← Sticker Wall AGENT · DETAIL

Cognosys Open-Source Desktop Agent Local Proxy Running, Monthly Income 15,000 RMB

Workflow: Receive client requests every day, build local agent scripts using Cognosys or OpenClaw, and write repetitive client ope

AGENT

Key Fields

FIELD STAMPS

IndustryE-commerce / Retail

RegionGlobal

ScaleSME

ChannelOnline

🔧 Workflow

Receive client requests every day, build local agent scripts using Cognosys or OpenClaw, and write repetitive client operations such as document organization, spreadsheet aggregation, and batch file renaming into natural language instructions. Clients place orders via WeChat or online to describe their pain points. Individuals first replicate the task scenario remotely or in a local environment, then write natural language prompts and simple scripts to let the agent automatically execute them on the client's computer or hosted cloud server. After the agent finishes running, it outputs result files or operation logs. Humans check item by item to ensure there are no accidental deletions, incorrect labels, or formatting errors. After confirming everything is correct, the deliverables are packaged and delivered, recording the time spent and the amount of API calls consumed, making it convenient to quote monthly packages for the client later. Spend half a day every week reviewing high-frequency service task types, and solidify processes that run stably into a skill package so that subsequent new clients can apply them directly, reducing marginal costs.

🛠 Setup Requirements

Requires an ordinary computer as a development and testing environment, installing open-source frameworks like OpenClaw or Cognosys, and preparing API keys to access large models. The technical barrier is mainly knowing how to use natural language to describe task steps; there is no need to write complex code from scratch, but knowing a little Python and regular expressions can significantly improve debugging efficiency. The tool itself is free, and spending 1 to 2 days in the early stage to run the first automated task is enough to start taking external orders. If you need to provide hosting and proxy running services for clients, you also need to prepare a low-spec cloud server or an old computer left on long-term, and install remote control software for easy login and maintenance at any time. It is recommended to prepare a secondary WeChat account or a Xianyu account specifically for taking orders, organizing service introductions and case screenshots into a standardized quotation sheet to reduce communication costs.

🧰 Toolchain

🔧 Cognosys
🔧 OpenClaw
🔧 Ollama
🔧 Python
🔧 Remote Control Software

💰 Revenue

15,000 RMB/month, mainly composed of monthly subscriptions from 20 to 30 small e-commerce or administrative clients. Each client is charged 300 to 800 RMB per month for automation task complexity, and one-time deployment can be charged 500 to 1,200 RMB. Local proxy running services are charged per occurrence, at 50 to 150 RMB per single instance, and peak monthly revenue can reach 20,000 RMB. In terms of revenue structure, monthly subscriptions contribute about 60%, one-time deployment contributes about 25%, and per-instance proxy running contributes about 15%. The unit price per customer is relatively low, but repurchases are stable, and the customer churn rate is low, because once daily operations are handed over to the agent, it is difficult to return to manual operation. In peak seasons such as before e-commerce mega-sales, proxy running demand for product listing and order organization surges, and single-month revenue is expected to spike to over 30,000 RMB.

💸 Cost

API call fees are about 200 to 500 RMB per month, depending on the frequency of large model calls and model tiers. If a large number of clients choose local Ollama deployment, API costs can be reduced to under 100 RMB. The open-source framework itself is free, with only electricity and network fees on a daily basis. If a more stable available cloud server hosting is required, an additional 150 to 300 RMB per month. The basic version of remote control software is free, and the advanced version is about 30 to 60 RMB per month. Overall costs are controlled within 10% of monthly revenue, with the core cost being time rather than money.

⏱ Time Investment

3 to 4 hours per day, concentrated in the evening to process client requests, debug instructions, and inspect deliverables. Spend 1 to 2 hours on weekends to update the skill library and record instructional videos for traffic generation. During order-taking peaks such as before e-commerce mega-sales, daily time investment may reach 6 hours, but corresponding revenue will also double. Time investment has high flexibility, making it suitable for operation during spare time from a main job.

🚀 Getting Started

First, automate daily operations on your own computer using OpenClaw, such as batch organizing download folders or applying formulas to spreadsheets. Record an operation video and post it to a technical community or Xianyu to take the first order helping clients organize repetitive reports. After inspection, rely on word-of-mouth and repurchases to roll forward. Beginners should not pursue large clients in the first step; instead, first post desktop agent service introductions on Xiaohongshu and Xianyu, with a before-and-after comparison screenshot, pricing starting from 50 RMB per instance. After accumulating 5 to 10 successful cases, organize the client's high-frequency tasks into a standardized service menu, and then gradually guide clients to convert to monthly subscriptions.

🔑 Keys to Success

✅ Choose the right high-frequency repetitive tasks to cut in, prioritizing client pain points such as invoice organization, product listing, and spreadsheet aggregation.
✅ Manual review before delivery ensures quality, building client trust and avoiding compensation caused by AI misoperation.
✅ Use natural language instructions to lower the learning barrier, making clients feel they can get started themselves and be willing to pay continuously.
✅ Package services into a subscription model rather than a one-time delivery to stabilize monthly income and increase repurchase rates.
✅ Keep pace during the rapid iteration period of open-source frameworks, turning new skill modules into products to sell to old clients immediately.
✅ Continuously publish operation cases on Xianyu and Xiaohongshu, using visual before-and-after comparisons to attract keyword search traffic.

⚠️ 风险

⚠️ AI misoperation leads to client data errors, requiring prior signing of disclaimer agreements and retaining backups.
⚠️ Open-source frameworks update quickly and skills become invalid; time must be spent testing new version compatibility every month.
⚠️ Core reliance on external large model APIs; provider price increases or rate limits will affect costs and delivery timeliness.
⚠️ Changes in platform rules may lead to platforms like Xianyu delisting services, requiring multi-platform diversification for taking orders.
⚠️ Client data privacy is sensitive; high-pressure areas such as bank accounts and internal financial systems should be avoided during remote proxy running.

📌 Real Cases

📌 Domestic individual developers used OpenClaw to help e-commerce sellers with batch product image renaming and spreadsheet aggregation, with a single client monthly fee of 300 to 800 RMB, cumulatively serving more than 20 small shops.
📌 Cognosys review pages show that users can use natural language to create agent application programs to obtain insights, and some early users earned side-income by helping small and medium-sized enterprises with data organization automation.
📌 In the 2026 OpenClaw community cases, someone packaged desktop document classification and archiving and code debugging into standardized skill packages, posting orders on Xianyu and Taobao, stably taking 8 to 12 orders per month.

Sources