LLMRunner is alpha software.

Only the latest commit on the main branch is currently supported.

LLMRunner is designed to bind to 127.0.0.1 by default. The server now binds to the configured host instead of listening on every interface.

If you change the host in ~/.llmrunner/config.json, enable API key authentication and use firewalling and transport security appropriate for your environment. Do not expose LLMRunner directly to the public internet.

API key authentication is enabled when at least one key is configured.

The recommended local setup is an environment variable:

export LLMRUNNER_API_KEY="$(openssl rand -hex 32)"
llmrunner start

Clients can send either:

Authorization: Bearer <key>
x-api-key: <key>

You can also store keys in security.apiKeys in ~/.llmrunner/config.json, but environment variables are safer for shared config files.

Health checks at /health and /v1/health do not require authentication.

Request logging is enabled by default. Logs include method, path, status, and duration. LLMRunner does not log request bodies, response bodies, API keys, or authorization headers.

HTTP request bodies are capped by security.maxRequestBodyBytes, which defaults to 10485760 bytes.

Use:

llmrunner logs
llmrunner logs --errors
llmrunner logs --follow

llmrunner models pull downloads GGUF files from Hugging Face or direct URLs. Only download and run models from sources you trust.

For private or gated Hugging Face models, LLMRunner reads HF_TOKEN from the environment and sends it to Hugging Face API and download endpoints.

If you find a security issue, please open a GitHub security advisory or contact the maintainers privately before posting exploit details publicly.