fomoxx

Why I Don’t Treat Investing as an Escape When My Capital Is Small

Wed, 30 Sep 2026 10:49:00 +0800

There is a sentence I disliked the first time I heard it:

If you have less than RMB 100,000 in cash, do not rush into trading and investing.

It sounds like “people with less money should not invest.”

That is not how I interpret it now.

The more useful point is that when capital is small, the absolute investment return may be too small to change your life, while the volatility can still be large enough to change your mood and behavior.

WeChat Official Account Cold Start: Recommendation Traffic, Search, and Monetization

Wed, 30 Sep 2026 10:48:00 +0800

Scope: this article summarizes real conversations among WeChat Official Account operators in May 2026. It is not an official Tencent ranking document and not a guaranteed monetization playbook. Account condition, content category, platform cycles, and luck all matter. Treat the notes below as field observations, not as a formula.

The short version

The first useful goal in a cold start is not follower count. It is producing enough content for recommendation testing and learning what the system distributes.
Reads and ad revenue are not linearly related. Posts with similar view counts can earn very different amounts.
WeChat Search (搜一搜), recommendation traffic, and private sharing are separate distribution paths.
Artificial engagement, careless originality claims, and repeated deletion/reposting can create noise or risk instead of useful growth.
The first 7–15 days are better treated as a sequence of content experiments than as a verdict on whether the account will work.

Watch recommendation traffic before follower count

New accounts often watch the wrong metric first.

Two AI Frontend Design Experiments: GPT vs Gemini vs Claude vs Kimi vs Grok vs GLM vs MiniMax

Wed, 30 Sep 2026 10:47:00 +0800

My conclusion after two rounds: AI-generated frontends have not converged to one universal aesthetic. Under low-constraint design prompts, different model/harness combinations repeatedly exposed different visual preferences. OpenAI tended to be strong at physical spaces and mature commercial branding; Gemini leaned toward concepts and interactive experiments; Claude Sonnet behaved more like a product/brand designer; Kimi was strong at Chinese typography and editorial direction; Grok leaned cinematic and theatrical. Instead of asking which model has the “best taste,” I now use those recurring biases as a multi-model design pitch pool.

ChatGPT Plus vs Team/Business: A Real-World Usage Capacity Test

Wed, 30 Sep 2026 10:46:00 +0800

Current-status note — September 30, 2026: this article is a historical usage test from May 11–12, 2026, not a current official quota table. ChatGPT Business has since introduced Standard and Premium seats. OpenAI’s current documentation says Premium includes 5× the usage of Standard and is not subject to the five-hour usage window. The Plus-A / Team-B / Team-C ranking below describes only the three accounts I measured at the time. For current limits, check OpenAI’s Business models and limits documentation .

Bitget Wallet Card Guide: Fiat24 Setup, Top-Up, Alipay, WeChat Pay, and Google Pay

Wed, 30 Sep 2026 10:45:00 +0800

What the card is

Bitget Wallet and Fiat24 launched a crypto-linked payment card that I originally tested as a virtual “U card.”

The things that stood out to me were:

Relatively clear fee structure. The current Mainland China fee page lists no issuance fee, no annual fee, and no spending fee, while top-ups and some payment scenarios can still carry fees.
Exchange rates and fees need to be checked in the live app. My original test had relatively low FX loss, but that can vary by region, funding currency, and payment method.
Multiple payment platforms. The card can be connected to services such as Alipay, WeChat Pay, Google Pay, and Apple Pay where supported.
GetGas. Bitget Wallet can automatically handle gas replenishment after funding, which is convenient.
It worked in my own payment tests. I used it for services including ChatGPT and API-related payments.

⚠️ Important: Fiat24 is not a bank deposit account with deposit insurance. I treat this as a spending tool: fund it when needed instead of leaving a large balance there.

Tencent Cloud Lighthouse Has No Automatic Snapshots: My Two-Slot Rotation with systemd

Wed, 30 Sep 2026 10:44:00 +0800

Tencent Cloud Lighthouse does not provide an automatic snapshot feature for the server workflow I was using, and my instance only has two snapshot slots.

If I rely on manual snapshots, I either consume the available recovery points quickly or simply forget to create them.

I ended up running a script on the Lighthouse server itself. The script calls the Tencent Cloud API, deletes the oldest automatic snapshot, creates a new one, and is triggered by a systemd timer every Sunday at 04:30.

Why I Self-Host Waline on a VPS with SQLite

Wed, 30 Sep 2026 10:43:00 +0800

I recently wanted to add a comment system to my blog.

Looking at the usual options, I realized how quickly comments can turn into a small infrastructure project: GitHub Issues, Discussions, Vercel, serverless functions, cloud databases, object storage, CAPTCHAs, email notifications, and anti-spam services.

None of those pieces is especially difficult on its own. Put enough of them together, though, and a comment box starts to look like another system to operate.

Slock Is Now Raft: Installation, Agent Setup, and Common Problems

Wed, 30 Sep 2026 10:42:00 +0800

2026 update: Slock has been renamed to Raft. This article keeps the real-world experience from the Slock phase and adds the current rename, compatibility, and migration context. Raft is still evolving quickly, so installation commands and UI details should always be checked against the current product page.

Slock, now renamed to Raft, is a chat-based way to dispatch local agents. You connect command-line tools such as Claude Code, Codex CLI, Gemini CLI, OpenCode, and OpenCLI , then @ them in channels to assign work, review plans, and collect results.

Enterprise AI Agents Don’t Always Need Multi-Agent Architectures

Wed, 30 Sep 2026 10:41:00 +0800

A while ago I reviewed a real AI customer-reception system and ran into a point that is easy to miss:

In many enterprise workflows, the most useful agent architecture is not several agents collaborating with each other. It is one agent constrained by a stable process.

