ld
Reference

Configuration file (config.yaml)

The contract between llama-dash and llama-swap.

config.yaml is the interface between llama-dash and llama-swap. It declares your models and peers; llama-swap owns its format. Start from llama-swap's example upstream config when building a new file. When you edit it through the dashboard's config editor, llama-dash validates the file before writing and checks for external modifications.

Write contract

  • Validate before writing - content is parsed as YAML and validated against llama-swap's published config-schema.json before any write.
  • Detect external edits - the editor sends the mtime it loaded; if the file changed on disk, llama-dash rejects the save and asks you to reload.
  • Write exactly what is in the editor - comments and key order are preserved when you do not use the Format action. Format intentionally reparses and reprints YAML.

The current config editor writes the file directly and does not create an automatic backup. Keep config.yaml in version control or make a manual backup before high-risk edits.

Hot reload

llama-swap watches the file with -watch-config (fsnotify) and reloads automatically on change. There is no reload endpoint — llama-dash writes the file and llama-swap picks it up.

Set INFERENCE_CONFIG_FILE to the absolute path of config.yaml to enable the in-dashboard config editor.

llama-swap surface llama-dash uses

  • POST /v1/* — forward OpenAI/Anthropic calls (after llama-dash middleware)
  • GET /running — currently loaded models
  • POST /models/unload — unload a model
  • GET /v1/models — model list, including peers
  • GET /health — backend health
  • GET /logs/stream* — SSE log streams (proxy, upstream, per-model)

Last updated on

On this page