Configuration & providers

One function, four providers, no config files.

The mental model

Reveilio stores a single process-global ReveilioConfig object. You set it with reveilio.configure(...), and every other call reads from it. There is no .reveilio.yaml, no Settings class, and no dependency injection.

If you never call configure() but an environment variable such as GEMINI_API_KEY or OPENAI_API_KEY is set, reveilio will auto-configure on first use. An explicit configure() call always overrides environment values.

Gemini (Google AI Studio)

reveilio.configure(
    provider="gemini",
    api_key="AIza...",                   # or leave out and set GEMINI_API_KEY
    model="gemini-2.5-flash",            # optional; this is the default
)

Environment fallback: GEMINI_API_KEY or GOOGLE_API_KEY.

OpenAI

reveilio.configure(
    provider="openai",
    api_key="sk-...",                    # or OPENAI_API_KEY env var
    model="gpt-4o-mini",                 # or gpt-4o, gpt-4-turbo, etc.
)

Any chat-completions-compatible model ID that your account has access to is supported.

Azure OpenAI

Azure requires three additional values: endpoint, deployment name, and (optionally) API version. Note that the model name is the deployment name you configured in the Azure portal.

reveilio.configure(
    provider="azure",
    api_key="...",                               # AZURE_OPENAI_API_KEY env var
    azure_endpoint="https://my-rsrc.openai.azure.com",
    azure_deployment="gpt-4o",                    # deployment name, not model name
    azure_api_version="2024-02-15-preview",       # optional
)

Environment fallbacks: AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT, AZURE_OPENAI_DEPLOYMENT, AZURE_OPENAI_API_VERSION.

Ollama (self-hosted and local)

reveilio.configure(
    provider="ollama",
    base_url="http://localhost:11434",  # default
    model="llama3.2",                    # or any model you have pulled
)

No API key is used. Environment fallback: OLLAMA_BASE_URL. Be sure to run ollama pull llama3.2 (or your chosen model) before invoking reveilio.

Note: smaller local models sometimes return JSON that does not match the schema exactly. Reveilio catches these cases and fills in safe defaults, but scores will be less reliable than those produced by a hosted frontier model.

Tuning the scoring weights

The default weights sum to 1.0 and reflect a general-purpose hiring rubric:

Dimension	Default weight
skills	20%
semantic_skills	20%
experience	25%
education	15%
certifications	10%
soft_skills	5%
domain_relevance	5%

Hiring for a compliance role? Increase certifications. Hiring for a research lab? Increase education. Pass a custom weights dictionary to configure():

reveilio.configure(
    provider="openai",
    api_key="sk-...",
    weights={
        "skills": 0.30,
        "semantic_skills": 0.20,
        "experience": 0.25,
        "education": 0.05,
        "certifications": 0.15,
        "soft_skills": 0.03,
        "domain_relevance": 0.02,
    },
)

The prompt sent to the LLM includes these weights verbatim so the model knows what you are optimizing for.

Switching providers at runtime

Call configure() again. The most recent call wins.

reveilio.configure(provider="gemini", api_key="AIza...")
quick_pass = reveilio.analyze_folder("./resumes", jd)

reveilio.configure(provider="openai", api_key="sk-...", model="gpt-4o")
deep_pass  = reveilio.analyze_folder("./resumes", jd)

Continue to Guides.