Self-hosting
Run the Vectorless engine yourself with Docker, Neon, Cloudflare R2, and Upstash QStash.
This page is a placeholder. A complete, step-by-step deployment guide lands in a later release. The architecture below is the intended self-host stack.
Vectorless is built to be self-hosted with a small, serverless-friendly
footprint. There is no vector database to run — the engine stores document
trees and orchestrates treewalk over managed primitives.
The stack
| Concern | Service | Role |
|---|---|---|
| Database | Neon (Postgres) | Document trees, metadata, job state |
| Object storage | Cloudflare R2 | Original document files |
| Background jobs | Upstash QStash | Async ingestion / parsing queue |
| Runtime | Docker | The engine container |
| LLM | your provider | Powers the treewalk agent |
Shape of a deployment
Provision the backing services
Create a Neon Postgres database, a Cloudflare R2 bucket, and an Upstash QStash queue. Collect their connection strings and credentials.
Configure the engine
Supply the engine container with the connection details and your LLM provider key via environment variables.
DATABASE_URL="postgres://...neon..."
R2_BUCKET="vectorless-docs"
R2_ACCESS_KEY_ID="..."
R2_SECRET_ACCESS_KEY="..."
QSTASH_TOKEN="..."
LLM_API_KEY="..."Run with Docker
docker run -p 8080:8080 --env-file .env ghcr.io/hallelx2/vectorless-engine:latestPoint an SDK at it
Set VECTORLESS_API_URL to your deployment and start asking. See the
Quickstart.
Why this stack
- No vector DB to provision, scale, or keep in sync.
- Serverless-friendly — Neon, R2, and QStash all scale to zero.
- Portable — the engine is a single Docker image you can run anywhere.