Skip to content

configuration.md

Latest commit

 

History

History
117 lines (84 loc) · 4.43 KB

configuration.md

File metadata and controls

117 lines (84 loc) · 4.43 KB
Raw

Configuration

All configuration is sourced from the environment. Nothing is required for offline, deterministic analysis — every setting has a safe default. Configuration is read once per process and is dependency-light (no pydantic-settings requirement), so the core stays importable in minimal environments.

Core variables

Variable Default Purpose
DEVOPS_AI_PROVIDER null Active AI provider: null, anthropic, openai, gemini, ollama
DEVOPS_AI_TIMEOUT 60 Provider HTTP request timeout, in seconds
DEVOPS_AI_MAX_CHARS 200000 Max input characters analyzed (input is truncated above this)

null is the offline default: the engine runs in pure deterministic mode and never calls out.

Provider keys and models

Variable Default Purpose
ANTHROPIC_API_KEY (unset) Anthropic API key
OPENAI_API_KEY (unset) OpenAI API key
GEMINI_API_KEY (unset) Gemini API key (GOOGLE_API_KEY also accepted)
OLLAMA_HOST http://localhost:11434 Ollama server base URL (local, no key)

A provider is considered available only when its key (or, for Ollama, its host) is present. If you request enrichment without an available provider, the engine returns deterministic results and adds a low-severity warning — it never fails. See AI providers.

Per-provider model and endpoint overrides

Variable Default
DEVOPS_AI_ANTHROPIC_MODEL claude-sonnet-4-6
DEVOPS_AI_OPENAI_MODEL gpt-4o-mini
OPENAI_BASE_URL https://api.openai.com/v1
DEVOPS_AI_GEMINI_MODEL gemini-2.0-flash
DEVOPS_AI_OLLAMA_MODEL llama3.1

The OPENAI_BASE_URL override lets you point the OpenAI adapter at any OpenAI-compatible endpoint (local gateways, proxies, alternative hosts).

Alternative key names

For convenience these alternates are also recognized:

  • DEVOPS_AI_ANTHROPIC_KEY → Anthropic key
  • DEVOPS_AI_OPENAI_KEY → OpenAI key
  • DEVOPS_AI_GEMINI_KEY, GOOGLE_API_KEY → Gemini key

Examples

Offline (default — nothing to set):

devops-ai analyze app.log

Anthropic enrichment:

export DEVOPS_AI_PROVIDER=anthropic
export ANTHROPIC_API_KEY=sk-ant-...
export DEVOPS_AI_ANTHROPIC_MODEL=claude-sonnet-4-6   # optional override
devops-ai analyze app.log --enrich

Local, fully private LLM via Ollama:

export DEVOPS_AI_PROVIDER=ollama
export OLLAMA_HOST=http://localhost:11434
export DEVOPS_AI_OLLAMA_MODEL=llama3.1
devops-ai analyze app.log --enrich

OpenAI-compatible gateway:

export DEVOPS_AI_PROVIDER=openai
export OPENAI_API_KEY=sk-...
export OPENAI_BASE_URL=https://my-gateway.internal/v1

Tighten resource limits:

export DEVOPS_AI_MAX_CHARS=50000   # analyze at most 50k chars
export DEVOPS_AI_TIMEOUT=30        # 30s provider timeout

Settings in code (SDK)

You can bypass the environment entirely by constructing Settings and injecting them:

from devops_ai_toolkit import AnalysisEngine
from devops_ai_toolkit.utils.config import Settings

settings = Settings(provider="anthropic", anthropic_api_key="sk-ant-...", max_input_chars=50_000)
engine = AnalysisEngine(settings=settings)

Settings.from_env() builds the same object from a dict of environment variables, which is useful in tests. See the Testing guide.

Precedence

  1. Explicit arguments (e.g. CLI --provider, SDK provider=/settings=)
  2. Environment variables
  3. Built-in defaults (offline null provider)

See also

  • AI providers — adapter details and how to add one
  • Security — how keys are handled (read from env, sent only to the chosen provider)