🚀 HugAgentOS Desktop 1.0.1

🚀 HugAgentOS Desktop 1.0.1

The first update since the 1.0.0 milestone. It delivers the personal evolution controls that 1.0.0 listed as coming next, gives agents working in group chats the ability to actually see the group, and governs what the long-term memory is allowed to keep.

🔖 Build note: artifact names now carry the product version directly (1.0.1). The 1.0.0 release retained the internal build number 0.2.9 in its file names; from this release on, the two are the same. Existing installations update in place through the built-in updater.

✨ Highlights

🧬 Personal evolution

The 1.0.0 release noted that personal-evolution controls were planned for the next version. They are here.

• Decide which mechanisms are allowed to run for your account: memory growth (settling facts and preferences into long-term memory), skill consolidation (turning recurring approaches into reusable skills), and orchestration correction (adjusting step order and process strategy).
• Review capability candidates that were derived from your own work, and approve them individually before they become active for you.
• See which of your conversations contributed to a capability, and inspect what was settled for any single turn.
• Optionally validate consolidation against a domain ontology, either across all ontologies or a chosen package — evolution still works normally without one.
• Run an evolution cycle on demand instead of waiting for the scheduled one, and turn the whole thing off in one switch.
• Skills are compiled from layered evidence with graph-memory support, promotion of tool-sequence skills is gated on genuine procedural commonality, and only candidates that can actually be carried out are offered for approval.

🧠 Governed long-term memory

• Writes to vector memory are now deduplicated, given tiered retention, and passed through a write gate, so long-lived memory accumulates what matters rather than everything said.
• Local long-term memory works as soon as the models are configured, with no extra setup.

💬 Group chat that sees the group

An agent added to a Feishu or DingTalk group could previously see only messages that @-mentioned it, plus its own replies — everything else the group said, and every file it shared, was invisible.

• Group listening (opt-in, per bot). Messages that do not address the bot are recorded as background context and never replied to, then folded into the next mention so the agent answers with the conversation in view. Off by default, with the privacy implication stated next to the switch.
• Group history is pulled deterministically by code, on every mention, rather than depending on the agent to ask for it. Pulls are incremental from a cursor; only the first reaches back over a 7-day cold-start window, and it takes the newest end of that window rather than week-old chatter. Every pull is bounded by message count, character budget, and a hard timeout, and stops being attempted after repeated failures.
• Files and documents are fetched on demand. Files seen in a group are recorded by handle rather than downloaded, and retrieved individually when the agent decides one matters. DingTalk attachments are supported for the first time, along with files shared through DingTalk Drive and online documents shared as link cards.
• Failures say what actually happened. When a platform refuses for a specific reason, that reason reaches you verbatim instead of being flattened into a generic download failure — a cross-organization refusal reads nothing like an expiry and cannot be fixed by retrying.

Group listening and history pulling depend on permissions granted in the Feishu or DingTalk admin console, and the DingTalk pull path additionally needs a connected DingTalk account. With neither configured, group behaviour is unchanged: the bot answers when mentioned, exactly as in 1.0.0.

⏰ Scheduled tasks deliver the whole result

Scheduled reports pushed to a chat channel were cut to 500 characters before being stored, so every delivered report stopped after its opening paragraph with nothing marking it as truncated. The complete result is now stored and delivered; truncation happens only where results are displayed, with consistent limits across the run history, notification cards, and model context.

🧭 Interface

• The sidebar was reworked: capabilities collapse into a nested capability center, scheduled tasks and sites get dedicated entries, and the knowledge base moved into My Space. The layout remains configurable.
• Files can be dragged straight into the chat composer to upload.
• The desktop client now defaults to Local mode during first-run setup.
• Fixed IME composition being lost in an empty composer, and loop status appearing in the project composer.

🛠️ Reliability

• Hardened guardrails for long-running tasks, including model watchdogs, iteration budgets that force an orderly finish, and sandbox keep-alive.
• The plan bar now finalizes when a run completes, instead of leaving steps visibly unfinished.
• Channel replies no longer leak inline <think> reasoning to DingTalk, Feishu, or WeChat, and channel-bound agents no longer end a turn without replying in some terminal states.
• Fixed the capability health view failing outright whenever a built-in skill was found never to have been opened.
• Registered the MCP marketplace routes in the Community Edition.

📦 Downloads

Platform	Package	SHA-256
Windows x64	HugAgentOS_1.0.1_x64-setup.exe	46446cea7d06b9c3820ac2e3d595724fb505c9401c136c98127ecf16ca2a0354
macOS Apple Silicon	HugAgentOS_1.0.1_aarch64.dmg	e151f1ad89e9d78960cbd347fcd209297351080d834ef78bb8bcacec768efb0c
macOS Intel	HugAgentOS_1.0.1_x64.dmg	15006f9633855919d35c0ae5913e59507df45700327f8edd9440dd8215a68d82
Linux x86_64	HugAgentOS_1.0.1_amd64.AppImage	4bd0b702e407d700a49154b30d50378321fb5f4d52bcb2ccca3d7dbc973f4eec
Debian / Ubuntu x86_64	HugAgentOS_1.0.1_amd64.deb	0ac704292acef6863bb16e1f4639077e17e72aff096aab3be253d05d3a541a47

The .sig files, macOS .app.tar.gz archives, and latest.json are updater artifacts and are not needed for a normal first-time installation. Linux automatic updates use the AppImage package; users of the .deb package should install future versions manually.

🚀 Upgrading from 1.0.0

Open the built-in update checker and install the signed update; user data is preserved. A fresh installation follows the same steps as 1.0.0 — download the installer for your operating system, then choose Local, Cloud, or Dual mode during first-run setup.

🍎 macOS installation note

The macOS packages in this release are ad-hoc signed but not Apple-notarized. On first launch, macOS may block the application because it cannot verify the developer. Open System Settings → Privacy & Security, find the HugAgentOS notice, and choose Open Anyway. Download packages only from this official repository.

❤️ Thank you for using HugAgentOS. Bug reports, feature requests, and community contributions are welcome through GitHub Issues.