We build command-line tools for orchestrating, managing, and scaling AI agent workflows.
Every tool ships with llms.txt and JSON APIs — designed for agents to read, not just humans.
aid (ai-dispatch)
Multi-AI CLI team orchestrator
Dispatch work to gemini, codex, opencode, cursor, kilo, ob1, and codebuff from one CLI. Track progress, manage worktrees, enforce methodology, and control cost.
- Unified dispatch to 7+ AI CLIs
- Task tracking with SQLite persistence
- Auto agent selection via classifier + capability matrix
- Git worktree isolation for parallel tasks
- Batch dispatch with DAG dependencies
- TUI dashboard with live event streaming
- MCP server for tool-use integration
- Cost tracking and budget enforcement
- Skills/methodology injection
- Retry with feedback loops
ai-summary
Web search & summarization compression proxy
Compresses web content for AI coding agents — search, fetch, extract, and summarize in one CLI call. Replaces 3-5 Claude tool calls with a single command, reducing token consumption by 10-50x.
- Search + Summarize via Gemini/DDG/Brave
- Fetch + Summarize any URL with readability extraction
- Fast no-LLM text compression
- Claude Code PreToolUse hook (real token savings via command rewriting)
- Test output compression (cargo test, npm test, pytest, go test)
- Multiple LLM backends (opencode free, oMLX local, OpenAI-compatible)
- Jina Reader API fallback for JS-heavy pages
- Rich token savings statistics with ROI tracking
- JSON output, stdin piping, standard exit codes
- Cloudflare Browser Rendering crawl support
- 268 tokens vs 5,000-50,000 raw (10-50x reduction)
hiboss
AI Agent ↔ Boss communication tool
Let AI agents send messages to their human boss and receive replies through Discord and Telegram. Agents call hiboss send or hiboss ask — the message lands in your chat app, and your reply flows back to the agent. Agents can also boss other agents for autonomous multi-agent workflows.
- Agent-to-boss messaging via Discord & Telegram
- Blocking ask with quick-reply buttons
- Action buttons that trigger shell commands
- Priority-based channel routing
- Agent-as-boss for multi-agent orchestration
- Cross-session agent-to-agent messaging
- Session auto-discovery and peer listing
- MCP server for AI coding tools
- File attachments via R2 storage
- Self-hosted on Cloudflare Workers + D1
crpc
Chain-aware RPC CLI — like cast, but better
A blockchain RPC command-line tool with auto ABI encode/decode, 2600+ chains via chainlist.org, Etherscan V2 integration, Multicall3 batching, smart RPC fallback, and offline encode/decode.
- Auto ABI encode/decode — no manual hex
- 2600+ chains via chainlist.org, zero config
- Etherscan V2 unified API for 60+ chains
- Batch Multicall3 — multiple calls in one RPC request
- Smart RPC fallback with L2 auto-detection
- Offline encode/decode without network
- 18 commands: call, balance, tx, logs, trace, gas, and more
- Built-in token aliases and chain name resolution
- Transaction decoding with event log parsing
- Storage slot reading and cross-block diff
ghost-code
Ghostty + Claude Code Telegram bridge
Get notified on your phone when Claude finishes a task or needs input — reply to inject text back into the terminal, or chat with Claude directly via streaming responses. Hooks into Claude Code's lifecycle events with a single binary.
- Task completion summaries via Telegram (silent notifications)
- Attention request forwarding with details
- Tool approval via inline Allow/Deny buttons
- Real-time statusline (model, cost, plan usage, context, aid stats)
- Streaming chat — send messages to Claude from Telegram
- Terminal injection — replies route back to originating Ghostty tab
- Multi-machine support with hostname identification
- Noise filtering for quota recovery and system events
- Sleep prevention with automatic caffeinate
- Single-instance flock-based daemon with graceful shutdown