Bring every AI coding agent to every place your team works.
HotPlex is a self-hosted gateway that connects AI coding agents to Web, Slack,
Feishu, and enterprise messaging through one consistent, production-ready interface.
简体中文 | English
Quick Start · Use Cases · Documentation · Changelog
AI coding agents are powerful, but they usually live inside one terminal, one vendor, or one developer's workflow. HotPlex puts a stable gateway in front of them so the same agents can serve a browser, a team chat, an automation, or an internal platform—without rebuilding the agent integration each time.
| What your team gets | Why it matters |
|---|---|
| One gateway for every agent | Connect Claude Code, Codex CLI, OpenCode Server, or any ACP-compatible agent through the same AEP v1 interface. |
| Access where work already happens | Use agents from the embedded Web Chat, Slack, Feishu, or Yuanxin instead of moving every request back to a terminal. |
| Conversations that survive real work | Resume deterministic sessions, stream long-running turns, and accept follow-up input even while an agent is busy. |
| Control you can operate yourself | Self-host authentication, permissions, audit trails, persistence, metrics, tracing, and lifecycle management in one Go binary. |
Give a Slack or Feishu channel access to a real coding agent. Teammates can investigate code, ask follow-up questions, approve tool use, and receive streamed results without sharing a developer workstation.
Continue an agent session from the embedded Web Chat or a mobile messaging client. Deterministic session identity and persisted history keep the conversation connected across network drops and restarts.
Turn natural-language requests into one-time reminders, recurring checks, or scheduled engineering jobs. HotPlex runs them through the selected agent and delivers the result back to the target platform.
Place one governed boundary between users and multiple agent runtimes. Centralize API authentication, permission ceilings, bot configuration, session visibility, audit records, and operational telemetry.
macOS / Linux:
curl -fsSL https://raw.githubusercontent.com/hrygo/hotplex/main/scripts/install.sh | sudo bash -s -- --latestFor a user-local installation without sudo, add --prefix ~/.local instead. See the installation reference for Windows, source builds, Docker, and version-pinned installs.
Windows (PowerShell 5.1+):
Invoke-WebRequest -Uri https://raw.githubusercontent.com/hrygo/hotplex/main/scripts/install.ps1 -OutFile install.ps1
.\install.ps1 -Latesthotplex onboard
hotplex doctorThe setup wizard detects installed agents, creates the local configuration, collects required secrets, and optionally enables Slack or Feishu. It is safe to run again when you want to reconfigure an existing installation.
hotplex gateway start -d
curl http://localhost:8888/healthOpen http://localhost:8888 for the embedded Web Chat. The same gateway can then be connected to Slack, Feishu, or Yuanxin.
Tip
Building from source? Clone the repository, run make hooks, then make quickstart. The full workflow is in the contributor setup guide.
HotPlex separates the place where a user talks from the agent that performs the work:
- Clients and messaging platforms send AEP events through WebSocket or a platform adapter.
- HotPlex Gateway authenticates the user, resolves the session and agent policy, persists lifecycle state, and streams events with backpressure and ordered sequence numbers.
- Worker adapters translate the unified session into the native protocol of Claude Code, Codex CLI, OpenCode Server, or an ACP agent.
- Results flow back through the same session to the originating browser, chat, or scheduled task.
| Channel | Experience |
|---|---|
| Embedded Web Chat | Built-in Next.js interface with streaming conversations, workspace controls, and an admin console. |
| Slack | Socket Mode adapter with streamed replies, slash commands, interactions, and file tooling. |
| Feishu | WebSocket adapter with interactive cards, commands, voice input, and voice summaries. |
| Yuanxin | Pulsar-backed enterprise messaging adapter with session routing and scheduled-result delivery. |
| Worker | Best fit |
|---|---|
| Claude Code | Full coding-agent sessions with tool interactions and mid-turn follow-up injection. |
| Codex CLI | Codex app-server sessions with streaming events and mid-turn follow-ups. |
| OpenCode Server | Long-lived OpenCode HTTP/SSE runtime managed behind the gateway. |
| ACP | Any Agent Client Protocol-compatible runtime over JSON-RPC 2.0 stdio. |
Worker choice can be set per bot or platform, with shared defaults for the rest of the deployment.
- Keep talking while an agent is busy. Follow-up input is injected into the current Claude Code or Codex CLI turn; ACP and OpenCode Server supplements are buffered and replayed after completion.
- One protocol contract across every SDK. A canonical AEP v1 schema and 38 golden envelopes are checked against Go, TypeScript, Python, and Java clients in CI.
- Operate from a complete admin experience. The v1.38 line includes dashboards for sessions, bots, cron jobs, users, API keys, skills, and activity.
- Resume with stronger reliability. Session sequence hydration and ACP resume fixes prevent history loss and event-sequence collisions across reconnects.
See the changelog for the complete release history.
- Security: timing-safe API key checks, permission ceilings, SSRF and DNS-rebinding protection, path safety, and isolated worker environments.
- Persistence: deterministic session identity, event history, and lifecycle storage on SQLite by default or PostgreSQL for shared deployments.
- Observability: structured JSON logs, Prometheus metrics, OpenTelemetry traces, and W3C TraceContext propagation.
- Operations: one cross-platform binary with 14 top-level commands, 27 diagnostic checks, hot-reloadable configuration, service management, and self-update support.
- Cross-platform: Linux, macOS, and Windows support with native process and service lifecycle handling.
| Language | Start here |
|---|---|
| Go | Go client SDK |
| TypeScript | TypeScript client |
| Python | Python client |
| Java | Java client |
All four clients consume the same AEP v1 contract and conformance corpus. Protocol details are available in the AEP reference and event catalog.
| Goal | Guide |
|---|---|
| Get running in five minutes | Getting Started |
| Connect a team chat | Slack Integration · Feishu Integration |
| Configure workers and platforms | Configuration Reference |
| Deploy and operate HotPlex | Enterprise Deployment · Observability |
| Integrate a custom client | WebSocket Integration · AEP v1 Protocol |
| Automate recurring work | Cron Scheduled Tasks |
| Manage the gateway | CLI Reference · Admin API |
HotPlex also serves its Chinese-first documentation portal directly from the binary at http://localhost:8888/docs.
Contributions are welcome. Start with CONTRIBUTING.md and the development setup guide. Install the repository hooks with make hooks before making changes.
Do not open a public issue for a suspected vulnerability. Follow the private reporting process in SECURITY.md.
HotPlex is distributed under the Apache License 2.0.