# pyproject.toml: [tool.quell] settings

> Reference for every [tool.quell] key the Python quelltest package defines, with types and defaults, and which keys the code actually reads today.

Canonical page: https://quelltest.com/docs/configuration/pyproject

The Python package reads its settings from a `[tool.quell]` table in `pyproject.toml`. Run [`quell init`](https://quelltest.com/docs/cli/init.md) to add the defaults, then edit the keys below. Settings are optional: with no table, quell uses safe defaults (rule-based only, no LLM).

## How is the configuration loaded?

The command reads `pyproject.toml` in the project root (`--root`), takes `[tool.quell]`, and validates it with a Pydantic model. Keys the model does not define are ignored. If the table cannot be parsed or a value has the wrong type, quell silently falls back to the safe defaults, so check that a setting really takes effect after you change it.

## Which keys exist?

The last column says where the key is read in the code tested for these docs (the repository at 2.0.1). Keys marked "defined only" are part of the model and written by `quell init`, but no command path read them, so changing them has no effect yet.

| Key | Type | Default in the model | Read by |
| --- | --- | --- | --- |
| `llm_provider` | string | `"anthropic"` (`quell init` writes `"groq"`; with no table, `"none"`) | LLM client when `--use-llm` is passed: `groq`, `anthropic`, `openai` or `ollama` |
| `llm_model` | string | `"claude-sonnet-4-5"` (`quell init` writes `"llama-3.3-70b-versatile"`) | LLM providers |
| `ollama_base_url` | string | `"http://localhost:11434"` | Ollama provider |
| `verification_timeout_seconds` | int | `30` | Verifier: timeout for each pytest run |
| `backup_dir` | path | `".quell/backups"` | Verifier and fix suggester: temporary files and backups |
| `audit_log_path` | path | `".quell/audit.jsonl"` | Writer: one log entry per write |
| `scaffold_dir` | path | `"tests/scaffold"` | Scaffold stubs for gaps that cannot be auto-tested |
| `enable_pyspark` | bool | `false` | Python SDK and MCP tools: read PySpark schemas (needs the `pyspark` extra) |
| `sync` | bool | `false` | `quell sync status`: opt in to cloud sync; never enabled automatically |
| `sync_project_alias` | string | `""` | Cloud sync: project name in the dashboard and in Claude |
| `use_llm` | bool | `false` | defined only: use the `--use-llm` flag |
| `max_verification_attempts` | int | `3` | defined only |
| `auto_write` | bool | `false` | defined only for `quell find` |
| `enable_docstring`, `enable_types` | bool | `true` | defined only |
| `enable_mutations` | bool | `false` | defined only (mutmut integration is off by default and not required) |
| `score_threshold`, `diff_only` | float, bool | `0.0`, `false` | defined only |
| `prs_threshold` | int | `60` | defined only: `quell ci` takes its threshold from `--threshold` |
| `sync_include_scaffolded`, `sync_include_flagged_locations` | bool | `true` | defined only |

Keys that earlier documentation listed, such as `paths`, `exclude`, `test_dir`, `threshold`, `enable_llm` and `min_confidence`, do not exist in the model.

## What is a minimal configuration?

```toml
[tool.quell]
llm_provider = "groq"
scaffold_dir = "tests/scaffold"
```

## Where do I read more?

See [LLM providers](https://quelltest.com/docs/configuration/llm-providers.md) and the [connector setup](https://quelltest.com/docs/connector.md).

The npm package has no configuration file: it uses command-line flags and a `.quellignore` file. See the [npm CLI reference](https://quelltest.com/docs/npm/cli.md).
