Quickstart

Five minutes from install to a working agent.

1. Connect a model channel

The interactive wizard is the fastest way:

hipmmcode model

It walks you through: pick a channel (OpenAI / Anthropic / Gemini / DeepSeek / Kimi / Qwen / xAI / Codex OAuth / custom / …) → API key or device-code login → live model list → default model. Everything is persisted to ~/.hipmmcode/config.json (OAuth tokens under oauthCredentials).

hipmmcode model deepseek-anthropic   # recommended DeepSeek path (native search)
hipmmcode model xai-oauth            # SuperGrok device-code login
hipmmcode model openai-codex         # ChatGPT Codex subscription

Prefer plain commands? Set a key and defaults directly:

hipmmcode config set anthropic.apiKey=sk-ant-...
hipmmcode config set defaultProvider=anthropic
hipmmcode config set defaultModel=claude-opus-4-8

Environment variables work too: OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, DEEPSEEK_API_KEY, KIMI_API_KEY, QWEN_API_KEY, XAI_API_KEY.

2. Start chatting

hipmmcode

The fullscreen REPL opens. Try:

› summarize this repository and suggest three improvements

The agent reads files, greps, and runs commands as needed. New installs default to Auto permission mode (v0.14.1+): lower-risk calls are classified automatically; higher-risk ones still ask or block. Press y/n when prompted, or Shift+Tab to cycle modes (default / acceptEdits / plan / auto / …).

A few things to try immediately:

Try	What happens
/model	Arrow-key model picker across all configured channels
/help	Command palette — all 75 slash commands
Ctrl+O	Fullscreen transcript view of the whole conversation
Drag a file into the terminal	Attaches it to your next message (images and videos are analyzed on capable models)
@	Fuzzy file mention with completion
Shift+Tab	Cycle ordinary modes: default → acceptEdits → plan → auto; a dangerous launch conditionally inserts bypass

3. Run one-shot from the shell

hipmmcode "what's the latest tokio release?"          # positional prompt → headless exec
echo "review this diff" | hipmmcode --json            # pipe stdin, get a JSON envelope
hipmmcode --stream --max-turns 8 "research X"         # NDJSON event stream

Perfect for scripts, CI, and composing hipmmcode into other tools. See Headless mode.

4. Spawn your first agent team

Inside the REPL:

› Spawn three named background Agents in parallel; they share this session's implicit team:
  one on openai/gpt-4o-mini researching A,
  one on anthropic/claude-haiku-4-5 researching B,
  one on gemini/gemini-2.5-flash researching C.
  Poll them with TaskOutput and aggregate the findings.

Three teammates run in parallel on three different vendors, and the lead aggregates. See Agent teams.

5. Where to go next

• Wire in tools from any MCP server: hipmmcode mcp add --transport http notion https://mcp.notion.com/mcp
• Ensure the default skills pack is present: hipmmcode skill install-defaults (or add more with hipmmcode skill add …)
• Tune permissions (defaultMode: auto is the lean default) or enable the OS sandbox
• Automate with hooks, the server & task queue, or the SDK protocol
• Read the built-in integration handbook: hipmmcode integration | less