Explicit AI actions

AI is optional and explicit. Hokan recognizes natural-language intent locally, and only sends a network request after you select the AI action. The returned command is inserted for review; it is not executed automatically.

This keeps the useful part of AI assistance close to the command line while preserving the same deliberate action model as local suggestions.

Set up a provider

Run the interactive provider wizard in a real terminal:

Bash
hokan ai setup

The wizard uses a colored, step-by-step terminal UI: errors are red, successful checks are green, and provider/model choices are easy to scan. It tests the connection before atomically writing the configuration. Set ui.color = "never" or NO_COLOR=1 when a plain terminal is preferred.

Route	Providers
Account sign-in	ChatGPT/Codex, Gemini Code Assist, xAI Grok
Direct API	OpenAI, Anthropic, Gemini, xAI, DeepSeek, Groq, Mistral, Together AI, Kimi/Moonshot (global and China), Z.AI/GLM, StepFun, Alibaba/DashScope, Hugging Face, NVIDIA NIM, Xiaomi MiMo, Tencent TokenHub, Fireworks AI, NovitaAI, DeepInfra, MiniMax (global and China)
Subscription or gateway	OpenCode Go, OpenCode Zen, OpenRouter, Kimi Coding Plan, Alibaba Coding Plan, Ollama Cloud, Vercel AI Gateway, Kilo Code
Local or self-hosted	Ollama, LM Studio, any custom OpenAI-compatible endpoint

The selected provider determines the wire protocol. Direct Anthropic, MiniMax, and Kimi Coding Plan use Anthropic Messages, OpenAI API keys use Responses, and OpenCode Go or Zen switch between Chat Completions, Messages, and Responses according to the selected model. Other gateways remain on their OpenAI-compatible route. Kimi/Moonshot chat requests omit temperature so the provider can apply the model's required server-side setting.

For OpenCode Go or Zen, create a key at https://opencode.ai/auth and select the matching route in the wizard. For OpenRouter, create a key at https://openrouter.ai/settings/keys and select OpenRouter; its model list is loaded from the OpenRouter /models endpoint when available. Kimi Coding Plan keys (sk-kimi-...) use the dedicated Kimi Coding Plan route, while Moonshot platform keys use Kimi / Moonshot.

The wizard also detects the common provider environment variables, including OPENAI_API_KEY, ANTHROPIC_API_KEY, GROQ_API_KEY, MISTRAL_API_KEY, KIMI_API_KEY, KIMI_CODING_API_KEY, ZAI_API_KEY/GLM_API_KEY, DASHSCOPE_API_KEY, OPENCODE_GO_API_KEY, HF_TOKEN, NOVITA_API_KEY, DEEPINFRA_API_KEY, and the gateway-specific variables shown during setup. Accepted values are copied into the private per-provider credential store; manually entered keys are read with terminal echo disabled and are never printed.

Nous Portal, Qwen OAuth, MiniMax OAuth, GitHub Copilot, AWS Bedrock, and Google Vertex AI are not presented as working options yet. They require a dedicated OAuth exchange, external process, or cloud identity flow; their tokens must not be pasted as ordinary API keys.

For scripted configuration, keep the API key in an environment variable:

Bash
export OPENAI_API_KEY='...'hokan config ai --enable \  --endpoint https://api.openai.com/v1 \  --model gpt-5-mini \  --api-key-env OPENAI_API_KEY

Credentials can also be piped from a password manager into a private 0600 credentials file:

Bash
password-manager read openai-key | hokan config ai \  --enable --model gpt-5-mini --api-key-stdin

Review before running

Press Esc to cancel a pending request. Returned commands are inserted into the editable command line. Review the text before running it; high-risk generated commands require a second confirmation.

Local completion remains available without enabling AI. Diagnostic logging is disabled by default. See configuration for what logs contain.