# letta-ai/letta Is Now a Landing Page: Where Letta Code Actually Lives > The letta-ai/letta repository no longer holds the V1 server. It points to letta-code, and the README tells you to install from npm. Here is what that means before you adopt it. **letta-ai/letta** — Platform for stateful agents: AI with advanced memory that can learn and self-improve over time. - Repository: https://github.com/letta-ai/letta - Website: https://docs.letta.com/ - Stars: 24,776 · Forks: 2,624 - Language: Unknown - License: Apache-2.0 - Published: 2026-09-09 · Updated: 2026-09-09 · Language: en - Canonical page: https://hysenlabs.com/projects/letta-ai-letta ## What letta-ai/letta Is Now, and Why the Repository Looks Empty The README opens with a note that reframes the whole repository: it now serves as a landing page for the Letta project. The current source code lives in letta-ai/letta-code, which the README describes as including the agent harness, an interactive terminal UI, an App Server, channels, and the runtime used by the desktop and web apps. That is a real structural change, not a cosmetic one. Anyone who cloned letta-ai/letta expecting to read the server implementation will find documentation files at the top level (.github/, AGENTS.md, AI_POLICY.md, CITATION.cff, CONTRIBUTING.md, LICENSE, PRIVACY.md, README.md, SECURITY.md, TERMS.md) and no application code. The retired Letta V1 server source is preserved on the archive branch for historical reference. If your evaluation process starts by reading the code, you are reading the wrong tree. Start at letta-ai/letta-code instead. The project is still moving: the most recent push to this repository was on 2026-09-08, and the latest release listed here is 0.16.8 from 2026-05-14. Note the gap between release cadence and push activity, because it tells you which artifact is actually being iterated on. ## Stateful Agents and the Memory Problem Letta Targets The one-line description is the whole pitch: a platform for stateful agents, meaning AI with memory that can learn and self-improve over time. The problem is concrete. A stateless chat call forgets everything the moment the request returns, so any assistant that needs to remember a preference, a decision, or a half-finished task has to reconstruct that context on every turn. Letta's answer is to make the agent a long-lived object with memory attached, rather than a function you call. The audience follows from that: developers building agents that persist across sessions, and teams that want the same agent identity available in more than one place. The README lists Letta Cloud as the option for keeping agent memory, identity, and conversations available across computers, which is the clearest statement of who this is for. If your use case fits in a single prompt and a single response, the stateful layer is overhead you will pay for and never use. ## How the Pieces Fit: Harness, Terminal UI, App Server, Channels, SDK The architecture visible in the README is a set of entry points over one runtime. The runtime is the part used by the desktop and web apps. The agent harness and interactive terminal UI sit on top of it for local work. The App Server is the piece you run when you want local or self-hosted agents served rather than driven from a terminal. Channels are the adapters that put the same agent into Slack, Telegram, Discord, or a custom integration. The Letta Agent SDK is the path for building agents into TypeScript applications. Read that list as a deployment spectrum: terminal for a single developer, App Server for a shared or self-hosted instance, channels for reaching users where they already are, SDK for embedding the agent in your own product. The README does not document the wire protocol between these pieces, the storage backend behind memory, or how identity is scoped across channels. Those are the questions to answer from the letta-code README and docs.letta.com before you design around it. ## Installing Letta Code and Running Your First Agent The README gives one install path: npm, globally. The package name is @letta-ai/letta-code, so the command is: ```bash npm install -g @letta-ai/letta-code ``` After that, the README says to launch the interactive terminal UI by running the bare binary name: ```bash letta ``` You should land in the terminal UI, which is where you talk to an agent. For a local or self-hosted server instead of the terminal, the README gives a second command: ```bash letta server ``` That starts the App Server. The README does not list a port, a config file, or environment variables for it, so do not assume a default; check the letta-code README or docs.letta.com for the actual bind address and configuration keys. If you would rather not install anything, the README points to the desktop app for macOS, Windows, and Linux, to chat.letta.com in a browser including on mobile, and to Slack, Telegram, Discord, and custom channels. The TypeScript route goes through the Letta Agent SDK. Pick one entry point for your first run rather than setting up all of them. ## The Archive Branch Is Not a Fallback This is the limitation that matters most, and the README states it without hedging. The archive