@kfk_ai: Poor Man's Agent Setup - Main model: deepseek-v4-flash (1M context), use SenseTime's free quota, register multiple keys, then bundle them with 9router to avoid rate limiting. Vision model: Qwen3-VL-30B-A3B-Thinking, SiliconFlow…

X AI KOLs Timeline Tools

Summary

A user shares a configuration plan for building an Agent using a combination of multiple free or low-cost AI models, including modules for main model, vision, image generation, video, and voice.

Poor Man's Agent Setup Main model: deepseek-v4-flash (1M context), use SenseTime's free quota, register multiple keys, then bundle them with 9router to avoid rate limiting. Vision model: Qwen3-VL-30B-A3B-Thinking, using the free credits from SiliconFlow, costs only a cent per image. Small tool model: Qwen3.6-35B-A3B, SiliconFlow, used for small tasks like title generation. Large tool model: Qwen3-30B-A3B-Instruct-2507, responsible for task decomposition and summarization. Image generation model: sensenova-u1 fast / Agnes-2.1-Image, both free, but mediocre quality, toy-grade. Video model: Agenes Video V2.0 Voice model: Just use WeChat's voice input to convert speech to text, then send it out. Emergency and heavy lifting: Antigravity reverse proxy for Claude Opus 4.6 to handle; non-urgent tasks go to GLM5.1 on trae; Codex still has some free quota; mimo code 2.5 can also be freeloaded.
Original Article
View Cached Full Text

Cached at: 06/28/26, 06:01 AM

Budget-friendly Agent Configuration

Main model: deepseek-v4-flash (1M context), using SenseTime’s free tier, register multiple keys, then combine with 9router to prevent rate limiting.

Vision model: Qwen3-VL-30B-A3B-Thinking, free credits from SiliconFlow, only 1 cent per image.

Small tool model: Qwen3.6-35B-A3B, SiliconFlow, used for tasks like title generation.

Large tool model: Qwen3-30B-A3B-Instruct-2507, responsible for task decomposition and summarization.

Image generation model: sensenova-u1 fast / Agnes-2.1-Image, both free, but quality is average — toy level.

Video model: Agenes Video V2.0

Speech model: Just use WeChat input method’s voice-to-text, then send it out.

Emergency and heavy tasks: Antigravity reverse proxy Claude Opus 4.6 to handle; less urgent ones go to GLM5.1 on Trae; Codex still has some free credits; mimo code 2.5 can also be used for free.

Similar Articles

@Lonely__MH: https://x.com/Lonely__MH/status/2055156505796866407

X AI KOLs Timeline

This article shares seven tips for configuring the Hermes AI agent, including setting its soul, memory system, skill solidification, scheduled tasks, auxiliary models, multi-platform gateways, and mobile phone control, and introduces how to use models provided by Nous Research for free.

@php_martin: OpenAI Codex is now free. But what really shocked me isn't the free part — it's that local open-source models can deliver AI Agent performance close to the cloud experience. The video demonstrates 4 real-world scenarios: fixing a crashed space game, building a Whac-A-Mole web game in minutes, generating an Apple-style product homepage, and even launching a browser to search, download, and save files on its own.

X AI KOLs Timeline

OpenAI Codex is now free, but even more surprising is that local open-source models can achieve AI Agent performance close to the cloud, demonstrating scenarios like fixing games and developing web games without requiring API keys or internet.

@Saccc_c: Want to add multimodal capabilities to DeepSeek v4 flash? I strongly recommend using it with qwen3.7-flash — currently the best value model combination. qwen3.7-flash is a lightweight multimodal model that is fast and well-suited to most image understanding tasks. Key point: the price is low enough, and new registrations get 100…

X AI KOLs Timeline

Recommends pairing DeepSeek v4 flash with qwen3.7-flash to add multimodal image understanding capabilities to DeepSeek at low cost, and provides simple configuration steps using Alibaba Cloud Bailian and Codex.

@seclink: Recently this open-source tool has been quite popular. It looks like an open-source version of DingTalk Wukong and ByteDance Aily. You can use it to implement your own agent and integrate it into the aforementioned instant messaging platforms. Some guys tweaked it and used it to demo to investors, obtaining a considerable valuation. What makes investors remember...

X AI KOLs Following

CowAgent is an open-source AI assistant framework based on large language models. It supports autonomous task planning, long-term memory, knowledge base, multi-model switching, and multi-channel access (WeChat, Feishu, DingTalk, etc.), enabling rapid construction and deployment of personalized AI agents.