# Optional settings for running the sidecar from source. A basic local run # needs none of them: keys are entered in the app and kept in the encrypted # vault (~/.waggle/vault.json). # ── Provider API keys ────────────────────────────────────────────── # Set keys in the app under Settings > Models (Provider Keys) or in the Vault # app. The server loads them from the vault at startup, so you normally do not # set them here. Providers: anthropic, openai, google (gemini), xai, deepseek, # openrouter, mistral, moonshot (kimi), alibaba (qwen), minimax, zhipu (glm), # perplexity, genspark, and tavily for web search. # # These are also read from the environment, for CI and headless runs: # ANTHROPIC_API_KEY= # OPENAI_API_KEY= # VOYAGE_API_KEY= # embeddings only (EMBEDDING_PROVIDER=voyage) # ── Data directory and port ──────────────────────────────────────── # WAGGLE_DATA_DIR=~/.waggle # WAGGLE_PORT=3333 # ── LiteLLM (optional) ───────────────────────────────────────────── # Off by default; the built-in provider proxy handles model calls. Point the # sidecar at an external LiteLLM-compatible proxy only if you run one. # LITELLM_BASE_URL=http://localhost:4000 # LITELLM_MASTER_KEY= # ── Embeddings ───────────────────────────────────────────────────── # auto (default) tries in-process, then Ollama, then an API provider. # Options: auto, inprocess, ollama, voyage, openai, mock # EMBEDDING_PROVIDER=auto # In-process model: downloads about 23 MB on first use, cached in # ~/.waggle/models/, then works offline. # EMBEDDING_MODEL=Xenova/all-MiniLM-L6-v2 # Ollama: needs Ollama running (https://ollama.com). # OLLAMA_HOST=http://localhost:11434 # OLLAMA_EMBED_MODEL=nomic-embed-text