> ## Documentation Index
> Fetch the complete documentation index at: https://docs.graphify.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Configuration
> Configure model backends, output paths, and local logging.
Code-only extraction works without a provider key. These backend settings matter when you explicitly run semantic extraction or model-assisted CLI workflows.
## Semantic backends
| `--backend` | Main configuration | Optional package extra |
| ------------ | --------------------------------------------------------------------- | --------------------------------------------------- |
| `claude` | `ANTHROPIC_API_KEY`; optional `ANTHROPIC_BASE_URL`, `ANTHROPIC_MODEL` | `anthropic` |
| `gemini` | `GEMINI_API_KEY` or `GOOGLE_API_KEY` | `gemini` |
| `openai` | `OPENAI_API_KEY`; optional `OPENAI_BASE_URL`, `OPENAI_MODEL` | `openai` |
| `deepseek` | `DEEPSEEK_API_KEY` | Use the client dependency required by your version. |
| `kimi` | `MOONSHOT_API_KEY` | `kimi` |
| `ollama` | Optional `OLLAMA_BASE_URL`, `OLLAMA_MODEL` | `ollama` |
| `bedrock` | Standard AWS credential chain | `bedrock` |
| `azure` | `AZURE_OPENAI_API_KEY`, `AZURE_OPENAI_ENDPOINT`, deployment setting | `openai` |
| `claude-cli` | Installed and authenticated Claude Code CLI | No separate model API key. |
Set credentials in your shell, secret manager, or CI environment. Never commit them into the documentation or repository. Use a model available in your account or local runtime rather than assuming a particular model name.
## Output and resource controls
| Variable | Meaning |
| --------------------------------------------------- | ------------------------------------------------------- |
| `GRAPHIFY_OUT` | Output directory, useful for separate worktrees. |
| `GRAPHIFY_MAX_WORKERS` | Structural extraction worker count. |
| `GRAPHIFY_MAX_OUTPUT_TOKENS` | Semantic model output cap. |
| `GRAPHIFY_API_TIMEOUT` | Backend request timeout in seconds. |
| `GRAPHIFY_MAX_RETRIES` | Retry count for rate-limited requests. |
| `GRAPHIFY_MAX_RETRY_DEPTH` | Limit recursive recovery of truncated chunks. |
| `GRAPHIFY_MAX_GRAPH_BYTES` | Override the graph file size guard; default is 512 MiB. |
| `GRAPHIFY_API_KEY` | Authentication for a shared graph MCP HTTP server. |
| `GRAPHIFY_TRIAGE_BACKEND` / `GRAPHIFY_TRIAGE_MODEL` | Backend and model for CLI PR triage. |
## Local query logging
The current implementation keeps query logging **off by default**.
| Variable | Effect |
| --------------------------------- | ------------------------------------------------------------------ |
| `GRAPHIFY_QUERY_LOG_ENABLE=1` | Enable a local JSONL query log at `~/.cache/graphify-queries.log`. |
| `GRAPHIFY_QUERY_LOG=/path/to/log` | Enable logging to a chosen file. |
| `GRAPHIFY_QUERY_LOG_DISABLE=1` | Force logging off, overriding enable settings. |
| `GRAPHIFY_QUERY_LOG_RESPONSES=1` | Include full responses when logging is enabled. |
Logs can contain questions and corpus paths. Full responses are opt-in. Keep access and retention appropriate to your source material.
## Ignore rules
Use `.graphifyignore` to exclude dependency folders, build artifacts, or irrelevant content. See [incremental updates](/guides/incremental-updates) for examples.