Tech Share on TorchTree

WeChat Official Account Traffic Revenue Rules in 2026: Long-Form Articles, Image Posts, and Channels

Thu, 13 Aug 2026 07:20:51 +0800

In 2026, WeChat Official Accounts are no longer a “subscription-based private-domain media platform.” The platform has raised the weight of public-domain recommendation to unprecedented levels: non-follower traffic can account for 70%-90%, with the algorithm doing three-dimensional matching across account tags, article tags, and user interests to actively push content to people who haven’t followed but might be interested. An account with a thousand followers can still get 100,000+ reads if the content is precise and the tags are right.

This shift directly changes the logic of choosing content formats. Long-form articles, image posts (commonly called “Little Green Book”), and WeChat Channels — the three formats have completely different traffic mechanics, revenue rates, and rule red lines.

1. Five Key Changes to Official Account Rules in 2026

Before discussing content formats, let’s look at the platform rule changes themselves — the underlying constraints shared by all formats.

1. Public-Domain Recommendation Weight Soars; Follower Dividends Fade

In 2026, non-follower traffic can account for 70%-90% of reads, of which recommendation traffic is about 50%-60%. The algorithm does three-dimensional matching across account tags, article tags, and user interests, actively pushing content to non-followers. Follower count is no longer the decisive factor — the key is content precision and tag matching.

In the message feed, avatars of accounts you follow and interact with frequently still appear in the pinned “Frequently Viewed” section, and the first 2-3 screens are mostly the latest content from followed accounts. The safe strategy is to walk on two legs: use the subscription message stream to retain existing followers, and use the recommendation stream to acquire new users.

2. Algorithm Metric Ranking: Completion Rate First

Based on multiple data sources, the approximate 2026 algorithm weight ranking is: completion rate (~35%) > share/forward rate (~30%) > bookmark rate (~20%) > interaction quality (~15%).

Completion rate has become the #1 core metric. Articles with overly long lead-ins, low information density, or content users click away from immediately will be stopped by the algorithm. If a user closes the article before reading 30%, the system marks it as “low-quality content” or “title fraud,” affecting future traffic allocation.

Sharing, forwarding, and other social virality behaviors still carry high weight. Bookmark rate’s importance has risen notably compared to the past — the system treats “bookmarking = high value” as a key trigger for long-tail traffic. Interaction quality is now assessed more granularly, looking not only at comment length and quality but also at the author’s reply rate.

3. Longer Lifecycle for Quality Old Articles

Articles with high originality, clear keyword placement, and complete structure are more likely to be associated with search. The recommendation lifecycle of quality old articles has extended from the past 24 hours to 7 days or even months. This means the long-tail effect of content has significantly strengthened: a single high-quality article can keep bringing traffic, and search traffic is becoming an increasingly important entry point.

4. AI Governance Upgraded: Non-Human Automated Creation Explicitly Banned

On March 27, 2026, the WeChat Official Accounts Platform updated its Operating Rules, adding a new special clause on “non-human automated creation behavior.” The following three types of behavior are explicitly listed as violations:

Generating, rewriting, splicing, or repurposing content using AI or similar means, detached from real creators’ expression
Mass, continuous publishing of content through automated methods such as scripts or program hosting
Spreading tutorials, methods, or services for non-human automated creation

Penalties range from content removal to account bans — quite heavy. In its response, the WeChat team made clear: the platform encourages creators to use tools reasonably to assist creation, but opposes fully automated programs replacing humans in content production.

The platform does not prohibit AI-assisted writing (such as checking typos, polishing sentences, or generating charts), but emphasizes that topic selection, ideas, frameworks, and opinions must be decided by humans. AI-generated drafts need at least three kinds of manual processing: adding real cases or data, injecting personal opinions and judgments, and adjusting the tone to remove the “AI flavor.”

5. Traffic-Guiding Behavior Is Being Throttled

Starting in May 2026, Official Accounts tightened traffic-guiding rules. Directly or indirectly embedding contact info, external links, QR codes, or other guiding information — through the account bio, article body, comments, or Official Account messages — to funnel users to private accounts, external pages, or third-party platforms counts as prohibited traffic guiding. Articles containing such content are flagged as “not eligible for content recommendation.”

Leaving a WeChat ID in articles is no longer a zero-risk operation. If your goal is content rather than funneling to private domain, keep the content clean; if you really need traffic guiding, be aware that it will affect recommendation traffic.

2. Traffic Mechanics of the Three Content Formats

1. Long-Form Articles: The Foundation of the Search Economy

Long-form articles are the most traditional Official Account format and the core carrier of search value. For content like tech tutorials, product comparisons, and deep analysis, reader search intent is clear, and traffic from Search (搜一搜) is precise and stable.

Relying purely on long blocks of text for traffic is getting harder, but the search-economy logic hasn’t changed: each article is an independent search answer that doesn’t depend on a follower base. The more articles you have, the more search entry points; the more articles in the same niche, the more the platform treats the account as authoritative. This niche-authority effect gives long-form articles long-term value in vertical domains.

For content with high knowledge density, long-form remains the only format that can carry it fully. From a GEO (Generative Engine Optimization) perspective, structured long-form articles with high data density are more likely to be extracted by Tencent’s AI products (Yuanbao, Doubao, Kimi) as authoritative answers and become cited sources for AI — a value short-form content can’t replace.

2. Image Posts (Little Green Book / TieTu): An Officially Supported Traffic Dividend

Image posts are the Official Account “image/text” feature, nicknamed “Little Green Book” after its 2023 redesign because it resembles Xiaohongshu. In March 2026, WeChat officially renamed it “TieTu” (sticker posts) in the creation area and gave it explicit traffic support: a dedicated “TieTu” section on the Official Account homepage, with TieTu recommendations appearing every 2-3 screens in the feed.

In form, image posts support up to 9 images with a title and up to 1,000 characters of description. On mobile, users swipe horizontally, with a default 3:4 aspect ratio. Static images, animated GIFs, and live photos are supported.

Its traffic advantages are threefold:

First, official traffic favoritism. Both the dedicated homepage section and interleaving in the recommendation feed show the platform is actively pushing this format. For creators, this is currently the incremental entry point with the highest return on effort.

