Skip to content

README.md

Latest commit

 

History

History
46 lines (32 loc) · 3.15 KB

README.md

File metadata and controls

46 lines (32 loc) · 3.15 KB
Raw

llame

A self-hosted, multi-user personal AI assistant platform — not just a chat UI, but an AI operating layer you run yourself.

llame treats the things a real assistant needs — groups, projects, goals, todos, skills, commands, connectors, model credentials, memories, artifacts, and knowledge bases — as first-class, durable, governed entities. It is designed to serve a single person, a family, a team, or an organization from the same core, with explicit ownership, permissions, and auditability throughout.

What it does

  • Multi-user from the ground up — nested groups (org → team → project), roles, and policy-based access (deny-overrides-allow).
  • Projects — shared workspaces with their own chats, knowledge, connectors, skills, artifacts, and members.
  • Durable agent runs — every message becomes a worker-processed run with a replayable event stream, so progress survives a page refresh.
  • Wiki-centric knowledge — Obsidian vaults, Notion, local Markdown, and Git repos normalized into searchable Knowledge Spaces; your notes are the assistant's memory, not an afterthought.
  • Artifacts — versioned, shareable, executable work products (documents, diagrams, components) linked to chats, projects, and Git.
  • Bring your own model — works with no instance-wide provider; users/groups/projects supply their own credentials (OpenAI-compatible, Anthropic, local via Ollama, …).
  • MCP & connectors — first-class Model Context Protocol host plus connectors (GitHub, filesystem, Notion, …) under per-scope capability policy.
  • Skills, slash commands, goals & todos — durable agent-control primitives, with skills following the open SKILL.md Agent Skills format.
  • Messaging channels — reach the same assistant from the web app, Telegram, Discord, and more.

Status

Early/WIP. A chat proof-of-concept (auth, model selection, persisted chats, agent orchestration) is in place; active work is building toward the self-hosted MVP. See ROADMAP.md for what's planned and CHANGELOG.md for what's shipped.

Getting started

pnpm install
pnpm dev        # run all apps in watch mode

Copy each app's .env.example to .env.local: apps/api owns the database and the chat loop, so it needs POSTGRES_URL and a model-provider key (OPENAI_API_KEY) — without the key, chat messages return 402 and no reply is generated; apps/web is a thin client and only needs NEXT_PUBLIC_API_URL. Full developer setup, commands, and architecture live in AGENTS.md (and per-app AGENTS.md files).

Documentation

  • SPEC.md — full product & architecture specification
  • ROADMAP.md — planned milestones
  • CHANGELOG.md — shipped history
  • AGENTS.md — how to work in this repo (dev setup, conventions, structure)

Architecture at a glance

A pnpm + Turborepo monorepo, TypeScript end-to-end:

  • apps/web — Next.js front end and BFF
  • apps/api — NestJS backend, database, and (in progress) the durable run worker
  • packages/ui — shared shadcn/ui component library
  • packages/config-* — shared ESLint / TypeScript configs