EasyAI Work changelog · Easy Series

v0.3.2

2026-09-29 Latest Group wizard and command rules

A four-step wizard for creating group chats: pick members from cards, members are spread across different models by default, and an empty group name is filled in from the topic
Command rules with remembered approvals, so the same kind of command does not ask every time; a Connector tool's risk level follows what you confirmed
A conversation's execution record can be opened at any time
A "Use all installed Skills" switch: every agent can use all Skills installed on this computer without binding them one by one
Conversation compaction can use a model of its own, and writes anything worth remembering into today's notes
Two safeguards on the Advanced page of the Expert editor: maximum tool calls per run and maximum run time
Sleep prevention is configurable: off / only on power (default) / always
Archived conversations can be expanded at the bottom of the list and unarchived
Background Sub-agent cards gain a Stop button that stops only that task

Check for updates only tells you a new version is available; you download it with Download update. The app also checks once on every start
Very long tool results are written to a file and only their beginning, end and path stay in the conversation, saving context on long tasks
A foreground Sub-agent that times out moves to the background and keeps running, and its card says so
While a task list still has open items, the agent sees it at every step and no longer loses track
When a bound Skill is excluded because it failed validation, the conversation says so once and links to the Skills page
Web search summaries share one budget; when a fetched page is empty or thin, the agent is told to try another source
Sub-agents reuse an already running process for the same model and start faster
The Zhihu export Skill is faster: it fills gaps sooner, fetches five pages at a time and slows down by itself when the site pushes back
Approval and waiting texts match what actually happens, with no English or internal names left in the Chinese interface

The Thinking switch beside the input box was back on after every restart (it is now remembered per conversation)
The application log studio.log was never written
Importing an Expert that uses a WorkBuddy model failed with "model.endpoint is invalid"
Connector tools confirmed as having external effects were rejected outright
The restart prompt always showed zero running tasks
After attached Claude Code ended a turn on a network or API error, the desktop kept showing Running and Stop did nothing
Attached Claude Code lost the desktop after /clear in the terminal (it now continues in the new session and moves the Channel binding)
A model set for one member of a group chat did not take effect
The Thinking switch had no effect on WorkBuddy models
The Limits field on the Sub-agents page had a white background in the dark theme

2026-09-27 Resumable background tasks

Long-running background tasks (bulk exports, crawls) keep going: after rate limits, network drops or stalls they resume automatically, even after the app is closed and reopened
A Mac on power does not sleep while a background task is running
Background task cards in the conversation gain a Stop button, and the agent can read a task's live progress instead of reporting stale data
The Zhihu export Skill resumes exactly where it stopped instead of paging from the start, and can skip deleted questions

When a model service glitch (empty response, network drop) interrupts a turn, the agent retries a little later instead of stopping
An agent that repeats the same failing action is stopped early and told to change approach, so the conversation is not rejected by the model service

Background exports were stopped every ten minutes
An error in a turn, or pressing Stop, also killed exports running in the background
The Zhihu export stopped the whole batch on a deleted question or collection
Background task logs stayed empty for a long time, hiding progress
The top of the About dialog was squashed while downloading an update
The Thinking switch beside the input box kept its old label after changing the interface language

2026-09-26

Plug-in management moved into the avatar menu at the bottom left, below About; the activity bar is simpler
The WorkBuddy model bridge setting moved to the end of the Model management group, after Sub-agents
The website, user manual, changelog and feedback links in About point to the new site address

After switching between light and dark themes, some text on the Channels, Claude Code takeover and Group chat pages kept its old colour and was hard to read in the light theme
Some pages only showed a newly selected interface language after a restart; they now switch immediately without touching what you are typing
The WorkBuddy model bridge showed a request address with an extra /v1 in Model settings, so Fetch model list failed

2026-09-26 First public release

First public release: task-centred workspace, multiple model providers, Experts, Skills and the Skill market, MCP Connectors, Sub-agents and group chat, tool approval and auditing, web search, mobile Channels and Claude Code takeover