Second, high completion rate. An image-first, text-secondary format naturally fits fragmented reading scenarios: users can finish consuming content by quickly swiping in the feed, giving completion rates far higher than long-form articles. Completion rate is the #1 algorithm metric, so this advantage translates directly into recommendation weight.

Third, low creation barrier. No long-form layout needed — a few images plus a short description and you can publish, making it ideal for high-frequency updates. The mobile Subscription Account Assistant app allows publishing anytime, and even has a built-in “text-to-image” capability.

The limitations of the Little Green Book also need to be known: publishing image posts uses the same once-per-day mass-send opportunity; you can’t mark original or add topic tags; information density is low, so it’s not suited to deep content; and it currently sits below private-domain account content, with exposure tied to account weight.

WeChat Channels is now deeply integrated with Official Accounts: the Channels feed is shown on account homepages, video content is interleaved into the information feed, and the mobile creation entry points are merged. WeChat is shifting from a subscription-based private-domain platform to an algorithm-recommended, all-domain content platform, and Channels is the core of this transformation.

In 2026, Channels’ traffic mechanics are a dual engine of “social recommendation + algorithm recommendation.” Content exposure is composed of four layers: follower traffic (subscription logic), friend-relationship traffic (friends’ likes/views triggering secondary spread), algorithm recommendation, and search. Social attributes are what distinguish Channels from Douyin: friend interaction signals carry equal weight with algorithmic judgment, and content starts from WeChat’s acquaintance relationship chain rather than a stranger public-domain pool.

Channels’ support direction in 2026 is clearly focused on originality: works declared original get significantly higher weight in the recommendation algorithm, and real-person filmed, vertical professional, and positive-energy content gets extra cold-start and exposure support. Meanwhile, new rules from April 2026 require AI-generated or AI-assisted content to be labeled “AI-generated,” or it will be throttled and taken down. The platform explicitly cracks down on unlicensed reposting and low-quality remixing — reposting, simple secondary creation, and unauthorized film/TV music clips are all banned from being declared original.

Channels’ revenue logic differs from Official Accounts and is covered separately below.

3. Revenue Comparison Across the Three Formats

1. Official Account Traffic Monetization: Priced by Niche

The threshold for enabling Official Account traffic monetization has dropped from 500 followers to 100 followers (adjusted in early 2026). Revenue is determined by eCPM (earnings per thousand impressions), with the core formula: single ad revenue = ad impressions × eCPM ÷ 1000.

eCPM is heavily influenced by content niche, follower quality, and account verticality, and varies enormously:

Content Niche	Revenue per 1,000 Reads	Notes
High-value verticals: finance, education, parenting, career	8-15 RMB	eCPM can reach 20-60 RMB
General entertainment, motivational content, reposted content	1-3 RMB	eCPM only 2-8 RMB

At the same 10,000 reads, professional niches can earn 300-1,500 RMB while lifestyle content might only get 50-300 RMB. Vertical professional content commands far higher rates than general entertainment — this is the #1 determinant of monetization revenue, more important than traffic scale itself.

On ad click-through: high-quality Official Accounts in vertical niches typically see 1%-3%, while general entertainment may be as low as under 0.5%. Ad rates are equally divergent: CPC (cost per click) in finance, education, and medical niches can reach 2-5 RMB, while lifestyle content is usually 0.3-1 RMB.

The Channels creator revenue-share program is an ad-impression-based revenue model within the Channels ecosystem: qualified original creators can earn income from ads displayed in the comment sections of their original videos and in the video feed on their profile pages. In late 2025, the threshold dropped from 1,000 to 100 followers, with individual creators receiving roughly 60% of the share.

Industry-tested revenue ranges for 2026:

Content Type	Revenue per 1,000 Valid Plays
General entertainment content	0.3-1 RMB
Quality vertical content (finance, legal, education, career)	1.5-5 RMB

High-quality content can earn 3-5x the rate of ordinary content, with revenue strongly tied to completion rate, interaction rate, comment-section dwell time, and originality. A viral vertical video with 100,000 plays earns roughly 300-1,000 RMB in ad share. Top accounts’ monthly share revenue stably sits in the 10,000-50,000 RMB range.

3. Little Green Book (Image Post) Revenue

Image posts follow the same revenue logic as long-form articles — paid by impressions through traffic monetization. When a single sticker post’s views pass a thousand, the content already has reach. Industry data shows quality sticker posts can boost user dwell time by 30%, indirectly lifting ad click-through by 15%-25%. But image posts have low information density and lower per-unit rates than deep long-form articles, making them better as a traffic entry point than a revenue mainstay.

4. How to Choose: Decide by Content Goals and Production Capacity

The three formats are not substitutes but a division of labor. Based on the mechanics and revenue data above, here’s a selection framework by account type.

High-Knowledge-Density Vertical Accounts: Long-Form First, Little Green Book as Support

For fields requiring deep explanation — tech, finance, education — long-form is the foundation. In the search economy, long-form is the only format that fully carries knowledge density and is the core carrier of GEO citations and Search traffic. The Little Green Book suits repackaging data comparisons, key checklists, and operation screenshots from long articles for secondary distribution, capturing the officially supported recommendation traffic.

Seeding, Lifestyle, and Visual-First Accounts: Little Green Book Is the Mainstay

For photography, travel, beauty, food, and fashion content, visual presentation is the content itself. The Little Green Book’s 3:4 large images, horizontal swiping, and low creation barrier are a natural fit. Such accounts can use the Little Green Book for high-frequency updates to stay active, with long-form as periodic deep output.

Accounts With Real-Person On-Camera Capability: Channels Is the Increment

Channels’ 2026 support clearly favors real-person filming and vertical professional content, and the social virality mechanism makes acquaintance relationship chains a cold-start advantage. The mandatory labeling rule for AI-generated content narrows the room for pure AI content on Channels. If you can appear on camera or do real-voice narration, Channels is worth investing in as a distribution channel.

The Common Bottom Line for All Accounts

Regardless of format, the 2026 rule bottom line is consistent: topic selection, ideas, framework, and opinions must be decided by humans, with AI only assisting; no reposting, no low-quality remixing, no batch automated publishing; no contact info or traffic guiding in articles and comments. Content purity directly determines whether an account can keep receiving recommendation traffic.

