Configuration

All configuration is done via environment variables in a .env file at the project root.

Setup

cp .env.example .env

LLM Endpoint

# For local LLM (llama.cpp, Ollama)
LLM_BASE_URL=http://localhost:11434/v1
LLM_API_KEY=
LLM_MODEL=llama3.2

# For vLLM
LLM_BASE_URL=http://localhost:8000/v1
LLM_API_KEY=
LLM_MODEL=meta-llama/Llama-3-8b-Instruct

# For OpenRouter / cloud providers
LLM_BASE_URL=https://openrouter.ai/api/v1
LLM_API_KEY=your-api-key-here
LLM_MODEL=moonshotai/kimi-k2-thinking

Variable	Default	Description
LLM_BASE_URL	https://openrouter.ai/api/v1	OpenAI-compatible API base URL
LLM_API_KEY	(empty)	API key (leave empty for local servers)
LLM_MODEL	moonshotai/kimi-k2-thinking	Model identifier
LLM_TIMEOUT	120	Request timeout in seconds

Agent Configuration

Variable	Default	Description
AGENT_DEFAULT_MODEL	(same as LLM_MODEL)	Default model for new agents
AGENT_MAX_CONTEXT	4096	Maximum context length for agents
AGENT_WORKSPACE_DIR	./agent_workspaces	Directory for isolated agent workspaces

Local Model Settings

Variable	Default	Description
LOCAL_MODEL_RUNNER	ollama	Default local model runner (ollama, llama-cpp, vllm)
LOCAL_MODEL_ENDPOINT	http://localhost:11434	Local model API endpoint
LOCAL_MODEL_QUANTIZATION	auto	Preferred quantization level

Two-Pass Extraction

The two-pass system first asks the LLM to reason about a problem, then makes a second call to extract just the final answer in a strict format.

Variable	Default	Description
TWO_PASS_ENABLED	1	Enable two-pass answer extraction
TWO_PASS_TEMPERATURE	0.0	Temperature for the extraction pass

Flask Server

Variable	Default	Description
HOST	0.0.0.0	Server bind address
PORT	8080	Server port
DEBUG	1	Enable Flask debug mode
SECRET_KEY	dev-secret-key-...	Flask secret key (change in production)

Logging

Variable	Default	Description
LOG_FULL_THINKING	0	Include full thinking content in live logs
LOG_FULL_RESPONSE	0	Include full LLM response in live logs

Optional

Variable	Default	Description
ANTHROPIC_API_KEY	(empty)	For the improver module (analysis & training data generation)

Model Explorer Overrides

Override the default evaluator for specific domains via environment variables:

EVALUATOR_MATH=keyword        # Use keyword evaluator for math domain
EVALUATOR_CONVERSATION=two_pass  # Use two-pass for conversation

Available evaluator types: two_pass, keyword, sql_executor, tool_call.