Kimi Work Launches Desktop AI Agent for Financial and Office Automation
The latest release from Kimi positions itself as a local agent for deep workflows, featuring browser automation, scheduled task execution, and direct access to A-shares, HK stocks, and US equities.
Kimi Work, a newly launched desktop-based artificial intelligence agent, has entered the market targeting knowledge workers with a focus on deep workflow automation. Positioned as a companion to the existing Kimi web application, which is designed for quick queries, the new software operates as a local agent intended for continuous, system-level tasks. The platform mounts local folders and navigates the web autonomously, functioning as what the company describes as a "system-level digital employee."
The application utilises a built-in Cron engine to support continuous operation, allowing it to run tasks around the clock. This includes scheduling LLM agent calls for daily briefings or executing Python scripts for dataset processing. To ensure overnight task completion, users can enable a "Keep Computer Awake" option. The software also features WebBridge, a browser automation tool that allows the agent to navigate tabs, extract data, and execute multi-step web tasks without manual intervention.
For financial research, Kimi Work comes pre-integrated with deep data sources for A-shares, HK stocks, and US equities. The company states that this integration allows users to pull earnings reports, analyse market anomalies, and reconcile spreadsheets through natural conversation, bypassing the need for complex API setups. Once research is compiled, the tool can automatically generate PowerPoint decks and Excel sheets from the gathered insights.
A core component of the software is its "Swarm Intelligence" feature, which coordinates multiple specialised agents to break down and solve multi-layered tasks simultaneously. The promotional materials mention the use of up to 300 agents to empower financial research and office tasks, although the specific capabilities of these individual agents are not detailed in the source documentation.
Security and user control are central to the design, with an "Ask before acting" safeguard that prompts for explicit authorisation before the agent modifies, overwrites, or runs code within local directories. This ensures that no changes occur without user consent, addressing potential concerns regarding autonomous file management on desktop systems.