5. References

Why Warp Went Open Source and What It Means

Wed, 29 Apr 2026 03:32:22 +0800

On April 28, 2026, Warp’s official blog announced that its terminal client is now open source under the AGPL license, with the code hosted on GitHub. This decision is not merely a code release; it comes with a collaboration mechanism called the Agent-first Workflow, where community contributors work alongside AI Agents via Warp’s own Oz platform to drive development. OpenAI participated in the experiment as a founding sponsor.

This article analyzes Warp’s open-source motives, the impact on users, the differences between the free and paid tiers, and the product’s future direction, based on Warp’s official announcements and pricing information.

What Is Warp

Warp is a modern terminal application for developers, with core features including block-based command editing, autocompletion, history search, and AI-based Agent-assisted programming. Beyond the terminal client, Warp has also launched product lines including Agents (multi-Agent deployment and tracking), Code (building complex features in production codebases), Drive (knowledge base and context management), and Oz (cloud Agent orchestration platform).

Before the open-source move, Warp’s terminal client was closed source, with AI features billed by usage. After open-sourcing, the terminal client code is public under the AGPL license, but AI capabilities and cloud services remain paid.

Why Warp Chose to Open Source Now

Warp’s official blog gives two levels of reasons.

The primary reason is a bottleneck shift in development efficiency. In traditional software development, coding itself is often the most time-consuming part. But as AI Agent capabilities improve rapidly, the Warp team found the limiting factor has shifted from “writing code” to “specification and verification done by humans.” Agents can handle the heavy lifting of implementation, while human contributors focus on higher-leverage work: deciding what to build and ensuring it’s built right. Having the open-source community participate in managing Agents could, in theory, break through the manpower ceiling of an internal team.

The secondary reason is strategic positioning in a niche. Warp points out that there is currently no fully functional open-source Agentic Development Environment (ADE) on the market. By open-sourcing its client, Warp aims to become the benchmark alternative in this emerging field, differentiating against well-funded closed-source competitors.

It’s worth noting that Warp has had open-source plans since its initial release in 2022, but only executed them in 2026. The team explains that “the rise of Agents” changed the feasibility of open-source collaboration. Without Agent assistance, the review and integration costs of large-scale community contributions would be unbearable.

What Is the Agent-first Workflow

The Agent-first Workflow is the core collaboration model Warp adopted after going open source. It works like this: community contributors propose feature directions or submit requests, Warp’s Oz platform dispatches AI Agents to do the coding, planning, testing, and other implementation work, while the core team handles final quality control and merge decisions.

The difference from traditional open-source projects: in the traditional model, community PRs require manual review and testing by the core team, making manpower the bottleneck; in the Agent-first model, Agents take on most of the implementation and verification work, and the human contributor’s role shifts from “writing code” to “setting direction” and “doing verification.”

Warp uses public GitHub issues as the single source of truth for feature tracking and has published a public roadmap for the ADE. This means the community can directly influence product direction by filing issues and joining discussions.

What Open Source Brings to Users

Transparency and Auditability

The terminal is the developer’s core entry point for interacting with the system, and the commands and data it handles often involve sensitive information. Open source means users can directly audit what Warp executes locally and how data flows. This has real value in security and compliance scenarios, especially for enterprise environments that need to vet third-party tools.

Greater Customizability

Warp has also introduced a programmable settings file, supporting configuration portability across devices. Users can adjust the interface level to their preferences, from a minimal terminal to a full ADE interface with diff views and file trees.

More Model Choice

The open-source version adds support for open-source models such as Kimi, MiniMax, and Qwen, and introduces an “auto (open)” routing feature that automatically selects the best model based on task type, reducing the cost of manual switching.

Free vs. Paid: What’s the Difference

Warp’s business model is a hybrid of “open-source client + paid cloud services.” Core terminal features are free for all users, while AI capabilities and cloud resources are tiered and billed by usage.

Dimension	Free	Build ($18/month)
Core terminal features	Fully free	Fully free
Source code	Open source under AGPL; viewable/modifiable	Same open-source codebase
Monthly AI Credits	150 (first 2 months), then 60	1,500
Available AI models	Limited	Frontier models from OpenAI, Anthropic, Google
Concurrent cloud Agents	4	20
Cloud Agent compute	2 vCPU / 4 GiB	4 vCPU / 8 GiB
Codebase indexing	3 codebases, 3,000 files each	40 codebases, 100,000 files each
Warp Drive knowledge base	10 workflows + 3 notebooks	Unlimited
BYOK (bring your own API key)	Not supported	Supported

For pure terminal users, open source doesn’t change the experience. Warp’s block editing, autocomplete, history search, and other features remain free. The barrier to AI features is credits and model access.

In late 2025, Warp simplified its pricing from multiple tiers (Pro/Turbo/Lightspeed) into a single Build plan and introduced a Reload Credits mechanism: excess usage can be purchased at 50% less than the old pricing, valid for 12 months.

One option favorable to heavy users is BYOK (Bring Your Own Key): Build and above users can connect their own OpenAI, Anthropic, or Google API keys, so AI costs go directly onto their existing provider bills, with Warp charging only the $18/month base subscription. For developers who already have AI subscriptions, this can significantly cut total costs.

How Different Users Should Choose

User Type	Recommended Option	Expected Cost
Pure terminal users	Free tier	$0
Light AI users	Free tier credits	$0
Developers with existing AI API keys	Build + BYOK	$18/month
Teams needing collaboration	Business ($45/person/month)	Per-seat pricing
Enterprises with strict data sovereignty	Enterprise (custom)	Custom pricing

The free tier’s 60 credits per month is quite tight for any substantive AI-assisted development. This means most users who want Warp’s AI features will eventually need to go the paid or BYOK route.

Warp vs. Closed-Source Competitors

Warp’s open-source strategy sets it apart from closed-source competitors like Cursor and GitHub Copilot.

Dimension	Warp	Cursor / Copilot
Client code	Open source under AGPL	Closed source
Core revenue	Cloud Agent services + AI credits	Subscription fees
Model choice	Multiple providers, including open-source models	Mostly proprietary/partner models
Community involvement	Agent-first; community sets direction	Officially driven
Self-hosting options	Supported in Enterprise	Generally not supported

The strong copyleft nature of the AGPL license means any derivative work based on Warp’s code must also be open source. This may make some enterprises hesitate to deeply customize, but it also ensures community contributions flow back into the main project.

