Letta Inc: $10M seed, tiered agent memory, then a memory-first coding agent with 23k+ stars by mid-2026.
Letta
What the business is
Letta (formerly MemGPT) is an open-source platform for stateful AI agents with tiered memory — core blocks always in context, archival vector storage, and full conversation recall — plus the Letta Code coding agent, a desktop app, messaging Channels (Slack, Telegram, Discord, WhatsApp, Signal) and a hosted cloud with free, $20/month Pro and enterprise tiers.
How it started
The MemGPT paper, posted to arXiv on 2023-10-12, proposed treating the LLM context window like RAM with an OS that pages data in and out; it went viral on Hacker News before the team formally released the code, drawing 11,000 stars in days. Packer and Wooders, PhD students in Berkeley's Sky Computing Lab, incorporated Letta in 2024 and came out of stealth on 2024-09-23 with the rebrand and a $10M seed led by Felicis, telling TechCrunch they were positioning as the open alternative to OpenAI.
What happened
Letta's tiered memory lets agents edit their own core memory, store observations in archival memory and search full conversation history, with sleep-time compute that reorganizes memory between interactions. Through 2026 the company consolidated around Letta Code: launched December 2025 with 42.5% on Terminal-Bench, it gained a desktop app in April 2026, Channels replaced the standalone LettaBot in May 2026, and pricing was rebuilt around the coding agent (free tier up to 3 managed agents, Pro at $20/month).
How it ended up
Still operating as of 2026-09-06: the open-source core is Apache-2.0 and self-hostable, the cloud sells free, Pro and enterprise tiers, and co-founder Charles Packer told Tessl the company is deliberately not innovating at the UI layer — the bet is that context engineering becomes the moat; the reviewed sources disclose no revenue figures.
Background
Letta began as MemGPT, a UC Berkeley research project whose October 2023 paper proposed giving LLMs an operating system for memory — paging information between a finite context window and external storage — and went viral on Hacker News before the code was even formally released. The founders incorporated in 2024, rebranded to Letta, and raised a $10M seed led by Felicis on the day they came out of stealth, positioning as the open alternative to OpenAI.
The platform's core is tiered memory: core blocks always in the context window, archival vector storage for long-term facts, and full conversation recall, with the agent managing all three through tool calls. Agents edit their own memory as they learn, and sleep-time compute reorganizes it between interactions. The Apache-2.0 core is self-hostable, while Letta Cloud sells managed agents with an Agent Development Environment.
By 2026 the company's center of gravity moved to Letta Code, a memory-first coding agent that scored 42.5% on Terminal-Bench as the #1 model-agnostic open-source agent. A desktop app shipped in April 2026, Channels folded the old LettaBot into the product in May, and pricing was rebuilt around the coding agent. Ry Walker Research records 23.3k+ GitHub stars as of June 2026 and frames the open question as whether memory stays a durable differentiator before model providers absorb it.
What has to be true
- Stateless LLMs cap what agents can do, and the MemGPT paper gave the problem a mechanism — virtual context management — that others could build on.
- A coding agent is the sharpest proof of memory: a terminal agent that remembers project context across sessions has an immediately testable edge.
- Model-agnostic, self-hostable design removes the vendor-lock objection that enterprise buyers raise against agent platforms.
- Keeping the UI deliberately familiar lets Letta compete on the memory layer without re-educating users.
What can be applied
A viral research idea earns stars, not a business — the durable bet is that memory becomes a layer agents cannot buy elsewhere.
Aftermath
As of 2026-09-06 Letta is active and funded, converging on Letta Code: 23.3k+ GitHub stars, a desktop app, Channels and pricing from free to $20/month Pro. Documented traction includes the December 2025 HN launch (83 points), the 42.5% Terminal-Bench score, and MemGPT's original 11,000-stars-in-days virality; no revenue figures appear in the reviewed sources. Analysts flag the risk that frontier providers ship native memory and absorb the differentiator — Letta's answer is that context engineering is a layer worth owning.
Sources
- Letta | Ry Walker Research (23.3k+ GitHub stars, June 2026)
- Letta (MemGPT) | AI Wiki — history, architecture and funding
- letta-ai/letta — Platform for stateful agents (landing repository snapshot)
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card