Skip to content

docker.md

Latest commit

 

History

History
142 lines (98 loc) · 3.97 KB

docker.md

File metadata and controls

142 lines (98 loc) · 3.97 KB
Raw

Docker deployment

Run llmwiki in a container — no Python install, no pip, no venv. Two supported setups:

  1. Pull pre-built image from GitHub Container Registry (recommended)
  2. Build locally from the repo Dockerfile (for development)

Prerequisites

  • Docker Desktop (macOS / Windows) or Docker Engine 20+ (Linux)
  • docker compose plugin (bundled with recent Docker Desktop; on Linux install the docker-compose-plugin package)

Quick start — pre-built image

The image is published to ghcr.io/pratiyush/llm-wiki:latest on every release tag. To run:

# 1. Clone the repo (you need the Dockerfile + examples for the volume mounts)
git clone https://github.com/Pratiyush/llm-wiki.git
cd llm-wiki

# 2. Pull the latest image
docker compose pull

# 3. Start the server
docker compose up -d

# 4. Open http://localhost:8765

The server runs in the background with restart: unless-stopped, so it comes back after reboots / Docker restarts.

Running CLI commands in the container

Any llmwiki subcommand works inside the container via docker compose run:

# Build the static site from raw/ + wiki/
docker compose run --rm llmwiki build

# Sync session transcripts (needs host agent stores bind-mounted)
docker compose run --rm llmwiki sync --dry-run

# Run every registered lint rule (15 at last count — see `llmwiki lint --help`)
docker compose run --rm llmwiki lint --fail-on-errors

# Generate a knowledge graph
docker compose run --rm llmwiki graph

Because the repo's raw/, wiki/, and site/ directories are bind-mounted, the output shows up on the host too.

Build locally

If you're developing llmwiki itself or need unreleased changes:

docker compose build
docker compose up -d

This uses the repo Dockerfile and re-builds on every code change.

Image details

  • Base: python:3.12-slim (~45 MB)
  • Runtime deps: markdown only (stdlib + optional graphifyy extra)
  • User: non-root (app, UID 1000) — mounted volumes stay host-owned
  • Port: 8765 (exposed + mapped)
  • Entrypoint: python -m llmwiki
  • Default CMD: serve --host 0.0.0.0 --port 8765 --dir site
  • Labels: OCI standard (title, description, source, licenses, authors)

Volumes

Host path Container path Purpose
./raw /wiki/raw Session transcripts (gitignored user data)
./wiki /wiki/wiki LLM-maintained wiki pages (gitignored)
./site /wiki/site Generated HTML (gitignored)
./examples /wiki/examples (read-only) Demo sessions + config templates

Publishing your own image

Only the repo maintainer can publish to ghcr.io/pratiyush/llm-wiki. To publish under your own namespace:

docker build -t ghcr.io/<your-user>/llm-wiki:latest .
docker push ghcr.io/<your-user>/llm-wiki:latest

Or fork the repo and the release workflow will publish under your fork's namespace automatically on tag push.

Privacy

The container reads from your bind-mounted directories. Nothing leaves the container — no telemetry, no external API calls. Same privacy guarantees as the CLI version.

Troubleshooting

"Permission denied" on mounted volumes

The container runs as UID 1000. If your host user is a different UID, either match UIDs in the Dockerfile or change host ownership:

sudo chown -R 1000:1000 raw/ wiki/ site/

Can't pull from ghcr.io

The GitHub Container Registry requires authentication for private images. For this repo (public), pulls should work without auth. If you see 404s, the image hasn't been published yet — fall back to local build:

docker compose build && docker compose up -d

Port 8765 already in use

Change the host port in docker-compose.yml:

ports:
  - "9999:8765"  # access at http://localhost:9999

Related docs