The Product’s Future Direction

Short Term: Competing on Iteration Speed

If the Agent-first collaboration model works, Warp’s feature iteration speed could significantly outpace closed-source competitors that depend on in-house engineers. Community contributors propose directions, Agents do the implementation, and the core team handles quality control. In theory, this pipeline can process a large number of feature requests in parallel.

Medium Term: Contesting the Open-Source ADE Niche

Warp explicitly positions itself as a pioneer of the “open-source Agentic Development Environment.” This niche is currently empty, but competition will intensify quickly. Whether Warp can build enough community scale and contribution quality will determine whether it can hold this position.

Long Term: Cloud Services as the Profit Core

Open-sourcing the client won’t shake Warp’s business model. AI compute, cloud Agent orchestration, and enterprise-grade management (SSO, data retention controls, self-hosted Agents) are the real revenue core. This resembles GitLab’s “open-source core + paid features” path. Warp’s Enterprise plan even supports Bring Your Own LLM and self-hosted cloud Agents, targeting large organizations with strict data sovereignty requirements.

Potential Risks and Uncertainties

The Agent-first collaboration model is experimental, and its effectiveness remains to be proven. Whether community contribution quality can be effectively amplified through Agents, and whether the core team’s review bottleneck will reappear in a new form, are open questions.

Additionally, the free tier’s low credit allowance may push many users toward paid plans. Warp needs to balance free-user growth against paid conversion, avoiding the free tier becoming a “trial trap” that harms the user experience.

Summary

Warp’s open sourcing is not the traditional “giving back to the community”; it is a business model experiment built on Agent capabilities. It uses an AGPL open-source client to build trust and transparency, and the Oz platform and cloud Agent services to build its commercial moat.

For users, pure terminal features are unaffected, and the cost of AI capabilities depends on usage intensity. For the industry, this marks a potential new phase of open-source collaboration where “humans set the direction, Agents do the execution.” Whether this model endures depends on Agent quality, community activity, and Warp’s ability to balance free and paid.

References

Warp Is Now Open-Source, Warp official blog, 2026-04-28
Warp Pricing, Warp official pricing page

Web-Based Presentations: Several Open-Source Options More Flexible Than Traditional Slides

Tue, 28 Apr 2026 02:03:53 +0800

Using the web for presentations was once a niche choice among technical practitioners. But over the past two years, more and more non-technical users have been trying it — in academic presentations, product launches, corporate training, and similar scenarios, web-based slides are moving from the fringe to the mainstream. The driving force behind this isn’t complicated: web technologies are naturally cross-platform, version-control friendly, and offer interactivity that traditional PPT cannot easily achieve.

This article reviews several web presentation tools worth attention, split into two layers — core frameworks and AI generation — with directly accessible project links.

1. The Core Framework Layer

The infrastructure of web presentations is built on a set of open-source frameworks. They liberate slides from proprietary formats and rebuild them on open web technologies.

reveal.js: The “De Facto Standard” for HTML Presentations

reveal.js is currently the most mature HTML presentation framework, developed by Hakim El Hattab. Its core positioning: anything that can be done on a web page can be used in a presentation.

The framework supports vertically nested slides, speaker view (with timer and next-slide preview), Auto-Animate, LaTeX formula rendering, and more. Slide transitions include fade, slide, convex, concave, zoom, and other modes. For those unfamiliar with code, the same author also provides a companion no-code editor, slides.com.

reveal.js suits technical presentations with high requirements for interactive effects — for example, embedding live code execution, iframe web content, or complex CSS animations.

Project link: github.com/hakimel/reveal.js

Slidev: The Developer’s Markdown Choice

Slidev is designed specifically for developers. Its tech stack is built on Vite + Vue 3 + UnoCSS, using Markdown as the content source format. Its biggest advantage is unifying the “writing code” and “making slides” workflows in the same environment.

Slidev’s standout features include Shiki-powered code highlighting, built-in live code demos, Mermaid diagram support, KaTeX math formulas, presenter pen annotations, and a built-in recording tool. For deployment, running slidev build outputs a static site that can be hosted on any platform.

For developers who already manage documents with Git, Slidev’s Markdown source files are naturally version-controlled — an advantage traditional PPT can’t match.

Project link: github.com/slidevjs/slidev

Marp: The Lightest Markdown Option

Marp stands for Markdown Presentation Ecosystem and offers an out-of-the-box experience as a VS Code extension. After installing the extension, create a new Markdown file, add the marp: true front-matter config, and you’re ready to make slides.

Marp has the lowest learning curve of the three, and supports exporting to PDF, PPTX, and HTML. It suits quick academic reports, work briefings, and other scenarios that don’t demand high design complexity.

Project link: github.com/marp-team/marp

Spectacle: A Presentation Option for the React Ecosystem

Spectacle is a React component-based presentation framework maintained by Nearform (formerly FormidableLabs), with slides written in JSX. It integrates seamlessly with the React ecosystem and supports live coding. For teams already deeply invested in the React stack, Spectacle enables component reuse — for example, embedding elements from your UI component library directly into a deck.

Project link: github.com/FormidableLabs/spectacle

2. The AI Generation Layer

Frameworks solve the “how to present” problem; AI tools are starting to solve “how to generate content.” Here are the open-source options worth watching.

Presenton: The Most Feature-Complete AI Presentation Generator

Presenton has 4.8k stars on GitHub and is open sourced under the Apache 2.0 license. It positions itself as an open-source alternative to Gamma, Beautiful.ai, and Decktopus.

The project’s core features include:

BYOK model support: Use APIs from OpenAI, Gemini, or Anthropic, or connect local models via Ollama
Template inheritance: Upload existing PPTX files to use as design templates
Fully local operation: Offers both an Electron desktop app and Docker self-hosting; all processing can be done locally
Built-in MCP Server: Supports the Model Context Protocol for integration with other AI toolchains

Deployment is very simple — a single Docker command starts the service. For data-privacy-sensitive scenarios, such as internal corporate reporting or presentations involving trade secrets, Presenton’s local operation capability is an important plus.

