ZenMind is a desktop-first AI agent platform for local, web, and mobile workflows.
It packages the core agent runtime into one Desktop experience, connects to modern Chinese model ecosystems such as DeepSeek V4 (deepseek-v4-pro and deepseek-v4-flash), MiMo, MiniMax M3, and Qwen/Bailian, and exposes a custom AGW UI protocol for rich agent interaction.
www.zenmind.cc provides official Desktop installers. You can also build ZenMind Desktop directly from this open-source project.
ZenMind starts from a simple idea: the computer staying at home should become a capable AI agent workspace, not just an idle machine. It is building toward an OpenClaw-inspired, desktop-first platform where local services, web clients, mobile clients, and Chinese model ecosystems work together through one protocol.
- One Desktop app to install, initialize, start, stop, and monitor the ZenMind services.
- Native support direction for DeepSeek V4 (
deepseek-v4-proanddeepseek-v4-flash), MiMo, MiniMax M3, Qwen/Bailian, and MiniMax office workflows. - A custom AGW UI protocol for streaming output, HITL approval, viewport rendering, usage telemetry, and sub-agent invocation.
- A local sandbox layer for long-lived agent sessions, tool environments, and office/document automation.
- Designed for the same agent experience across Desktop, web clients, and upcoming mobile clients.
| Model | Context | Max thinking effort | Official DeepSeek API Docs pricing |
|---|---|---|---|
deepseek-v4-flash |
maxInputTokens: 1048576 (1M context) |
reasoning_effort=max |
cache hit $0.0028, cache miss $0.14, output $0.28 per 1M tokens |
deepseek-v4-pro |
maxInputTokens: 1048576 (1M context) |
reasoning_effort=max |
cache hit $0.003625, cache miss $0.435, output $0.87 per 1M tokens |
The runtime registry uses the current V4 model IDs, maxOutputTokens: 393216, and pricing fields for cache-hit input, cache-miss input, and output tokens. See Model Support for the verified field mapping.
Placeholder: add the ZenMind walkthrough video here.
ZenMind is distributed through ZenMind Desktop. The Desktop app wraps the core services, prepares local configuration, starts the runtime in the right order, and gives users a single control center for the whole system.
Download the official installer from www.zenmind.cc, or package it yourself from this open-source repository.
ZenMind Desktop brings together four core services:
zenmind-app-server: authentication, OIDC, admin console, and app access tokens.agent-platform: agent runtime, model registry, tools, memory, HITL, usage, and sub-agent orchestration.agent-webclient: chat UI, timeline, model switching, viewport rendering, and usage display.agent-container-hub: local sandbox sessions, environment templates, and container-backed tool runtimes.
AGW UI is ZenMind's custom protocol between clients and the Agent Platform. It combines HTTP, SSE, and optional WebSocket transport with a rich event model for agent work:
- H2A streaming and attach recovery.
- HITL modes for question, approval, form, and plan confirmation.
- Builtin and HTML viewports for interactive agent UI.
- Usage snapshots for tokens, cache, reasoning, tool calls, and estimated cost.
agent_invokefor sub-agent tasks that stream back into the main timeline.
Read more in AGW UI Protocol.
See LICENSE.