Launchstack

A TypeScript engine for AI-native applications. Ingestion, OCR, RAG, knowledge graph, LLM abstractions, and background jobs — ports-based, and wired into a Next.js reference app that shows how the pieces fit together.

Run it · Repository layout · Packages · Architecture · Chat models · Contributing

---

Status

The publishable bricks in packages/ and the compositions package @launchstack/pipelines are not yet on npm. The engine aggregate is @launchstack/engine; @launchstack/design-tokens is private. The first release will publish the public packages together through the Changesets flow in release.yml, which builds the bricks and pipelines, validates the packed engine tarball with publint, and checks every declared publishable subpath for Node-ESM loadability. Until that first release lands, consume the engine by running this repository.

To try Launchstack today, run the app below.

---

Run it locally

Requirements: Node ≥ 20 and pnpm 10.15.1 (corepack enable picks up the pinned version).

git clone https://github.com/Deodat-Lawson/LaunchStack.git
cd LaunchStack
pnpm install
cp .env.example .env

apps/web/src/env.ts refuses to boot without DATABASE_URL and BETTER_AUTH_SECRET. Compose supplies the database, file storage, and every local service; .env.example already points DATABASE_URL at it. A usable instance — sign-up, searchable uploads, chat — needs these in .env:

Variable	Value
BETTER_AUTH_SECRET	openssl rand -base64 32
BETTER_AUTH_URL	http://localhost:3000 (the .env.example default). Without it, browser sign-up is refused with "Invalid origin".
Chat	Gemini: GOOGLE_AI_API_KEY, with CHAT_BASE_URL unset. Azure OpenAI: CHAT_BASE_URL=https://<resource>.openai.azure.com/openai/v1, CHAT_API_KEY, and CHAT_MODELS_CONFIG=config/chat-models.azure.yaml. See Chat models.
Embeddings	EMBEDDING_API_BASE_URL + EMBEDDING_API_KEY. There is no default: without them, uploads are read and chunked but never become searchable. Azure OpenAI fits the default EMBEDDING_INDEX (legacy-openai-1536) as-is, with text-embedding-3-large deployed under that name. Gemini needs EMBEDDING_INDEX=gemini-embedding-768 on a fresh database.

There is no per-vendor variable: a bare OPENAI_API_KEY, OPENROUTER_API_KEY or OLLAMA_BASE_URL will not configure chat, and none of them is forwarded to the Gemini default — every one of those providers speaks the OpenAI chat-completions protocol, so each is reached through CHAT_BASE_URL. Only AI_BASE_URL/AI_API_KEY, a straight rename of the canonical pair, is still translated for a release with a deprecation warning.

With Docker (recommended)

make up-prod     # lite stack, detached (~400MB RAM)
make up-ocr      # adds Docling for Office docs, detached (~1.2GB RAM)
make logs        # follow logs
make down        # stop containers (keeps volumes — DB + S3 data persists)
make down-clean  # stop + wipe volumes (fresh DB on next up)