Project link: github.com/presenton/presenton

PPTAgent: An Academic-Grade Generation Option

PPTAgent is a research project published at EMNLP 2025, developed by the Institute of Computing Technology, Chinese Academy of Sciences (ICIP-CAS). Its distinguishing feature is a systematic quality evaluation mechanism. The project uses a two-step generation pipeline and includes the PPTEval scoring system, which quantitatively evaluates generated results across three dimensions: design, flow, and content quality.

PPTAgent also supports Docker deployment in seconds. It suits research or enterprise environments that need quantifiable evaluation of generation quality — for example, when you need to prove to clients that AI-generated content meets a specific quality bar.

Project link: github.com/icip-cas/PPTAgent

3. The Toolchain Layer: Format Conversion

Dedicated conversion tools bridge web presentations and traditional workflows.

DeckTape: High-Quality PDF Export for HTML Presentations

DeckTape is built on Puppeteer and the Chrome rendering engine, and can export 13+ HTML presentation frameworks — including reveal.js, Slidev, impress.js, and remark — to high-quality PDF. Its usage is simple:

The tool also supports selective export (e.g., only pages 1, 3, and 5-10), multi-resolution screenshot capture, and Docker containerized operation. For offline distribution or print backup scenarios, DeckTape is an indispensable part of the web presentation workflow.

Project link: github.com/astefanutti/decktape

4. Selection Advice

Need / Scenario	Recommended Option
Already have Markdown content; want a web presentation quickly	Slidev or Marp
Need AI assistance to generate a complete deck from scratch	Presenton (self-hosted)
Want complex interactive effects and embedded web apps	reveal.js
React stack; need componentized reuse	Spectacle
Need PDF backups or offline distribution	Any framework + DeckTape
Data-sensitive; must run locally	Presenton + local Ollama models

5. A Workflow Worth Watching

These tools are currently forming a clear pipeline:

Markdown/text → AI-generated content → web presentation framework rendering → DeckTape PDF export

This means users can generate an outline with AI, refine it in Markdown with Slidev or Marp, build it into an interactive web presentation, and export a PDF with DeckTape as a fallback. This “web-first, PDF as backup” pattern is the technical foundation behind the claim that “web presentations are better.”

Web presentations have objective advantages in interactivity, version control, and cross-platform consistency. But they’re not a silver bullet: traditional PPT still holds ground in business compatibility, review/annotation features, and acceptance among non-technical users — real constraints web solutions must face. Which tool you choose ultimately depends on your specific use case and audience.

Unsloth: The Accelerator Democratizing AI Training

Thu, 23 Apr 2026 15:09:04 +0800

Founded by brothers Daniel Han and Michael Han in late 2023, Unsloth is an open-source framework for efficient LLM fine-tuning. Its core mission can be summarized in a single sentence:

Turning model training — which used to require expensive hardware and deep engineering skills — into something within reach of ordinary developers and even individual hobbyists.

1. Technical Innovation: Rebuilding Training Efficiency from the Ground Up

Unsloth does not simply wrap existing frameworks; it rewrites CUDA kernels from the ground up, delivering a series of hardcore technical breakthroughs:

Technical Breakthrough	What It Delivers
Custom Triton Kernels	Core operators such as RoPE embeddings and cross-entropy loss rewritten with OpenAI Triton to optimize forward and backward passes
2-5x Speedup	Training speed improved 2 to 5 times over a standard Hugging Face + PyTorch pipeline
80% Memory Savings	GPU memory usage reduced by roughly 80% via weight deprojection and gradient checkpointing optimizations
Lossless Precision	All optimizations maintain 32-bit precision without sacrificing model quality
Train Large Models on a Single GPU	Fine-tune 70B-parameter models on consumer GPUs such as the RTX 4090 (24GB)

1.1 Weight Deprojection: The Core of Memory Optimization

One of Unsloth’s key innovations is weight deprojection. Traditional fine-tuning requires storing both the original weights and optimizer states (such as Adam’s momentum and variance) simultaneously, which is a huge burden on GPU memory. Through a mathematically equivalent transformation, Unsloth computes weight updates on the fly when needed rather than persistently storing the full optimizer state, dramatically reducing memory usage.

1.2 Automatic Gradient Checkpointing and Flash Attention Integration

Unsloth automatically integrates Flash Attention 2/3 and intelligently selects gradient checkpointing strategies to balance compute and memory. Users don’t need to manually configure these low-level optimizations — the framework detects hardware capabilities and applies the best strategy automatically.

2. Leveling the Playing Field: From Enterprise to Individual

2.1 Enterprise Teams: Lowering Experiment Costs

For enterprise teams with A100/H100 clusters, Unsloth’s direct value is reducing experiment costs. A task that needed 8 A100 GPUs for 3 days might require only 4 GPUs for 1 day after Unsloth optimization. This means:

The same hardware budget can run more experiments
Model iteration cycles shrink from weeks to days
Small and mid-sized teams can afford LLM fine-tuning

2.2 Small Teams: Bypassing the Engineering Barrier

Fine-tuning large models traditionally requires deep CUDA and distributed training knowledge. Unsloth lowers this barrier through:

One-line enablement: FastLanguageModel.from_pretrained() replaces complex configuration
Automatic hardware adaptation: Automatically detects the GPU model and applies optimal kernels
Hugging Face ecosystem compatibility: Existing datasets and model repositories can be used directly

2.3 Individual Hobbyists: The Possibility of Consumer Hardware

This may be Unsloth’s most disruptive impact. This change in hardware accessibility means individual developers, students, and independent researchers can participate in large model training without relying on cloud computing resources or enterprise-grade hardware.

3. Community Validation: Real-World Feedback

Unsloth has drawn significant attention in the open-source community. As of early 2025, its GitHub repository had surpassed 15,000 stars, making it one of the most popular fine-tuning tools in the Hugging Face ecosystem.

