The 5 AI agents Lark (Feishu) teams should know in 2026 — Junior, Lark's built-in AI, open-source GPT bots, the official Lark CLI, and Base automation — and the honest pick by intent.
Lark (Feishu's international edition) is where a growing number of teams in Asia and beyond run their work and their conversation in one place. The AI-agent options for Lark are younger than Slack's, but real — here are the 5 worth shortlisting in 2026, picked for honesty rather than vendor spread.
Junior joins your Lark workspace as a bot — but it behaves like an AI employee, not a chatbot. It has its own user, a name your team picks, its own work email, and persistent memory of your team's context. It does work across 3,000+ connected tools, not just inside Lark, and it acts on a schedule or a signal without waiting to be pinged. Strongest fit when you want a teammate that ships outputs on its own.
Lark ships its own AI for summarizing long threads and docs, search, translation, and drafting inside Messenger, Docs, and Base. It's well-integrated and a good first layer. Its limits are the usual ones for built-in assistants: it stays inside Lark, doesn't carry work across other tools, and doesn't act when nobody asks. Pick this for in-app help, not autonomous work.
Projects like ConnectAI-E/Lark-OpenAI, lark_bot, and Nanobot wire OpenAI's models into a Lark bot you host yourself. You bring an API key and a server, and you get a capable in-thread assistant with custom personalities. Strong when you want full control and don't mind running the infrastructure; weak on team-level memory and autonomous, cross-tool work out of the box.
The official Lark/Feishu CLI exposes 200+ commands and 20+ AI Agent Skills across Messenger, Docs, Base, Sheets, Calendar, and more — and plugs into coding agents like Claude Code and Cursor. This is the right tool when a developer wants to drive Lark from the terminal or script it into an agent workflow. It is not a coworker your non-technical team interacts with in chat.
Lark Base's built-in automations and approval flows handle deterministic "when X happens, do Y" work entirely inside Lark — record updates, notifications, approvals. Free, native, and reliable for well-defined flows. No AI judgment between steps and no reach beyond Lark.
| You want | Pick |
|---|---|
| A coworker who ships work without being asked | Junior |
| In-app summaries, search, and drafting inside Lark | Lark's built-in AI |
| A self-hosted GPT bot you fully control | Open-source Lark bots |
| To drive Lark from your terminal or a coding agent | Lark / Feishu CLI |
| Deterministic record and approval flows | Lark Base automation |
The most common mistake is buying an in-app assistant and expecting it to act when nobody is in Lark. If that's what you need, you want an AI employee. The AI agent vs chatbot page goes deeper on the distinction.
Related reading
Follow Junior
