Environment variables¶

hlock reads these variables. Only the worker (the devbox and cluster backends) uses them. The local backend ignores them all.

Worker¶

Variable	Default	Description
`HL_PYTHON`	`python3`	Interpreter the Rust core uses to launch the worker. Point it at one with `hlock` installed.

LLM calls¶

Set these to call a real model through any OpenAI-compatible endpoint. Without HL_LLM_BASE_URL, the worker uses a deterministic offline reference LLM.

Variable	Default	Description
`HL_LLM_BASE_URL`	unset	Base URL of an OpenAI-compatible API, e.g. `http://localhost:8000/v1`
`HL_LLM_MODEL`	unset	Model name to request
`HL_LLM_API_KEY`	unset	API key, if the endpoint needs one

Fine-tuning¶

Used by hlock optimize --depth ft and --depth full.

Variable	Default	Description
`HL_FT_BASE_MODEL`	`Qwen/Qwen2-0.5B`	Base model for LoRA fine-tuning
`HL_FT_STEPS`	`3`	Training steps
`HL_GPU_COST_PER_H`	`2.0`	GPU price per hour in USD, used for cost evidence

Output¶

Variable	Description
`NO_COLOR`	Any value turns off colored CLI output

hlock backend status shows which of these are set. API keys are reported as set or unset, never printed.