Community feedback centers on the following areas:

Positive feedback:

Training speedup is significant and consistent with the claimed 2-5x
Memory optimization is real, allowing larger models on smaller hardware
Integration with PEFT/LoRA is seamless, with low migration cost for existing code

Known limitations:

Supported model architectures still have boundaries; not all models benefit from Unsloth optimization
Documentation and examples for some advanced features (such as custom loss functions) are relatively limited
Extreme optimization sometimes requires trade-offs — in some cases gains in speed come with a slight risk to numerical stability

4. Ecosystem Contribution and Impact

4.1 Strengthening the Open-Source Ecosystem

Unsloth does not aim to replace Hugging Face or PyTorch; it exists as a performance layer on top of the ecosystem:

Seamless integration with libraries such as transformers, peft, and trl
Supports mainstream architectures including Llama, Mistral, Gemma, and Qwen
Remains open source (Apache 2.0 license), permitting commercial use

4.2 Driving AI Democratization

Unsloth’s significance goes beyond pure technical optimization. It represents a trend: pushing AI training capabilities from resource-intensive labs out to a broader developer audience. Similar projects (such as llama.cpp and Ollama) did the same on the inference side, and Unsloth completes this loop on the training side.

5. Objective Summary

Dimension	Traditional Fine-Tuning	With Unsloth
Training speed	Baseline	2-5x improvement
Memory usage	Baseline	Reduced by ~80%
Hardware requirements	Multi-GPU A100/H100	Consumer single-GPU feasible
Engineering barrier	Requires CUDA/distributed knowledge	One-line enablement
Precision loss	None (32-bit)	None (keeps 32-bit)
Ecosystem compatibility	Hugging Face	Fully compatible

Glossary

The following terms appear in order of first mention, covering key concepts and abbreviations referenced throughout the article for readers’ convenience.

Term	Full English Name	Definition
LLM	Large Language Model	Large language model: neural network models with a huge number of parameters (typically billions to hundreds of billions) that excel at understanding and generating natural-language text.
CUDA	Compute Unified Device Architecture	NVIDIA’s parallel computing platform and programming model that lets developers use GPUs for general-purpose computing; the primary acceleration foundation for deep learning training.
Triton	OpenAI Triton	A Python-like GPU programming language and compiler developed by OpenAI for writing high-performance custom GPU kernels, simpler than writing CUDA by hand.
RoPE	Rotary Position Embedding	A technique that injects positional information into the Transformer attention mechanism via rotation matrices; adopted by mainstream models such as Llama and Mistral.
Hugging Face	—	A company focused on open-source AI whose transformers library is the industry’s most mainstream framework for downloading and running pretrained models; in this article the term broadly refers to that ecosystem.
PyTorch	—	Meta’s (formerly Facebook) open-source deep learning framework, known for its dynamic computational graphs and one of the most widely used training frameworks in academia and industry.
Adam	Adaptive Moment Estimation	An optimization algorithm with adaptive learning rates that accelerates convergence by maintaining the first moment (momentum) and second moment (variance) of gradients; the default choice for training large models.
Flash Attention	—	An IO-aware exact attention algorithm that reduces data movement between GPU memory and HBM, significantly accelerating attention computation and reducing memory usage while remaining mathematically equivalent.
A100 / H100	—	NVIDIA data-center-class GPU models: A100 on the Ampere architecture and H100 on the Hopper architecture; both are mainstream hardware for large-scale model training.
RTX 4090	—	NVIDIA’s consumer flagship GPU with 24GB of memory, built on the Ada Lovelace architecture; a common GPU for high-end personal workstations.
PEFT	Parameter-Efficient Fine-Tuning	A category of fine-tuning methods that update only a small number of model parameters (rather than all of them), significantly reducing the compute and memory required for fine-tuning.
LoRA	Low-Rank Adaptation	One of the most popular PEFT methods; fine-tunes by injecting low-rank matrices alongside the original weights, training only the small set of newly added parameters.
FP16	Half-Precision Floating Point	16-bit half-precision floating point numbers, saving half the memory compared to 32-bit single precision; a common numeric format in large model training and inference.
4-bit Quantization	4-bit Quantization	A technique for compressing model weights from 16/32-bit to 4-bit representations, dramatically reducing model size and memory usage while recovering precision at compute time via dequantization.
Llama	Large Language Model Meta AI	Meta’s open-source LLM family (e.g., Llama 2, Llama 3); thanks to open weights and excellent cost-performance, one of the most popular base models for community fine-tuning.
Mistral	—	An open-source LLM family from the French company Mistral AI, known for efficient architectures with strong performance at small parameter counts.
Gemma	—	Google’s open-source lightweight LLM family, released for both research and commercial applications.
Qwen	—	An open-source LLM family from Alibaba’s Tongyi Qianwen team; multilingual with particularly strong performance in Chinese-language scenarios.
Apache 2.0	Apache License 2.0	A permissive open-source license allowing free use, modification, and distribution, including commercial use, with only a requirement to retain the original copyright notice.
llama.cpp	—	An open-source project porting Llama models to C/C++ with support for multiple quantization formats, focused on efficient inference on consumer CPUs.
Ollama	—	An open-source tool for running local large models that wraps model download, management, and inference into a flow letting users “run with one click” on personal computers.
GitHub Stars	—	The star count of a GitHub repository, a common indicator of a project’s community attention and popularity.
TRL	Transformer Reinforcement Learning	Hugging Face’s Transformer-based reinforcement learning training library, supporting alignment methods such as SFT, PPO, and DPO.

References

Unsloth official website: unsloth.ai
GitHub repository: github.com/unslothai/unsloth
Official documentation: docs.unsloth.ai
Hugging Face integration docs: huggingface.co/docs

This article is compiled from public technical documentation and community discussions and does not constitute technical advice.

Z.ai Tightens GLM Coding Plan Usage Policy: Non-Coding Uses Trigger Throttling and Bans

Tue, 21 Apr 2026 01:26:07 +0800

Event Overview

Recently, AI platform Z.ai updated the usage policy for its GLM Coding Plan subscription, strictly restricting the plan to coding scenarios. According to user communities and official announcements, non-coding uses (such as running AI Agents, role-playing, translating websites, and general chat) now trigger the platform’s risk-control mechanisms; repeated violations may lead to permanent account bans, and subscription fees are non-refundable.

This change has sparked widespread discussion in overseas AI user communities, and some industry practitioners read it as a signal that AI subscription business models are shifting from “subsidizing to win users” to “precisely acquiring training data.”