The surrounding system should first tell the agent:

which business stage it is currently in;
what this stage is supposed to accomplish;
which tools are available;
what conditions allow a transition;
when the workflow must stop and hand over to a human.

I usually call this an SOP agent.

Multi-Agent Collaboration Patterns: Chat, Issue-Driven, SOP, and Hybrid Workflows

Wed, 30 Sep 2026 10:40:00 +0800

Once you start running several AI agents at the same time, one question shows up quickly: how should they actually collaborate?

Spinning up more agents is easy. The hard part is deciding who receives the request, who breaks it down, who executes, who reviews, where state lives, how permissions are controlled, and how the system recovers when something goes wrong.

Across my work with Slock/Raft , Multica , and an enterprise SOP-agent system, the biggest difference has not been the model. It has been the organizational pattern around the model.

Multica Guide: Issues, Agent Workflows, and Multi-Agent Collaboration

Sun, 27 Sep 2026 14:26:08 +0800

Multica is a multi-agent collaboration platform. You connect CLI agents such as Claude Code, Codex, OpenCLI , and Hermes into it, then organize tasks using workspaces, issues, automations, and CLI tools.

If Slock / Raft feels like a chatroom, Multica is closer to an office. A chatroom works well for calling someone in, quick brainstorming, and immediate course corrections. An office needs tasks, assignees, statuses, handoffs, and verification.

When you run multiple agents, multiple projects, and multiple terminals at once, Multica provides a shared space where work can retain its state, handoffs, and acceptance criteria.

OpenCLI for AI Agents: Turning Browser Workflows into Reusable CLI Commands

Thu, 24 Sep 2026 15:30:00 +0800

OpenCLI is easy to describe as “letting an AI click around websites.” That is not wrong, but it is too narrow.

A better mental model is that OpenCLI can do two things at once: let an agent operate a logged-in browser through Browser Bridge, and turn browser workflows that already work into reusable adapters or CLI commands. A flow can open a page, reuse an authenticated session, click, paginate, download, submit, and return a result. Codex, Claude Code, Slock / Raft, Multica, or another agent can then call the capability again instead of rebuilding the same interaction from scratch.

Does Multica Reduce Work or Add More? Lessons from a Real Squad Workflow

Tue, 22 Sep 2026 10:30:00 +0800

I previously wrote a Multica guide that focused on the basics: how Issues work, how agents are organized, and how multi-agent collaboration is structured.

Getting those individual features working is one thing. Putting an actual multi-stage workflow into Multica and letting several agents hand work to each other is another.

I took a real multi-stage task I was already working on and ran it through a full Squad workflow. It was not a demo invented to test Multica. It was a real workflow that needed several roles to participate in sequence, including upfront analysis, task decomposition, work by different agents, verification of key results, and an independent review step.

I Used 1.05 Billion Tokens in Cursor: What a 93.31% Cache Hit Rate Taught Me

Tue, 15 Sep 2026 15:10:00 +0800

When Cursor’s per-request pricing period was approaching its end, I exported my Usage Events from March 1 to May 19, 2026 and added them up.

The result was larger than I expected: across this Cursor workflow alone, 80 days accounted for 1,051,066,111 tokens of total throughput.

That was not my total AI usage. I was using other tools during the same period; they are not included here. This dataset only covers what I could see in Cursor’s Usage Events: requests, uncached input, cache writes, cache reads, and output.

Fix Codex App instructions_required 400 Behind CLIProxyAPI

Tue, 15 Sep 2026 15:05:00 +0800

When I routed the Codex App through CLIProxyAPI (CPA) to a third-party upstream, requests failed immediately with a 400 error:

instructions_required · 400
The instructions parameter is required and must be a non-empty string.

The failure was not fixed by upgrading CPA, turning on the built-in Codex instructions option, or trying to remove the field with payload.filter. The working fix was much smaller: keep the built-in injection disabled and use payload.default to supply a minimal non-empty compatibility value only when the field is missing.

How to Connect a Local stdio MCP Server to ChatGPT Web with OpenAI Secure MCP Tunnel

Tue, 15 Sep 2026 15:00:00 +0800

A local MCP server that communicates over stdio is easy for a desktop client to launch. ChatGPT Web is different: it runs in the cloud and cannot simply cross into your Mac and execute a local process.

I ran into exactly that boundary while trying to use Google’s official google-analytics-mcp from ChatGPT Web. The path that finally worked was OpenAI Secure MCP Tunnel:

Local MCP server (stdio)
 ↓
OpenAI Secure MCP Tunnel (tunnel-client)
 ↓
ChatGPT Web / Developer Mode / custom MCP app

This article documents the complete setup I used on macOS, including the parts that actually took time: Google OAuth, isolating failures by layer, local proxy settings, launchd, and verifying that ChatGPT could execute a real Analytics report.

Privacy Policy

Tue, 01 Sep 2026 00:00:00 +0800

Privacy Policy

Last updated: September 2026

Fomoxx is a personal blog.

This site uses Cloudflare for network and security services, Google Analytics for aggregate traffic measurement, and Waline for comments. These services may process information such as IP addresses, device information, and pages visited as needed to provide their services. If you leave a comment, Waline also processes the nickname, email address, and comment content you choose to submit.

The site also loads fonts or front-end libraries from Google Fonts and jsDelivr. When your browser requests those resources, the relevant provider may receive ordinary network-request metadata such as your IP address, user agent, and referrer.

About

Mon, 01 Jan 0001 00:00:00 +0000