I'm building AgentHub: a multi-model, multi-agent terminal workspace for real engineering work.
AgentHub keeps the practical Codex-style coding workflow — repository context, shell execution, file search, patch editing, approvals, sandboxing, tool calls, and resumable threads — then extends it with multi-provider routing, sidecar tabs, delegated agents, background teammates, plugins, and headless automation.
- GitHub: https://github.com/hoteye/AgentHub
- Latest CLI release: https://github.com/hoteye/AgentHub/releases/tag/cli-v0.1.11
- One-line install for Linux/macOS:
curl -fsSL https://raw.githubusercontent.com/hoteye/AgentHub/main/scripts/install_agenthub_cli.sh | bashRun it:
agenthub-cli
agenthub-cli --headless --prompt "/provider"AgentHub is not just another chat UI. It is a terminal-based engineering workspace designed to let coding agents participate in real repository workflows:
- read, search, and modify project files
- run shell commands and tests
- apply patches and review diffs
- preserve and resume long-running task context
- switch between OpenAI, Anthropic, GLM, DeepSeek, and other providers
- delegate scoped work to child agents or background teammates
- extend workflows through plugins and gateway integrations
- run interactively in TUI mode or non-interactively in headless/server mode
Single-agent coding tools are useful, but real software work is rarely a single prompt. A serious task usually needs context, tool use, approvals, tests, model switching, decomposition, review, and recovery after interruption.
AgentHub is my attempt to make that workflow explicit and reliable: a recoverable, auditable, extensible runtime for multi-model coding agents.
- CLI/TUI usability
- Provider compatibility and diagnostics
- Codex-style sidecar execution
- Multi-tab and multi-agent orchestration
- Startup performance and portable releases
- Plugin architecture for GitHub, browser, document, and domain-specific tools
我正在做 AgentHub：一个面向真实工程任务的多模型、多工具、多 Agent 终端工作台。
它保留 Codex 风格的代码工作流能力，包括仓库上下文、shell 执行、文件搜索、patch 修改、审批策略、沙箱控制、工具调用和会话恢复；同时扩展多 provider、多 tab、子 Agent、后台任务、插件和 headless 自动化能力。
如果你关注 AI 编程助手、Agent 编排、多模型协作或终端开发工具，欢迎试用和反馈：