Policy Details: What the Official Statement Says

According to a screenshot of Z.ai’s official announcement shared by Reddit user JustSomeGuy3465, the new GLM Coding Plan policy includes the following core terms:

Usage restriction: The GLM Coding Plan is designed specifically for Coding Scenarios. If the system detects the subscription being used for requests unrelated to programming, the platform may restrict the relevant subscription benefits.
Risk-control measures: After violating the Usage Rules and triggering risk control, an account may face high-intensity throttling, account suspension, or permanent ban.
Violation red line: Violating the Usage Rules three or more times will result in a permanent ban, and subscription fees are non-refundable.

The user also noted that the 1302 and 1303 rate-limit errors many users have recently encountered are related to this policy update.

User Feedback: What’s Happening in the Community

In the Reddit post mentioned above, the poster explicitly warned: “If you are thinking about buying or renewing a Z AI coding plan subscription for anything other than coding: Don’t do it.”

Comments and related discussions show that some users have already reported their accounts being banned after using them for non-coding scenarios (such as running Agents via OpenClaw or engaging in role-play conversations), without receiving clear warnings in advance.

Notably, the plan previously attracted quite a few non-coding users because of its relatively low price. After the sudden policy change, these users became the most directly affected group.

Industry Commentary: Observations From the OpenClaw Founder

OpenClaw founder Peter Steinberger (@steipete) commented on the matter on X:

Steinberger’s view places this policy change in a larger industry context: some AI platforms’ low-priced subscriptions don’t make money from the subscription fee itself; their purpose is to obtain high-quality code data generated by users in real workflows.

Third-Party Analysis: Re-Examining the Subsidy Logic

Chinese tech commentator AYi (@AYi_AInotes), while sharing Steinberger’s view, expanded the analysis of this phenomenon:

The difference in data value: The commentator argues that real private code produced by users in daily work is far higher in quality than public code on platforms like GitHub, giving it greater data value for training next-generation code models.
The cost-structure conflict: Activities like running Agents, chatting, and role-playing consume a lot of GPU compute while producing no data returns useful for training the platform’s models. From the platform’s perspective, this usage pattern constitutes a “net cost.”
Industry trend judgment: The entire AI industry may be shifting from a phase of “acquiring user scale through subsidies” to a phase of “filtering high-value users through precise pricing.”

It should be noted that the above analysis is personal observation and inference, and has not been officially confirmed by Z.ai or the relevant platforms.

Observations and Summary

Based on available information, the facts of Z.ai’s policy adjustment are clear: the scope of the GLM Coding Plan has been significantly narrowed, and non-coding uses now face explicit risks. For users who have subscribed or plan to subscribe to this plan, the first step is to confirm whether their actual use cases match the platform’s definition.

As for the “data in exchange for subsidies” hypothesis proposed by Steinberger and the Chinese commentator, it offers one lens for explaining low-priced subscription strategies, but generalizing it as a universal rule for the whole industry requires more evidence. Different platforms have different cost structures, business models, and competitive strategies; a single case is not enough to capture the full picture.

For ordinary users, the more practical takeaway is: before subscribing to any AI service, read the updated terms of its usage policy carefully, to avoid having your account or service restricted due to policy changes.

Get a US Company and Physical Debit Card for $60: Anatomy of an Overseas Payment Path Born From a Tweet

Mon, 13 Apr 2026 03:58:11 +0800

A recent post on X (formerly Twitter) about “registering a US company at low cost and getting a physical debit card to reliably use overseas AI services” has drawn widespread attention. The post has received more than 270,000 views and over 3,000 bookmarks. This article parses its logical content, practicality, and potential risks from an informational perspective, for readers’ reference.

Source of Information

The following content is compiled from a post by X user Kai (@xkajon), published on April 11, 2026:

The original title of the post reads: “Easier than opening a Xiaohongshu store: Get a US company card for sixty bucks — including a shady trick for getting an EIN instantly.”

Core Logic Walkthrough

1. The Pain Point: Risk-Control Woes of Subscribing to AI Services with Virtual Cards

The post points out that the common practice among domestic users — subscribing to overseas services like Claude and OpenAI with virtual cards — is facing increasingly strict risk control. The root causes:

Shared card segments: Virtual card BINs usually belong to a “shared by thousands” card pool, easily flagged by risk-control systems;
Fabricated addresses: Billing addresses are often randomly generated and lack consistency;
Unstable IPs: Login IPs drift frequently and don’t match the registration information.

The author argues this makes accounts extremely vulnerable to bans, and since some virtual card platforms go unresponsive after blocking a card, users are forced to buy new cards — a cycle of harvesting.

2. The Solution: Registering a Wyoming LLC in Three Steps

To address the above issues, the author proposes a path to “build your own US corporate identity,” centered on registering an LLC (Limited Liability Company) in the state of Wyoming, with a total cost of roughly $60 plus a $20 annual fee, broken down into three steps:

Step 1: Hire a Registered Agent

Recommended provider: Wyoming Agents (wyomingagents.com)
Cost: about $20/year
Purpose: provides a physical Wyoming address, receives state government and IRS mail, and scans it to your email.

Step 2: Register the LLC on the State Government Website

Entry point: Wyoming Secretary of State - Business Filing
Action: File the Articles of Organization online, filling in the company name, registered agent information, and contact details
Fee: $60 (paid online)
Timeline: In fast cases, you can receive the receipt the same day

Step 3: Call the IRS to Obtain an EIN

Phone: +1 267-941-1099 (the IRS dedicated line for EIN applications from non-US domestic companies)
Action: Tell the representative you are a foreign-owned LLC applying for an Employer Identification Number (EIN). The information you provide must match what you used when registering in Wyoming.
Timeline: You can get your EIN number before you hang up.

3. Next Steps: Opening a US Business Bank Account

Once you have the EIN, you can apply for a business account with US-based or neobanks (such as Mercury, Relay, or Chase) and request a physical debit card. The author emphasizes that your payment profile is now completely different:

The card is in your name;
The billing address is the real registered company address;
Spending records resemble normal commercial procurement;
The risk-control system sees “a legitimate Wyoming company purchasing AI services.”

4. Why Wyoming?

The author summarizes three advantages in the post:

No state income tax: For LLCs with no US domestic business income, state tax burden is light;
Low annual report fees: Only a few dozen dollars in annual report fees per year are needed to keep the company active;
Shareholder information is not public: LLC member information is not shown in public records, offering stronger privacy.

