Download self-host

Pull the public images from GHCR. Download the config package, pin the release tag, then docker compose pull. No sign-in or GitHub account is required to pull the images. The same install guide is in the public README.

Release v1.0.0

Images: ghcr.io/pullsight-inc/pullsight-db:v1.0.0, pullsight-api, pullsight-web

Download config package (14.4 KB)

Install

The config package includes SELF_HOST.md with webhook and OAuth details. The same steps are in the public README. Postgres and Redis stay on the Compose network.

Requirements

Docker Engine 24+ and Docker Compose v2
A network path to GHCR
Disk for the images and the Postgres volume
Optional: Ollama for local LLM inference (see “Choose a model” below)
Public HTTPS URL (or tunnel) for webhooks and OAuth callbacks in production

1. Unpack the config package

Unzip pullsight-selfhost-v1.0.0-config.zip. It expands to a self-host/ folder with the Compose file, env templates, and an empty images/ directory. It does not contain the application images.

2. Configure environment

cp .env.example .env
cp web.env.example web.env

Edit .env and web.env. Compose will not start until POSTGRES_PASSWORD and REDIS_PASSWORD are set. Also set PULLSIGHT_SECRET_KEY, PULLSIGHT_PROVISION_SECRET, PULLSIGHT_ENCRYPTION_KEY, PULLSIGHT_API_KEY_SALT, and AUTH_SECRET.

3. Pin the release

Use the same tag for the database, API, and web images:

PULLSIGHT_IMAGE_TAG=v1.0.0
PULLSIGHT_IMAGE_PREFIX=ghcr.io/pullsight-inc

latest moves only when a stable release is published. Do not pin production to it.

4. Pull the images and start

docker compose pull
docker compose up -d

Open http://localhost:3000, sign in, and connect a repository. redis:7-alpine is pulled from Docker Hub with the stack. It is not a Pullsight image.

Choose a model

Pullsight needs an LLM to review code. Run a local model with Ollama for free, or point it at a hosted provider with your own API key.

Option A — Local model with Ollama (free, default)

Download Ollama

Install it on the API server host, or on another machine the API can reach. Then pull a model:

ollama pull qwen2.5-coder:3b

Configure the API in your .env:

PULLSIGHT_DEFAULT_LLM_PROVIDER=ollama
PULLSIGHT_OLLAMA_MODEL=qwen2.5-coder:3b

# Ollama on the same Docker host:
PULLSIGHT_OLLAMA_BASE_URL=http://host.docker.internal:11434

# Ollama on a separate machine (start it with OLLAMA_HOST=0.0.0.0,
# open port 11434), then point the API at that host instead:
# PULLSIGHT_OLLAMA_BASE_URL=http://<ollama-host>:11434

Option B — Your own API key (no Ollama required)

Set PULLSIGHT_DEFAULT_LLM_PROVIDER to the provider you want and add its key in .env:

OpenAI (keys): PULLSIGHT_DEFAULT_LLM_PROVIDER=openai, PULLSIGHT_OPENAI_API_KEY, PULLSIGHT_OPENAI_MODEL
Anthropic (keys): PULLSIGHT_DEFAULT_LLM_PROVIDER=anthropic, PULLSIGHT_ANTHROPIC_API_KEY, PULLSIGHT_ANTHROPIC_MODEL
Google Gemini (keys): PULLSIGHT_DEFAULT_LLM_PROVIDER=gemini, PULLSIGHT_GEMINI_API_KEY,

Back home