╭━━╮╭━━━╮╭━━━╮╭━╮╱╭╮ ╭━━━╮╭━━━╮╭━━╮╭━╮╱╭╮ ╰┫┣╯┃╭━╮┃┃╭━╮┃┃┃╰╮┃┃ ┃╭━╮┃┃╭━╮┃╰┫┣╯┃┃╰╮┃┃ ╱┃┃╱┃╰━╯┃┃┃╱┃┃┃╭╮╰╯┃ ┃╰━╯┃┃┃╱┃┃╱┃┃╱┃╭╮╰╯┃ ╱┃┃╱┃╭╮╭╯┃┃╱┃┃┃┃╰╮┃┃ ┃╭╮╭╯┃╰━╯┃╱┃┃╱┃┃╰╮┃┃ ╭┫┣╮┃┃┃╰╮┃╰━╯┃┃┃╱┃┃┃ ┃┃┃╰╮┃╭━╮┃╭┫┣╮┃┃╱┃┃┃ ╰━━╯╰╯╰━╯╰━━━╯╰╯╱╰━╯ ╰╯╰━╯╰╯╱╰╯╰━━╯╰╯╱╰━╯
Route tasks to the right model at the right time. Use any provider — Anthropic, OpenAI, Ollama, or your existing CLI subscriptions.
Orchestrate multiple models in parallel. The main model delegates exploration and execution to specialized slots, then integrates results.
Works with Anthropic, OpenAI, Ollama, LM Studio, or any OpenAI-compatible API. Use CLI subscriptions too — Claude Code, Codex, Gemini CLI.
Automatically routes file edits to your execute slot, grep/search to explore, and strategy to main. Configurable per tool type.
Generate a PRD and task breakdown with /plan. Review, approve, and auto-execute through Forge with optional auto-commit per task.
Run tasks repeatedly with /loop until a condition is met. Auto-detects when stuck and suggests new strategies.
Auto-discover markdown skill files as slash commands. Connect MCP servers for external tool integration.
SQLite-backed sessions with cross-session lessons. Thinking level control (off/low/medium/high) per slot across all providers.
Inject files, directories, git state, or images into prompts with @./file, @git:diff, @dir:src/. Multimodal image support included.
Add context while the agent is streaming. The response pauses, your input is injected, and generation resumes with full history.
Responsive multi-agent display with status dots, tool-call trees, and per-card stats. Adapts to terminal width automatically.
Extend Iron Rain with plugins. Hook into dispatch lifecycle events, modify tasks, and add custom behaviors.
Run /init to analyze your codebase — maps repo structure, extracts tech stack and conventions, and stores findings as persistent lessons.
Review staged changes or branch diffs with /review. Get structured feedback on bugs, security, performance, and style.
Git-based checkpoint system with /undo to restore previous states. Automatic checkpoints during plan execution.
Per-model cost registry with configurable input/output token pricing. Track spend across slots and sessions.
Define coding standards in IRON-RAIN.md or .iron-rain/rules/. Rules are injected into every system prompt automatically.
Auto-generated repository map with file structure and extracted symbols. Gives the model structural awareness of your codebase.
Run commands in isolated sandboxes — macOS Seatbelt, Docker, or gVisor. Control network access and write paths.
Mistype a command? Iron Rain suggests the closest match using Levenshtein distance. /upgade → "Did you mean /update?"
Each slot serves a distinct purpose. Assign any model to any slot.
Handles conversation, planning, and high-level reasoning. Your most capable model goes here.
Read-only exploration. Grep, glob, search, and codebase understanding. Can be a fast, cheap model.
File writes, bash commands, and code generation. Needs a model strong at following precise instructions.
Iron Rain is part of the howlerops family of open-source AI infrastructure tools.