LobeHub 2.3 Lets Every Bot Decide When to Join a Channel Conversation, and Logs Each Call It Makes

LobeHub has changed the default social rule for its bots: they no longer wait to be called on. Version 2.3.0 of the open-source agent workspace, released on GitHub on November 1, 2026, turns a feature called Proactive Bot on for every bot. The release notes put it plainly: “Bots stop waiting to be @-mentioned.”

LobeHub describes itself on GitHub as “your Chief Agent Operator,” a workspace that organizes a team’s agents into round-the-clock operations. The repository has more than 83,000 stars. The 2.3.0 release collects 145 merged pull requests from seven contributors since v2.2.21, and its notes are marked as generated with Claude Code.

From preview flag to default

Bot wake, the mechanism behind Proactive Bot, was previewed behind a rollout flag in v2.2.21. In 2.3.0 the flag is gone. Each bot now “judges each message in the channels it watches and decides for itself whether to speak up, stay quiet or wait.”

Owners keep an off switch. A new Proactive participation block sits at the top of the Wake policy tab in channel settings and saves as soon as it is changed. With it off, the Duties, Channels and Pace options disappear, and only watch keywords can still wake the bot.

When a bot decides to speak

The release spells out when a bot will step in. If its owner has written duties for it, the bot no longer waits for a message that falls inside them. It also joins in when someone leaves an open question or problem that nobody has answered and the bot can settle it.

Two smaller rules shape the behavior. A line that comes right after a person’s message is passed to the model rather than skipped, and the bot still stays out of replies addressed to someone else. LobeHub says the change “adds no extra model call,” which matters for teams paying per token across many bots and busy channels.

An audit trail for every judgment

Letting bots choose when to talk raises an obvious question: why did it say that, or why did it stay silent? LobeHub’s answer is a new Wake records tab in bot settings. It lists every message the bot judged, with a 30-day summary of how many it woke on, skipped, deferred or treated as proactive wakes.

For each message, the tab shows the outcome, the model’s score and “a readable reason,” with a link back to the original message on Discord. Owners can filter by outcome or show only proactive wakes, and mark each decision as right or wrong. The release does not say whether those marks feed back into the bot’s later choices.

Failures in chat platforms also get clearer messages. Replies now explain sandbox dispatch errors and Feishu error 41050 instead of reporting a bare failure, and internal metadata is kept out of public replies.

Channels, memory and goals

Much of the rest of the release builds the space bots now speak into. Pages support real-time editing by several people, with page Agent edits running in a sandbox. Channel conversations gain quoted replies and threaded Agent replies. Every new Agent gets a direct channel, every new project a General channel, and in a group with only one Agent, every message goes straight to it.

Memory is reorganized into contexts and dossiers shown as “stories,” and an interrupted consolidation run now resumes. Dashboards gain a declarative card layout and a createDashboard tool. Goals can be shared across a workspace, and the persistent sandbox leaves preview.

For agents running elsewhere, LobeHub now shows Claude Code project memory in its workspace panel and picks the Claude Code account per device for Agent quota. The lh command-line tool gains commands to wake or start conversations, and a generic text-to-speech model type arrives with MiniMax t2a_v2 support.

Notes for self-hosters

Self-hosted installs pick up six database migrations, 0193 to 0198, adding goal and file sharing tables, structured user memory tables and new memory indexes. The notes flag one: migration 0196 builds indexes on the user_memories_* tables inside a transaction, which blocks writes to those tables while it runs. For large tables, LobeHub suggests creating the same indexes with CREATE INDEX CONCURRENTLY IF NOT EXISTS before upgrading. The release also pins the OIDC provider’s request host to APP_URL.

Proactive bots are a bet that teams want agents that behave more like colleagues than command lines. The Wake records tab is LobeHub’s acknowledgment that such a bet needs a paper trail, and the per-channel switch is there for teams that would rather keep the old rule.