Configuration

Configuration

Sentinel is configured via sentinel.toml in the repository root. All settings are optional — Sentinel uses sensible defaults.

Create a config file

sentinel init /path/to/repo

Full reference

[sentinel]
# Model provider: "ollama" (default), "openai", or "azure"
provider = "ollama"

# Model name for LLM judge and LLM-assisted detectors
model = "qwen3.5:4b"

# Ollama API URL (only used when provider = "ollama")
ollama_url = "http://localhost:11434"

# API base URL (for openai/azure providers)
api_base = ""

# Environment variable containing the API key (for openai/azure)
api_key_env = ""

# Database path (relative to repo root)
db_path = ".sentinel/sentinel.db"

# Output directory for reports
output_dir = ".sentinel"

# Skip the LLM judge step (use raw detector findings)
skip_judge = false

# Skip LLM-assisted detectors (semantic-drift, test-coherence)
skip_llm = false

# Embedding model for semantic context (empty = disabled)
embed_model = ""

# Embedding chunk size and overlap
embed_chunk_size = 50
embed_chunk_overlap = 10

# Custom detectors directory (empty = built-in only)
detectors_dir = ""

# LLM context window size in tokens
num_ctx = 2048

# Model capability tier: "none", "basic", "standard", "advanced"
# Controls which LLM features activate. Higher tiers unlock richer analysis.
model_capability = "basic"

# Minimum confidence threshold for report inclusion (0.0 = show all)
min_confidence = 0.0

# Explicit list of detectors to run (mutually exclusive with disabled_detectors)
# enabled_detectors = ["todo-scanner", "complexity", "dep-audit"]

# Detectors to skip (mutually exclusive with enabled_detectors)
# disabled_detectors = ["lint-runner"]

Capability tiers

Tier	Models	What it enables
none	—	Deterministic detectors only, no LLM
basic	4B+ (Qwen3.5 4B, Phi-3.5)	LLM judge, semantic-drift, test-coherence (binary signals)
standard	9B+ or small cloud (GPT-5.4-nano)	Enhanced analysis, finding cluster synthesis, richer prompts
advanced	Frontier cloud	Deep semantic analysis (reserved for future use)

See Capability Tiers for details on how tiers affect detector behavior, the synthesis gate, and enhanced vs basic mode.

Per-detector provider overrides

Route specific detectors to different models or providers. See Per-Detector Providers for details.

[sentinel.detector_providers.semantic-drift]
provider = "openai"
model = "gpt-4o-mini"
api_base = "https://api.openai.com/v1"
api_key_env = "OPENAI_API_KEY"
model_capability = "standard"

CLI overrides

CLI flags override sentinel.toml values for a single run:

sentinel scan /repo --model qwen3.5:8b --capability standard --skip-judge

Environment variables

Variable	Used by
SENTINEL_GITHUB_TOKEN	create-issues — GitHub API token
SENTINEL_GITHUB_OWNER	create-issues — GitHub repo owner
SENTINEL_GITHUB_REPO	create-issues — GitHub repo name
Value of api_key_env	Provider authentication (e.g., OPENAI_API_KEY)