--- title: "Use DeepAgents in Lark — setup and first task — AgentSky" description: "Use DeepAgents in Lark through AgentSky. Test the agent in the shared Composer, connect the channel, review permissions, and verify the first real task." url: https://agentsky.dev/agents/deepagents/channels/lark --- [Log in](https://agentsky.dev/auth/start?returnTo=%2Fagents%2Fdeepagents%2Fchannels%2Flark) [Sign up](https://agentsky.dev/auth/start?returnTo=%2Fagents%2Fdeepagents%2Fchannels%2Flark&mode=signup) # DeepAgents + Lark Lark is the conversation entry point; DeepAgents is the agent that runs the work. Test DeepAgents with Claude Sonnet 5 below, then connect Lark and choose this agent in channel settings. [Repo](https://agentsky.dev/auth/start?returnTo=%2Fagents%2Fdeepagents%2Fchannels%2Flark%3Fresume%3D1) A task starts here in your browser. To receive messages through Lark, [connect Lark on your Channels page](https://agentsky.dev/channels?connect=lark) . Channel setup opens separately; it does not automatically bind this new task. How the combination works ## DeepAgents does the work. Lark carries the conversation. Runs the task through its ACP harness in a persistent cloud workspace while the selected AgentSky model supplies reasoning and generation. Connect your organization once. Everyone shares the workspace; each conversation can continue a task or start a new one. - DeepAgents Owns the work loop, model calls, files, cloud computer, enabled tools, and the result returned to the conversation. - Lark Lets people reach the selected AgentSky agent from Lark. It does not grant access to repositories or other applications mentioned in a message. - AgentSky Owns the selected responder, channel binding, task state, model and runtime usage, and the separate permissions for tools and connectors. Setup and first task ## Connect DeepAgents to the Lark conversation you intend to use. The Composer tests the exact agent in the browser. Channel setup remains a separate account or workspace authorization, so neither step can silently stand in for the other. 01 ### Test DeepAgents in the Composer The Composer above selects DeepAgents and its current recommended model, Claude Sonnet 5. Use a small task with an answer you can inspect before connecting a shared channel. 02 ### Connect Lark and choose DeepAgents In Channels, choose Lark. Register your organization’s bot and publish it. Send the temporary connection code to your bot, in a direct chat or group. 03 ### Verify the real channel path Send a small task and check the reply in Lark. Use /new for a separate session. Check that the reply arrives in the intended Lark conversation and that the dashboard still names DeepAgents as the responder. Access, cost and limits ## What DeepAgents + Lark establishes Tasks supported by this ACP harness through AgentSky's verified platform-model path. Configure and verify - DeepAgents and Claude Sonnet 5 are preselected for the browser evaluation. - Lark is connected or paired to the intended account, workspace, server, device or conversation. - The first real message reaches the selected agent and returns a useful reply through the same channel. Do not assume - A browser task does not automatically bind Lark; finish the channel setup separately. - Lark membership does not authorize repositories, email, documents or other application data. - Model tokens and active computer time are billed separately from any channel or provider plan. ## DeepAgents + Lark questions ## Related resources [DeepAgentsReview DeepAgents's supported models, capabilities and browser Composer.](https://agentsky.dev/agents/deepagents) [Lark channelReview the channel-specific authorization, pairing and troubleshooting guide.](https://agentsky.dev/channels/lark) [DeepAgents + Claude Sonnet 5Review the current default agent and model combination.](https://agentsky.dev/agents/deepagents/with/claude-sonnet-5) [PricingReview model, active-computer, tool and connector charges.](https://agentsky.dev/pricing) ## Test DeepAgents, then connect Lark. [Open the Composer](https://agentsky.dev/agents/deepagents/channels/lark#composer) ## Frequently asked questions ### Can I use DeepAgents in Lark on AgentSky? Yes. AgentSky currently offers DeepAgents as a runnable agent and Lark as a channel. Connect Lark from channel settings, choose DeepAgents as the responder, and verify one real message after setup. ### What does DeepAgents do when I send a message in Lark? DeepAgents runs the agent workflow: it reads the task context, uses Claude Sonnet 5 by default, calls the permitted tools or connectors, works in its cloud computer, and returns the result through the connected channel. ### Does connecting Lark give DeepAgents access to my other apps? No. Lark is the conversation surface. Repositories, Gmail, Notion and other systems require their own project or connector authorization, and the task still needs to state what changes are allowed. ### Can I test DeepAgents before connecting Lark? Yes. Use the Composer on this page to run DeepAgents in the browser first. The note below the Composer then links to channel settings; connecting the channel never happens silently from a browser task. ### What happens if I am signed out? The shared Composer preserves the draft and the selected DeepAgents configuration through sign-in, then returns to this page. Lark authorization opens separately and returns to its own settings flow. ### What happens if my AgentSky balance is empty? AgentSky checks the account and balance before dispatch. If funding is required, it opens the billing path and keeps the draft rather than silently changing the agent, model or channel. Channel-provider subscriptions remain separate. ### How much does it cost to run DeepAgents from Lark? There is no fixed message price. AgentSky records model token usage and active cloud-computer time for the task, plus any paid tools or connectors it calls. Lark's own plan or administrator requirements are separate; review current AgentSky pricing before a larger workflow. --- Markdown copy of https://agentsky.dev/agents/deepagents/channels/lark. Every agentsky.dev page has one: add `.md` to its address or send `Accept: text/markdown`. Start from https://agentsky.dev/llms.txt.