branch contains the retired Letta V1 API server as it existed when this repository was archived. Existing tags and releases remain available for reproducibility, which is a deliberate choice in favour of reproducible builds. Then the README is explicit: that source is unsupported, receives no fixes or security updates, and should not be used in production. Treat that as a hard boundary. If you have an existing integration against the V1 API, the tags will still resolve, but you are pinning to a tree nobody patches. The honest reading is that the archive branch is an escape hatch for reproducing old behaviour, not a supported deployment target. A second limitation is structural: because this repository is a landing page, its issue tracker and release list do not describe the code you will actually run. Bug reports and version questions belong with letta-ai/letta-code. Filing them here is likely to go nowhere. ## Letta Code Versus a Plain LLM SDK The real alternative for most readers is not another agent platform but the model provider's own SDK plus your own state handling. The difference in approach is where memory lives. With a plain SDK you own the store: you persist conversation history, decide what to include in each request, and manage identity yourself. You get full control over the data model and no extra runtime, at the cost of writing and maintaining the memory layer. Letta Code inverts that. Memory, identity, and conversations are the platform's concern, which is why Letta Cloud can keep them available across computers and why the same agent can appear in Slack and in a terminal. You trade control for not building the plumbing. That trade is worth it when memory behaviour is the hard part of your product. It is not worth it when you want a thin wrapper over a completion endpoint, or when your data cannot leave your own infrastructure and you have not yet confirmed how the App Server stores state. ## Licence, Upgrade Cost, and What the Tags Tell You The repository carries the Apache-2.0 licence, which permits commercial use and modification and includes a patent grant. That is the licence on this landing-page repository; the README does not state the licence of letta-ai/letta-code, so confirm it there before you rely on it for a distributed product. This is a factual gap, not legal advice. On upgrade cost, the release list here shows 0.16.6 on 2026-03-04, 0.16.7 on 2026-03-31, and 0.16.8 on 2026-05-14, roughly a monthly patch cadence on the 0.16 line, while pushes to this repository continued through 2026-09-08. Patch-level releases on a 0.x line usually mean the API is still settling, so budget for reading release notes before each bump rather than assuming drop-in upgrades. The npm global install also means upgrades are a single command, which lowers the mechanical cost but not the review cost. ## Conclusion Adopt Letta Code if you want a terminal-first agent harness with memory and channels, or an App Server you can self-host, and you are willing to follow a project whose source moved to letta-ai/letta-code. Do not adopt it if you need the old V1 API server: the README states that source is unsupported, receives no fixes or security updates, and should not be used in production. Before committing, verify that @letta-ai/letta-code exists on your registry and that the letta binary launches on your platform, then confirm the letta server command binds where you expect. ## FAQ ### How do I install Letta? The README gives one install path: npm install -g @letta-ai/letta-code, then run the letta command to launch the interactive terminal UI. The desktop app for macOS, Windows, and Linux and the browser version at chat.letta.com are alternatives if you prefer not to install from npm. ### What is Letta AI? The README describes Letta as a platform for stateful agents: AI with memory that can learn and self-improve over time. It was formerly known as MemGPT, and the current source lives in the letta-ai/letta-code repository. ### How do I use Letta? The README lists several entry points: the interactive terminal UI launched with letta, the App Server started with letta server for local or self-hosted agents, the desktop app, chat.letta.com in a browser, channels such as Slack, Telegram, and Discord, and the Letta Agent SDK for TypeScript applications. ### What is Letta? The README describes Letta as a platform for stateful agents, formerly known as MemGPT, and states that this repository now serves as a landing page while the current source code lives in letta-ai/letta-code. ### How do I install Letta? Install the package globally with npm install -g @letta-ai/letta-code, then run letta to open the interactive terminal UI or letta server to start the App Server. The README does not document a port or configuration file for the server. ## Sources - [letta-ai/letta on GitHub](https://github.com/letta-ai/letta) - [License: Apache-2.0](https://github.com/letta-ai/letta/blob/main/LICENSE) - [Project website](https://docs.letta.com/) - [README](https://github.com/letta-ai/letta/blob/main/README.md) - [Releases](https://github.com/letta-ai/letta/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/letta-ai-letta