更新日志
AgentConnect 的新功能、改进与重要修复。
在 GitHub 查看所有版本 ↗2026年8月
v1.30.0
- Session Retention — Choose how long a daemon keeps completed sessions, including a custom number of days or no expiry.
- Workspace Management — Browse session worktrees and automatically retire workspaces for merged pull requests and closed issues.
- Model Usage Analytics — Break down usage by model to understand where agent costs come from.
Behavior change: The Slack status bar is off by default unless explicitly enabled.
Upgrade note: Deploy the Control Plane and daemons together. Hand-authored integrations must use the new
coreandconfigformat.v1.29.0
The Console's conversation rail and chat layout work better across screen sizes.
Upgrade note: Legacy integration and interaction message formats are retired. Upgrade the Control Plane, relay, and daemons together to avoid dropped interactions or rejected integration configuration during a mixed-version rollout.
v1.28.0
- Cross-platform Scheduling — Send scheduled tasks and webhook-triggered work to any connected chat platform.
- Improved Webchat — Attach files from the home composer and follow more of the agent's work as it happens. The Console also shows locally installed skills and their origins.
v1.27.0
- Organization Variables & Secrets — Define shared configuration once and assign it to all agents or selected agents.
- Session Workspaces — Give sessions separate Git worktrees and clean up expired workspaces automatically.
Upgrade note: Node.js 24.12.0 or newer is required.
Behavior change: Hand-authored per-user integration allowlists are no longer supported; platform access follows conversation admission rules.
v1.26.0
- Unified Conversations — See one conversation across its participating agents, with combined activity, usage, and session history.
- Memory Dreaming Restored — Dream execution and review return, with adoption tied to the exact content you reviewed.
This release also improves Lark identity linking and recovery from skipped or failed GitHub reviews.
v1.25.0
- Multi-agent Webchat — Bring several agents into one conversation, with shared context and replies attributed to each participant.
- Organization Knowledge & Skills — A shared Knowledge library brings approved knowledge and reusable skills together, with revision history and per-agent skill selection.
- Linux Sandboxing — Run agents with a private home and restricted filesystem writes, with added protection for Codex credentials from model tools.
- Improved Webchat — Use remote MCP servers and manage agent permissions from the conversation.
Behavior change: Dream execution and staged-content review are temporarily paused while their credential and adoption protections are completed.
2026年7月
v1.24.0
Organization members can leave without orphaning agents or other resources: ownership transfers to a remaining organization owner. Adding a daemon also lets you name it and choose who can access it during setup.
v1.23.0
- Agent Templates & Guided Setup — Start with a built-in AgentConnect assistant, connect it through Add to Slack, and follow a persistent setup checklist.
- Lark Setup — Create and connect Lark or Feishu apps through a guided flow, with HTTP callbacks and streaming replies.
- Dream Skill Review — Review, accept, or dismiss skills proposed by memory dreaming directly in the Console.
- Slack Sign-in — Sign in with Slack and manage linked sign-in methods from your profile.
v1.22.0
Agents now work in Slack group DMs, with mentions to start a conversation and follow-up replies in the same thread. Slack also gains a Manage Session shortcut, and background tasks deliver their completed results more reliably.
Upgrade note: Existing Slack apps need the new group-DM permissions and event subscription; update the app configuration and reauthorize it to use this capability.
v1.21.0
- Guided Setup — A resumable first-run wizard walks you through connecting a daemon, creating an agent, and adding an integration.
- Usage Analytics — Track spend over time and compare usage by agent or runtime in your local timezone.
- Agent Collaboration — Agents can follow delegated sessions and receive results when their peers finish.