EasyAI Work
A local-first desktop AI agent workspace
Hand tasks to agents that run on your own computer: pick any model provider, install and swap Experts and Skills, plug in MCP Connectors in one click, and let Sub-agents work in parallel. Every tool call is approved and audited, and your sessions, history and keys stay on your machine.
Beta: ready for everyday use, while screens and settings may still change between versions. In-app Check for updates delivers each new version right away.
Download EasyAI Work
The installer is not code-signed: if SmartScreen warns, choose Run anyway.
The package is ad-hoc signed: on first launch, right-click the app in Finder and choose Open.
EasyAI Work — New taskWhat can EasyAI Work do for you today?
Click the input box to type your own instruction
What EasyAI Work can do
Task-centred workspace
Agent, Ask and Plan modes. Thinking and execution appear step by step as cards, and deliverables preview right beside the conversation.
Any model
Five protocols (OpenAI-compatible, OpenAI Responses, Anthropic, Gemini, local Ollama) and 12 preset providers. Keys live only in the system keychain.
Experts
Role agents with a persona and memory documents that can wake up on a schedule. Import and export as JSON to share with your team.
Skills and the Skill market
One-click install, built-in integrity checks, enabled only when needed; scripts bundled with a Skill run safely on your machine.
MCP Connectors and market
OAuth 2.1 sign-in with My services and Market tabs; command-line Connectors sit in the same list as MCP servers.
Sub-agents and group chat
explore, worker and fork Sub-agents run in parallel; several agents discuss one topic in a Group chat and hand back a conclusion card.
Governance and approval
Built-in tools are denied by default, approved one call at a time and fully audited; permission requests from the terminal can be decided remotely.
Web search and fetch
Eight search engines to choose from, with traceable result cards; web fetch brings the source text back into context.
Mobile Channels and takeover
Follow progress, answer questions and approve actions from Telegram or Lark, and take over Claude Code sessions running in your terminal.
Where it fits
Write documents and reports
Drop in your material, say what you need, and the agent researches, organises and drafts, with a live preview beside the chat.
Tidy local files and data
Rename in bulk, sort, extract tables — all on your machine, and every step can require your approval.
Scheduled Expert check-ins
Have an operations Expert check your services every morning and push the verdict to your phone.
Connect internal systems
Hook up ticketing, knowledge bases and databases with MCP Connectors so the agent can query them directly.
Steer tasks on the road
A task runs on your desktop while you follow it, answer its questions and approve its actions from your phone.
Share know-how across the team
Package Experts, Skills and signed plug-ins for colleagues so everyone works the same way.
See it in action
Before you install
- Operating system
- macOS (Apple Silicon); Windows 10 / 11 (64-bit) coming soon
- Memory
- 8 GB or more recommended
- Disk
- About 1 GB of free space
- Runtime
- Bundled JDK 21; no Java installation needed
- Network
- Needed for online models and web search; local Ollama models work offline
- Models
- An API key for at least one model provider, or Ollama on this machine
- Interface language
- Simplified Chinese / English
The macOS package is ad-hoc signed: on first launch, right-click the app in Finder and choose Open.
About EasyAI Work
Do I need my own model API key?
Yes. EasyAI Work does not provide model access. Connect a preset provider such as DeepSeek, Zhipu, Qwen, Kimi or SiliconFlow, any OpenAI-compatible service, or a local Ollama.
Where are my data and keys stored?
Sessions, history and settings live in a local SQLite database. API keys are kept only in the system keychain (macOS Keychain / Windows Credential Manager) and never written to a file.
How does it relate to Claude Code or Cursor?
Those are built for writing code; EasyAI Work is built for everyday work tasks. It can also take over a Claude Code session in your terminal so you can follow progress and grant permissions from the desktop or your phone.
Can I use it offline?
The interface, history, Experts and Skill management all work offline; online models and web search need a connection. With a local Ollama model it can run fully offline.
How do updates work?
About → Check for updates reads the update manifest from this site and verifies the signature and SHA-256 before downloading. You can also download the latest installer here at any time.
Download EasyAI Work
v0.3.2 · Windows / macOS