knowledge
Browse documentation
DocumentationConfiguration

Configuration

knowledge is configured through a small set of environment variables. Most users only need to set the API key for their LLM provider; everything else has a sensible default.

LLM provider configuration

The model is selected by the model string passed to --model or Knowledge(model=...). Authentication uses provider-specific environment variables, which litellm reads directly.

OpenAI

Variable Required Example
OPENAI_API_KEY Yes sk-...
OPENAI_API_BASE No https://api.openai.com/v1 
export OPENAI_API_KEY=sk-...
knowledge --model gpt-4o create https://example.com/docs.html ./out

Anthropic

Variable Required Example
ANTHROPIC_API_KEY Yes sk-ant-... 
export ANTHROPIC_API_KEY=sk-ant-...
knowledge --model claude-3-opus-20240229 create https://example.com/docs.html ./out

Google Gemini

Variable Required Example
GEMINI_API_KEY Yes AIza... 
export GEMINI_API_KEY=AIza...
knowledge --model gemini/gemini-1.5-pro create https://example.com/docs.html ./out

Ollama (local)

Variable Required Example
OLLAMA_HOST No http://localhost:11434

Ollama runs locally, so no API key is required.

ollama pull llama3
knowledge --model ollama/llama3 create https://example.com/docs.html ./out

vLLM

vLLM exposes an OpenAI-compatible endpoint. Point litellm at it with OPENAI_API_BASE and any non-empty OPENAI_API_KEY.

export OPENAI_API_BASE=http://localhost:8000/v1
export OPENAI_API_KEY=not-required
knowledge --model open-mistral-nemo create https://example.com/docs.html ./out

HTTP client configuration

These variables control the resilience behaviour of fetch_url.

Variable Default Purpose
KNOWLEDGE_MAX_BODY_SIZE 52428800 (50 MiB) Reject responses whose Content-Length exceeds this value before reading the body.
KNOWLEDGE_REQUEST_TIMEOUT 30 Timeout in seconds for each HTTP request.
KNOWLEDGE_MAX_RETRIES 3 Maximum attempts for transient failures (network errors, HTTP 429, HTTP 5xx).

The current values are baked into the SDK at build time. To override them today, edit knowledge/sdk.py (the constants MAX_BODY_SIZE, REQUEST_TIMEOUT, MAX_RETRIES); runtime overrides are planned for v0.2.

User-Agent

The HTTP client sends User-Agent: knowledge-sdk/0.1.0. If your downstream service filters by User-Agent and you need a custom value, fork the USER_AGENT constant in knowledge/sdk.py.

.env.example

The repository ships an annotated .env.example at the root. Copy it to .env and fill in your values:

cp .env.example .env

The CLI does not read .env automatically. Source the file into your shell, or use a tool like direnv or dotenv to load it before running knowledge.

  1. Export the API key for the provider you will use.
  2. Run knowledge --model <model> create <url> <out> to make a small bundle and verify the provider works.
  3. Scale up to a larger source.
  4. If you hit a transient error, increase KNOWLEDGE_MAX_RETRIES. If you hit a body-size error, increase KNOWLEDGE_MAX_BODY_SIZE.