aibutler 1.1.3

Changes since 1.1.1 (1.1.2 was not published separately, so its changes are listed here too).

1.1.3

• No more getting stuck on big images or long content: images sent to the model are first shrunk to a size it can handle, and text in them stays readable. If the provider rejects an image as too large or the content as too long, the app shrinks the image or cuts the content and sends it again; if it is still too long, it saves what matters to memory, clears the conversation and sends once more. You no longer get a raw English error, and the rest of the conversation no longer keeps failing because of it.

1.1.2

• Create a persona in one sentence: at the top of the new-persona page, write one sentence about what it should do (e.g. "a cook who teaches me simple home dishes") and click "Ask the butler to build it". The butler picks a name, avatar and role, creates it, and tells you what it decided.
• Move your data to another drive: Settings → Advanced → Data folder shows where your data lives, opens the folder, and can "Move elsewhere…". On restart everything is copied over and the switch happens only if the copy succeeds; the old folder is kept so you can delete it once you're sure.
• Personas can hand work off to run in the background: a persona can send one or several helpers to look things up or run commands in the background while it keeps talking with you, and reports back when they finish. Running jobs show in the task panel. Helpers default to a cheaper model from the same provider as the main one.
• Replies even when its window is closed: personas connected to Telegram, Discord and other channels now wake up and reply when a message arrives, even with their window closed. Only one copy of the app can run on the same data folder, so two copies can't corrupt each other's data.
• Easier keys and model picking: keys with a custom address now auto-detect which API to use; you can pick Doubao models yourself; the persona model list shows only chat models, no image or video ones; and the English interface shows Qwen, Doubao and Zhipu by their English names.
• A cleaner chat, and a tip at startup: tool calls are no longer listed line by line in the chat by default; the status bar at the bottom shows what's running. Turn them back on in Settings → General → Tool calls. A usage tip appears at each startup; you can turn it off or see one now, also under General.
• Retries when a reply cuts off mid-way: if the model service starts replying and then drops mid-way, the request is now sent again automatically instead of leaving you with half a sentence.
• User guide updated for 1.1.3, in Chinese and English: latest at https://fantasyxxj.github.io/aibutler/docs/ , this version frozen at https://fantasyxxj.github.io/aibutler/docs-versions/v1.1.3/ .

On Windows, the installer is not code-signed yet; choose "More info" → "Run anyway" if SmartScreen appears.

SHA-256:

• aibutler-1.1.3-arm64.dmg: 9a7e4be6834f3fed1f8c508586af10cecf698e63d5d168bd9b09e40b7cfbbaf3
• aibutler-1.1.3-setup.exe: d2eea7a3b65442397753ca17a9bb4a97fb8d517a69c12733569f6a03dc9a86c0
• aibutler-1.1.3-x64.dmg: fc2d16bab04607ce070cb363d93b30936dc9baccfc85837b666d85eec76e1ca3

aibutler 1.1.1

Changes since 1.1.0.

• Pick a model by how capable you need it, not by its name: when you create a persona, choose Top, Strong, Everyday or Budget under "How capable?" and the app tests your key on the spot to find a model in that tier; models already in use on that key are marked as such. When adding a key in Settings, pick the provider first — the address is filled in, you only paste the key.
• Your butler speaks up: it opens the conversation once a day, and the first time you talk to it each day it mentions one thing that is due soon or unfinished after answering you. The quick buttons above the input box are now situation-based ("Remind me on time", "Look it up and compare"), and the weekly checkup offers to swap out any that went untapped for two weeks.
• Dropped connections are retried: when the model service drops mid-way (including mid-sentence), it is retried up to 6 times and the half-shown text is removed. Only if all 6 attempts fail do you see a short notice.
• Switch key and model from the top bar: turn on "Quick model switch" in Settings → General and a "key · model ▾" chip appears next to the persona name.
• Bing and DuckDuckGo removed from web search: they mostly returned unrelated pages or bot checks. Add Brave, Tavily or your own SearXNG; with none added, the model's built-in search is used.
• Every pop-up window has a visible Close button.
• macOS builds are signed and notarized by Apple: no more xattr workaround — just open the app.
• User guide updated for 1.1.1, in Chinese and English, and now kept per version: latest at https://fantasyxxj.github.io/aibutler/docs/ , this version frozen at https://fantasyxxj.github.io/aibutler/docs-versions/v1.1.1/ .

On Windows, the installer is not code-signed yet; choose "More info" → "Run anyway" if SmartScreen appears.

SHA-256:

• aibutler-1.1.1-arm64.dmg: 9a04ce89c8211a8efb57b6f4009bbf789fcd852c727d938b970cba7288d6027e
• aibutler-1.1.1-setup.exe: 17fc8d3cdca06e0e676951ca354217698a6889b9e7af3b6ce9c83571d886e9f5
• aibutler-1.1.1-x64.dmg: 27d5d32dc8c293a067acf9e73f7af3a5dab0e74f5df5a41d8e3d46a49b3d8c72