LLeMbas
The LLeMbas leaf

LLeMbas

Waybread for the long road of thought

Your own home for language models: a self-hosted server for chats, agents, knowledge and an API over every model you run — and a terminal agent that does the work where your code is.

Run the serverInstall the CLIRead the docs

Products

Three ways in, one account

The server is where your models, people and data live. The CLI and the desktop app sign in to it — or the CLI works entirely on its own.

1.0.0

LLeMbas Server

A self-hosted web UI and API for local and remote language models, for one person or a whole team.

  • Chats, agent chats, a library, reports and schedules
  • OpenAI-compatible and Anthropic connections
  • Users, groups, permissions and credits
  • An OpenAI-compatible /v1 API over all of it
ServerSource
1.0.0

LLeMbas CLI

A terminal coding agent and project manager: the lembas command. Any LLM API, a real TUI, git-native.

  • Four permission modes: manual, edit, auto, plan
  • Every turn snapshotted, /undo and /redo
  • Subagents, memory, skills, MCP and voice
  • Links to a server so agent chats run here
CLISource
Coming soon

LLeMbas Desktop

An installable app for your desktop that points at a LLeMbas server: your chats and agents in a window of their own.

  • Signs in to your own server
  • Same account, same models, same library
  • Nothing to host twice
What is coming

Why LLeMbas

Built for the models you already run

Local-first

The server runs on your hardware, keeps everything in one SQLite database on your disk, and loads nothing from a CDN — every script and font ships with it. API keys are encrypted at rest. Agent commands never run on the server: they run on a machine you linked, under that machine’s own limits.

Your models

Point it at llama.cpp, llama-swap, vLLM, LM Studio, Ollama, OpenRouter, OpenAI, DeepSeek — anything that speaks the OpenAI API — or at Anthropic. Every model gets a clear name, provider/model, and local models stay free beside priced ones.

One place

Chats, agent sessions, knowledge, notes, skills, reports and the API share one account. Start a session in the terminal and carry on in the browser; the library you build in one is the library the other searches.

Quick start

Up in a few minutes

1 · Run the server

With Docker. Put a TLS proxy in front before you use it from other machines. Then open it and create the first account — it becomes the administrator.

shell
git clone https://github.com/LLeMbas/LLeMbas.git
cd LLeMbas

# A secret key, generated once and kept: it signs sessions and encrypts stored API keys
echo "LEMBAS_SECRET_KEY=$(python3 -c 'import secrets; print(secrets.token_urlsafe(48))')" > .env

docker compose up -d --build      # serves on http://127.0.0.1:8080

2 · Link the CLI

On any Linux machine (x64 or arm64). Logging in reads the server’s models, voice and search — nothing to configure twice.

shell
curl -fsSL https://github.com/LLeMbas/LLeMbas-CLI/releases/latest/download/get.sh | bash

lembas login https://your-server
cd your-project && lembas

All install options · Add your first connection · Run agent chats from the browser