Prerequisites
Everything you need before starting a self-hosted DocuGardener deployment.
Software Requirements
| Software | Minimum Version | Notes |
|---|---|---|
| Docker | 24+ | Docker Desktop or Docker Engine |
| Docker Compose | v2 | Bundled with Docker Desktop; standalone on Linux |
| Node.js | 20+ | LTS recommended; used for the Next.js control plane |
| Python | 3.13+ | Only needed if running the analysis plane outside Docker |
| Git | 2.x | Used for cloning repositories during analysis |
Hardware Minimums
| Resource | Minimum | Recommended |
|---|---|---|
| CPU | 2 cores | 4+ cores |
| RAM | 4 GB | 8+ GB |
| Disk | 20 GB | 50+ GB |
Note: The Weaviate vector database stores embeddings for every analysed document and code file. Disk usage grows with the number of repositories and their size. Plan for at least 20 GB free.
GitHub Requirements
- A GitHub account (personal or organisation).
- Ability to create a GitHub App — you need admin access to the organisation, or use a personal account.
- The GitHub App requires these permissions: repository contents (read), check runs (write), pull requests (read), and webhooks.
LLM Provider
DocuGardener needs access to a large language model for drift analysis and fix generation. You have four options:
| Provider | Env Var | Notes |
|---|---|---|
| Google Gemini | GEMINI_API_KEY | Default provider. Fast and cost-effective. |
| OpenAI | OPENAI_API_KEY | GPT-4o recommended. |
| Anthropic | ANTHROPIC_API_KEY | Claude Sonnet 4.6 recommended. |
| Ollama | OLLAMA_URL | Local-only, zero data egress. Perfect for air-gapped environments. |
Ollama tip: Any model with 4K+ context window works for drift analysis. For embeddings, use
nomic-embed-text. Pull it with ollama pull nomic-embed-text.