Skip to content

Configuration

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 1 Off Creative writing (low halt rate)
customer_support Off 0.55 1 Off 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.