Skip to content

Configuration

Configuration Wizard

Use the wizard as the primary setup path when choosing profiles, preparing a facts file, or collecting calibration feedback:

director-ai wizard

The [ui] extra enables the Gradio interface:

pip install director-ai[ui]

The wizard includes profile selection, threshold and weight controls, facts.txt preparation, calibration feedback JSONL generation, and an advanced field editor for the full DirectorConfig surface. Use CLI mode when Gradio is not available:

director-ai wizard --cli --output config.yaml

DirectorConfig is a dataclass with environment variable, YAML file, and named profile loaders. All fields have sensible defaults.

Loading

export DIRECTOR_COHERENCE_THRESHOLD=0.6
export DIRECTOR_USE_NLI=true
export DIRECTOR_VECTOR_BACKEND=chroma
export DIRECTOR_METRICS_ENABLED=true

from director_ai.core.config import DirectorConfig
config = DirectorConfig.from_env()

# config.yaml
coherence_threshold: 0.6
hard_limit: 0.5
use_nli: true
nli_model: "lytang/MiniCheck-DeBERTa-L"
vector_backend: chroma
chroma_persist_dir: "./data/chroma"
metrics_enabled: true

config = DirectorConfig.from_yaml("config.yaml")

config = DirectorConfig.from_profile("thorough")

config = DirectorConfig(
    coherence_threshold=0.7,
    use_nli=True,
    scorer_backend="onnx",
)

Profiles

Profile NLI Threshold Candidates Metrics Use Case
fast Off 0.5 1 Off Development, low latency
thorough On 0.6 3 On Production default
research On 0.7 5 On Evaluation, benchmarking
medical On 0.30 3 On Healthcare (measured on PubMedQA)
finance On 0.30 3 On Financial services (measured on FinanceBench)
legal On 0.30 3 On Legal document review (not yet measured)
creative Off 0.4 3 On Creative writing (low halt rate)
customer_support Off 0.55 3 On Support agents
summarization On 0.15 1 On Document summarization
lite Off 0.5 1 Off Zero-dependency fast path

Building Components

config = DirectorConfig.from_profile("thorough")

# Build scorer with all config applied
scorer = config.build_scorer(store=my_store)

# Build vector store from config
store = config.build_store()

Combining Profile + Overrides

config = DirectorConfig.from_profile("medical")
config.nli_model = "lytang/MiniCheck-DeBERTa-L"
config.cache_size = 4096

Or combine YAML + env vars (env vars take precedence):

config = DirectorConfig.from_yaml("config.yaml")
# DIRECTOR_USE_NLI=true overrides use_nli in YAML

Key Field Groups

Scoring

coherence_threshold, hard_limit, soft_limit, use_nli, nli_model, scorer_backend, max_candidates, history_window

LLM Provider

llm_provider (mock | openai | anthropic | huggingface | local), llm_api_key, llm_model, llm_temperature, llm_max_tokens

Vector Store

vector_backend (memory | chroma), embedding_model, chroma_collection, chroma_persist_dir, reranker_enabled

Server

server_host, server_port, server_workers, cors_origins, rate_limit_rpm, api_keys

Caching

cache_size, cache_ttl, redis_url

Observability

metrics_enabled, log_level, log_json, otel_enabled

See DirectorConfig API Reference for the complete field table.