Setup

Run chanto in a terminal to choose a model or log in. The TUI offers arrow keys and Enter; --plain uses a numbered prompt. API keys are masked and stored in ~/.chanto/auth.json with private permissions (set CHANTO_KEYCHAIN=1 to keep them in the macOS login keychain instead, leaving only a placeholder in that file; toggling the setting back still resolves saved logins). Codex credentials are never reused: run chanto login chatgpt for a separate login.

Detection prefers an explicit Chanto default, then a Codex model with Chanto credentials, hosted credentials in catalog order, local coder models, and other local models. Local probes run concurrently with 300 ms limits: Ollama on 11434, LM Studio on 1234, and llama.cpp/vLLM on 8080. ChatGPT setup lists the models available to your logged-in account. Other recommended hosted models are bundled suggestions; provider access and model availability still depend on your account.

Free options. The picker's Free section lists three ways to start without paying:

• OpenRouter's free models: sign in, then pick from the models OpenRouter currently offers free. Only those that support tool calls and have at least 64k of context are listed, with their context size and how long they're free. Models Chanto has benchmarked come first, marked tested, then coding models. They're rate-limited (about 50 requests a day, 1,000 once the account has $10 of credit), and their providers may log or train on prompts.
• Gemini's free tier: paste a free key from Google AI Studio and pick a model, Flash models first. Google may use free-tier prompts to improve its products.
• A local model: free and private (Ollama, LM Studio, llama.cpp).

Don't use the free hosted options on code you can't share.

chanto setup                         # rerun the picker
chanto setup --plain                 # numbered prompt
chanto setup --model openrouter/anthropic/claude-sonnet-4.5
chanto setup --model qwen3-coder --provider openai --base-url http://localhost:1234/v1
chanto login openrouter              # browser login
chanto login chatgpt --as work       # a second account (`account = "work"` or `/login chatgpt work` to use it)
chanto login openai --with-api-key < /path/to/key-file
chanto login kimi-coding             # a coding plan's key (also zai, minimax-coding)
chanto doctor                        # settings, origins, reachability and fixes
chanto doctor --json                 # stable redacted diagnostic report
chanto models                        # available and recommended models
chanto models openrouter

Setup preserves existing config keys and comments. Non-interactive runs (-p, pipes, CI) never ask setup questions: exactly one usable candidate is selected temporarily with a stderr notice; otherwise pass -m provider/model or run setup in a terminal. /doctor uses the same report as the CLI. /model in the TUI searches all models with fuzzy matching; Enter switches and saves the choice as the default for new sessions (as in Claude Code); Ctrl+S, or /model --session <model>, switches this session only. The list is remembered in ~/.chanto/cache/models.json, so the picker opens at once and refreshes in the background; /models always fetches a fresh list.

To disable reading another agent's settings in place, put this in ~/.chanto/config.toml:

[compat]
claude = false
codex = false

Or set CHANTO_NO_CLAUDE_CONFIG=1 / CHANTO_NO_CODEX_CONFIG=1. Malformed foreign settings are skipped and reported by doctor with their location; they never prevent startup.