AI-powered terminal code assistant built with Rust.

• ✅ TUI Interface — Built with ratatui for an interactive terminal experience
• ✅ Streaming Responses — Real-time LLM response streaming with thinking/reasoning support
• ✅ Input History — Shell-like Up/Down navigation with persistent history.jsonl
• ✅ Configuration via TOML — Customizable models, providers, and settings in ~/.ironcode/config.toml

• ✅ Kimi Provider — Native Kimi API support with Coding Agent authentication headers
• ✅ Thinking Mode — Supports <think> reasoning content extraction
• ✅ OpenAI-Compatible Framework — Extensible provider trait for adding new models
• ❌ Context Compaction — Automatic token limiting and summary generation
• ❌ Auto-Retry with Backoff — Exponential backoff for failed requests

• ✅ ReadFile — Read file contents with line numbers and offset/limit
• ✅ WriteFile — Write or append content to files
• ✅ ReplaceFile — String replacement within files (single or multiple edits)
• ✅ Glob — Find files matching glob patterns
• ✅ Grep — Search file contents with ripgrep integration
• ✅ Bash / PowerShell — Execute shell commands with timeout support
• ✅ SearchWeb — Web search via DuckDuckGo
• ✅ FetchURL — Fetch and extract article content from URLs
• ✅ SetTodoList — Track task progress visually
• ✅ AskUserQuestion — Structured question definitions (TUI integration pending)
• ❌ MCP Support — Model Context Protocol server integration
• ❌ Git Tools — Diff, blame, log integration
• ❌ LSP Integration — Code completion and go-to-definition
• ❌ AST-Aware Tools — Code parsing and semantic analysis

• ✅ Session Persistence — JSONL-based storage (meta.json + context.jsonl)
• ✅ Resume by ID — ironcode --session <ID> to resume a specific session
• ✅ Resume Latest — ironcode --continue to continue the most recent session
• ✅ Auto-Save — Automatic persistence on messages, tool results, and history clears
• ✅ Session Metadata — Title, timestamps, and message tracking
• ❌ Session List UI — Browse and load historical sessions from within the TUI
• ❌ Checkpoints / D-Mail — Snapshot and rollback to previous conversation states

• ✅ Custom Config Directory — -c / --config to specify config location
• ❌ Print / Non-Interactive Mode — --print for single-shot queries and piping
• ❌ YOLO Mode — Auto-approve all tool executions
• ❌ Approval System — Diff previews and per-tool auto-approval settings

• ❌ Multi-Agent / Subagent System — Task delegation and LaborMarket-style agent pools
• ❌ Skill System — Reusable skills and flow-based workflows
• ❌ Web UI — FastAPI + WebSocket alternative interface
• ❌ ACP Protocol — Agent Client Protocol for IDE integration
• ❌ OAuth Authentication — Device flow token management
• ❌ Vision / Image Input — Multimodal message support

# Run with a new session (default)
ironcode

# Continue the most recent session
ironcode --continue

# Resume a specific session
ironcode --session <SESSION_ID>

# Use a custom config directory
ironcode -c /path/to/config

Place your config.toml in ~/.ironcode/:

default_model = "kimi/kimi-for-coding"

[providers.kimi]
type = "kimi"
base_url = "https://api.moonshot.cn/v1"
api_key = "${KIMI_API_KEY}"

[models."kimi/kimi-for-coding"]
provider = "kimi"
model = "kimi-for-coding"
max_context_size = 128000
supports_streaming = true

# Context compaction settings (auto-compression when approaching token limits)
[compaction]
enabled = true
trigger_ratio = 0.85
reserved_context_size = 50000

The [compaction] section controls automatic context compression to prevent exceeding the model's context window:

• enabled — Enable auto-compaction (default: true)
• trigger_ratio — Token usage ratio threshold (0.5-0.99) that triggers compaction (default: 0.85)
• reserved_context_size — Reserved tokens for LLM response generation (default: 50000)

Compaction triggers when either condition is met:

• current_tokens >= max_context_size * trigger_ratio
• current_tokens + reserved_context_size >= max_context_size