Value Assessment

Practical Value: Medium-High

For users who rely on overseas AI/cloud services over the long term and stably, and who frequently run into payment risk control, this post offers an actionable path. The author not only names specific providers and government website links, but also provides the IRS phone number — a high information density rather than vague talk.

Operability: Medium, With Barriers

Although the author simplifies the process to “three steps,” real-world execution still involves several parts:

Filling out legal documents in English (Articles of Organization);
Conducting phone communication with the IRS in English;
Passing the KYC (Know Your Customer) review when opening a US business bank account;
International shipping or forwarding of the physical debit card.

For people without overseas experience or limited English communication skills, each step can be a barrier.

Risks and Pitfalls: Needs High Vigilance

Even if the process is feasible, the approach carries compliance and tax risks that cannot be ignored:

IRS tax filing obligations: As a foreign-owned US LLC, even with no US income, you may still need to file informational returns with the IRS (such as Form 5472). Missed or incorrect filings can carry heavy fines;
Company maintenance costs: Beyond the annual report fee, if the business is complex or requires professional tax preparation, the cost of hiring cross-border tax advisers can far exceed the registration cost itself;
Tighter bank account approval: In recent years, neobanks like Mercury and Relay have increasingly tightened scrutiny of “purely overseas users + shell companies”; an EIN does not guarantee 100% approval;
Motivation and compliance: If the sole purpose of registering the LLC is “to get a payment card,” it may raise questions at the levels of business substance, tax compliance, and bank risk control.

Disclaimer

This article only retells, organizes, and objectively analyzes the above public post; it has not been verified through actual practice. The process, costs, timelines, and risk warnings herein come from the original author’s sharing and our preliminary research, and do not represent final operational results.

Cross-border company registration, tax filing, and bank account opening involve complex legal and compliance issues, and each individual’s actual circumstances differ greatly. The post’s author is also not a professional legal or tax adviser. We strongly recommend that readers consult qualified cross-border lawyers, certified public accountants, or professional tax advisers before taking any action, and never make impulsive decisions based on website articles alone.

If readers choose to try it themselves, please research the compliance requirements of each step thoroughly and maintain clear expectations about the resulting maintenance costs, legal liabilities, and financial risks.

Tech Share on TorchTree

WeChat Official Account Traffic Revenue Rules in 2026: Long-Form Articles, Image Posts, and Channels

1. Five Key Changes to Official Account Rules in 2026

1. Public-Domain Recommendation Weight Soars; Follower Dividends Fade

2. Algorithm Metric Ranking: Completion Rate First

3. Longer Lifecycle for Quality Old Articles

4. AI Governance Upgraded: Non-Human Automated Creation Explicitly Banned

5. Traffic-Guiding Behavior Is Being Throttled

2. Traffic Mechanics of the Three Content Formats

1. Long-Form Articles: The Foundation of the Search Economy

2. Image Posts (Little Green Book / TieTu): An Officially Supported Traffic Dividend

3. WeChat Channels: Social Virality Driven, Real-Person Video Supported

3. Revenue Comparison Across the Three Formats

1. Official Account Traffic Monetization: Priced by Niche

2. Channels Creator Revenue Share: Rates Also Depend on Verticality

3. Little Green Book (Image Post) Revenue

4. How to Choose: Decide by Content Goals and Production Capacity

High-Knowledge-Density Vertical Accounts: Long-Form First, Little Green Book as Support

Seeding, Lifestyle, and Visual-First Accounts: Little Green Book Is the Mainstay

Accounts With Real-Person On-Camera Capability: Channels Is the Increment

The Common Bottom Line for All Accounts

5. References

Why Warp Went Open Source and What It Means

What Is Warp

Why Warp Chose to Open Source Now

What Is the Agent-first Workflow

What Open Source Brings to Users

Transparency and Auditability

Greater Customizability

More Model Choice

Free vs. Paid: What’s the Difference

How Different Users Should Choose

Warp vs. Closed-Source Competitors

The Product’s Future Direction

Short Term: Competing on Iteration Speed

Medium Term: Contesting the Open-Source ADE Niche

Long Term: Cloud Services as the Profit Core

Potential Risks and Uncertainties

Summary

References

Web-Based Presentations: Several Open-Source Options More Flexible Than Traditional Slides

1. The Core Framework Layer

reveal.js: The “De Facto Standard” for HTML Presentations

Slidev: The Developer’s Markdown Choice

Marp: The Lightest Markdown Option

Spectacle: A Presentation Option for the React Ecosystem

2. The AI Generation Layer

Presenton: The Most Feature-Complete AI Presentation Generator

PPTAgent: An Academic-Grade Generation Option

3. The Toolchain Layer: Format Conversion

DeckTape: High-Quality PDF Export for HTML Presentations

4. Selection Advice

5. A Workflow Worth Watching

Unsloth: The Accelerator Democratizing AI Training

1. Technical Innovation: Rebuilding Training Efficiency from the Ground Up

1.1 Weight Deprojection: The Core of Memory Optimization

1.2 Automatic Gradient Checkpointing and Flash Attention Integration

2. Leveling the Playing Field: From Enterprise to Individual

2.1 Enterprise Teams: Lowering Experiment Costs

2.2 Small Teams: Bypassing the Engineering Barrier

2.3 Individual Hobbyists: The Possibility of Consumer Hardware

3. Community Validation: Real-World Feedback

4. Ecosystem Contribution and Impact

4.1 Strengthening the Open-Source Ecosystem

4.2 Driving AI Democratization

5. Objective Summary

Glossary

References

Z.ai Tightens GLM Coding Plan Usage Policy: Non-Coding Uses Trigger Throttling and Bans

Event Overview

Policy Details: What the Official Statement Says

User Feedback: What’s Happening in the Community

Industry Commentary: Observations From the OpenClaw Founder

Third-Party Analysis: Re-Examining the Subsidy Logic

Observations and Summary

Get a US Company and Physical Debit Card for $60: Anatomy of an Overseas Payment Path Born From a Tweet

Source of Information

Core Logic Walkthrough

1. The Pain Point: Risk-Control Woes of Subscribing to AI Services with Virtual Cards

2. The Solution: Registering a Wyoming LLC in Three Steps