AI Providers (LiteLLM)

Gemini is the bundled default (GEMINI_API_KEY). Use OpenAI, Anthropic, Mistral, Groq, Ollama or any LiteLLM-compatible model.

Configuration (`.loki/config.yaml`)

By default, LOKI uses Gemini (gemini/gemini-flash-latest) and reads GEMINI_API_KEY from your environment. You can override the provider, model, endpoint, or API key variable in .loki/config.yaml:

.loki/config.yaml

ai:
  provider: gemini              # Default provider (prefixes model if it has no "/")
  model: gemini-flash-latest    # Default model, or any LiteLLM id (e.g. openai/gpt-4o, ollama/llama3)
  api_base: https://host/v1     # Optional: custom or local OpenAI-compatible endpoint
  api_key_env: MY_PROVIDER_KEY  # Optional: environment variable holding the API key
  api_key: AIza...              # Optional: inline API key (prefer api_key_env for shared repos)

Model Profiles (`.loki/models.json`)

LOKI also maintains a registry of named AI profiles in .loki/models.json (editable via /model inside loki chat). The active profile takes priority over .loki/config.yaml across every AI feature (chat, business rules evaluation, fix, and --auto-heal).

.loki/models.json

{
  "active": "local-llama",
  "profiles": [
    {
      "name": "local-llama",
      "model": "ollama/llama3",
      "api_base": "http://localhost:11434/v1"
    },
    {
      "name": "work-mistral",
      "model": "mistral/mistral-large-latest",
      "api_key_env": "MISTRAL_API_KEY"
    }
  ]
}

Resolution Priority

1. Explicit --model CLI flag
2. Active profile in .loki/models.json (managed with /model in loki chat)
3. ai: section in .loki/config.yaml
4. Bundled default (gemini/gemini-flash-latest)

API keys

bash

# Default provider (Gemini)
export GEMINI_API_KEY=AIza...
# Or on Windows PowerShell:
$env:GEMINI_API_KEY="AIza..."

Avoid committing plain-text api_key values in .loki/config.yaml or .loki/models.json to public repositories. Prefer environment variables (api_key_env).

Previousloki update & version NextBusiness Rules