← Back to all sparks
L

LobeHub

COMMS
Velocity0.0

Open-source AI chat platform with multi-model support, plugins, knowledge bases, and a marketplace for AI agents.

LobeHub turned its chat client into an agent operator that delegates to Claude Code and Codex.

agent orchestrationcoding agentsself-directed agentsapproval flowgateway
Current state
Five weeks of weekly releases moving in one direction. CAO, the Chief Agent Operator, turns short exchanges into agents that check their own work, decide the next step and continue without prompting, pausing only when a decision genuinely needs a person. Agent Gateway moved agent execution to the server with WebSocket streaming and session resume across disconnects. Server-side human approval landed for agent actions, alongside a personalised Daily Brief and Notion-style document history. Most recently, coding agents themselves became delegable: Claude Code and Codex are created and controlled from inside LobeHub on desktop, with agent-grouped topics and a Review tab aggregating bulk git diffs. Model and provider additions ride along each release.
Where it's heading
The client is becoming a control plane rather than a chat window. Server-side execution, session resume, human approval gates and self-directed agents are the components of something that runs work while nobody is watching — and delegating to Claude Code and Codex means LobeHub is not trying to be the best coding agent, it is trying to be where you supervise the ones that exist. The Review tab reads that way too: diff review is a supervision surface, not an authoring one.
Prediction
Expect more third-party agents to become delegable alongside Claude Code and Codex, and the approval flow to extend into the coding-agent path where reviewing a diff is the natural gate.

Recent moves

  1. 3mo ago

    Delegate Claude Code and Codex

    ⚡ SPARK

    Delegating Claude Code and Codex from inside LobeHub, with agent-grouped topics and a Review tab aggregating git diffs, positions the client as the supervision layer over coding agents rather than a competitor to them.

    View source ↗
  2. 3mo ago

    Claude Code & Codex on Desktop

    Desktop delegation of Claude Code and Codex, screen-capture quick chat, and a batch of new models and providers including a reasoning-effort slider. Model roster additions on top of the coding-agent work that the following release expanded.

    View source ↗
  3. 3mo ago

    Daily Brief, Document History & Approval Flow

    Adds a personalised Daily Brief across agents, Notion-style document history, server-side human approval for agent actions, and expanded Discord support. The approval flow is the notable piece — it is the gate that makes server-side autonomy tolerable.

    View source ↗
  4. 4mo ago

    Agent Gateway & Customizable Sidebar

    Gateway mode runs agents on the server with WebSocket streaming, and sessions reconnect and resume through topic switches and short disconnects. Moves execution off the client, which is what lets an agent keep working when nobody is watching.

    View source ↗
  5. 4mo ago

    AI Auto-Completion & Real-Time Gateway

    AI auto-completion in the editor plus a WebSocket-based Agent Gateway for lower-latency streaming, with Feishu/Lark, Slack and QQ moving to WebSocket delivery. Latency and reliability groundwork ahead of the server-side agent work.

    View source ↗