Sandy Kadyan | AI × Marketing

Marketing Ops on Claude Code for Agencies, Coaches and B2B founders.
Building in public. Every guide tested first.
Free 8 min training ↓

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

Without an agent:

npx hyperframes init demo
cd demo
npx hyperframes preview
npx hyperframes render

Timing lives in HTML attributes: data-start, data-duration and data-track-index on each clip.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

The install lines, from the HyperFrames README.

In Claude Code:
claude plugin marketplace add heygen-com/hyperframes
claude plugin install hyperframes@hyperframes

Then run /hyperframes:hyperframes. For other agents: npx skills add heygen-com/hyperframes

[0xMarioNawfal @RoundtableSpace]

HeyGen just open sourced HyperFrames, an Apache 2.0 framework that lets AI coding agents write HTML and render it into MP4 videos, and it sits at 62.7K GitHub stars after adding 5.1K in a week.

It's inspired by Remotion but runs on plain HTML instead of React
Headless Chrome and FFmpeg turn HTML, CSS, media and seekable animations into deterministic MP4
You can install it as a Claude Code plugin, an agent skill or a command-line tool
It needs Node.js 22 or later and FFmpeg

Source: https://t.co/uWikFN3UIg

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

It ships 21 skills. /hyperframes reads the request first and picks a workflow:

/product-launch-video from a site URL
/pr-to-video from a GitHub PR
/embedded-captions on a talking-head video
/remotion-to-hyperframes to port a Remotion project

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

To run it, clone the repo, then:

cp .env.example .env
docker compose up --build

Open localhost:5175. It also exposes an MCP endpoint, so Claude Code can call it:

claude mcp add --transport http openshorts http://localhost:8000/mcp

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

How OpenShorts picks and frames the clips, from its README:

Gemini reads the transcript and scene cuts and picks 3 to 15 moments. Two people on camera get stacked, captions on the seam. One speaker gets a face-tracked crop. Subtitles come from faster-whisper.

[Alba @albacnt]

🚨 ESTO ES UNA LOCURA

Un proyecto de GitHub que convierte vídeos largos en clips cortos listos para publicar en redes.

Se llama OpenShorts y es open source.

¿Lo mejor? 👇

→ Encuentra los mejores momentos del vídeo
→ Genera clips verticales
→ Añade subtítulos automáticamente
→ Prepara contenido para Shorts, TikTok y Reels

Todo en una misma herramienta.

Ideal para reutilizar podcasts, entrevistas o vídeos largos sin tener que editar cada clip desde cero.

Y sí, puedes ver el código en GitHub.

Enlace abajo 👇

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

Self-hosted is free under MIT, with no watermark and no upload limits. It needs Docker and 8GB+ RAM, and takes 5 to 8 min per 8-min video on CPU.

No Google key? Point LLM_BASE_URL at Ollama or LM Studio and the moment picker runs on your own model.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

Scheduled tasks run Claude on a timer: a code review every morning, a weekly dependency audit.

It needs a Pro, Max, Team or Enterprise plan. On Linux it installs with apt.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

What else the Claude Code desktop app does, from its quickstart docs:

After an edit, click the +12 -1 counter to open the diff. Comment on any line and Claude revises. Or type /code-review and it reviews its own changes.

[ClaudeDevs @ClaudeDevs]

We've been working hard on making the Claude Code desktop app great for running lots of sessions side by side.

It's now much quicker to start sessions, easier to split work across a few, and you can always see what each of them is doing. Here’s what's new:

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

Once a PR is open, it watches the CI checks and can fix failures, or merge once every check passes.

Long jobs can go to the cloud, so they keep running after you close the app.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

A reply adds the part after listing: directories re-scan servers on a schedule, so keep the endpoint, TLS and health check up or the listing drops.

The registry is still in preview, so breaking changes or data resets can happen.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

How to get an MCP server into the official registry, from its own quickstart:

The registry stores metadata only. Ship the server as a package first (npm, PyPI, NuGet, OCI, MCPB), or list a public remote endpoint with no package at all.

[饭特稀 | Fentix @fantexi997]

分享一个免费拿 DR 91 外链的小技巧 🚀

如果你的网站有一些适合 AI 调用的功能，可以把它们开发成 MCP Server，然后提交到 MCP 官方 Registry：

👉 https://t.co/cyX8EHDLz6

提交成功后，等几天你可能会发现：

🔍 很多 MCP 导航站、工具站、聚合站会自动同步并收录你的 MCP。

相当于：

提交 1 次 → 多个网站自动收录 → 顺便拿到一批外链 🔗

如果��正在做 SaaS、API 或开发者工具，可以顺手提交一下。

💰 免费
🔗 能拿外链
🤖 还能多一个 AI / MCP 的流量入口
一鱼多吃。

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

Then four commands:

brew install mcp-publisher
mcp-publisher init
mcp-publisher login github
mcp-publisher publish

init writes server.json, and its name must match mcpName. login prints a code you enter on GitHub.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

Best use, per the model card: agents that reason and call tools. Pair it with retrieval or web search for knowledge-heavy work.

16 languages, Hindi included. 131K context.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

A 24B model built to run agents on a laptop. Only 2B parameters are active per token, so it decodes 184 tokens/s on an Apple M5 Max.

It was trained with reinforcement learning inside OpenClaw, Hermes Agent, Pi and OpenCode, so it already knows their tools and prompts.

[Liquid AI @liquidai]

Today we release 𝙻𝙵𝙼𝟸.𝟻-𝟸𝟺𝙱-𝙰𝟸𝙱, the largest 𝙻𝙵𝙼𝟸.𝟻 family member. It is an efficient local model for laptops and PCs to work through multi-step tasks inside popular agent harnesses.

> 24B total parameters, 2B active per token
> Context expanded to 131K tokens
> A 128K-token vocabulary
> 33T pretraining tokens
> It is a reasoning model
> Very easy to fine-tune and customize
> Performs competitively with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks.
> Open base and post-trained weights on @huggingface 🤗

1/

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

To point Hermes Agent at it, serve it locally on port 8080, then:

hermes config set model.provider custom
hermes config set model.base_url http://localhost:8080/v1
hermes config set model.default LFM2.5-24B-A2B

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

How to set it up:
1. Sign up at https://t.co/3q6NZJJX6E, no card needed
2. Import your repo from GitHub, or upload a ZIP
3. Set the start command and deploy

5 slots share 2 GB RAM and 1.5 CPU, so spread them across small jobs. Nothing goes to sleep when idle.

[Sandy Kadyan | AI × Marketing @sandyaimarketer]

Free always-on hosting for small projects, if you sign up before 15 Oct, 17:00 Italy time.

What fits in the free plan:
- A Discord or Telegram bot that stays online
- A small API or worker in Node, Python, Java, PHP or Rust
- A Postgres, MySQL or Redis database

[painn @painn_x]

THIS IS F**KING GOLD, STOP PAYING FOR HOSTING

botkeep gives you 5 hosting slots FREE forever if you sign up before oct 15.

it replaces the thing you're already paying for.

what you get:
-5 slots free forever
-24/7 hosting
-bots + sites + dbs + apps

takes under 30 seconds:
1. sign up https://t.co/Y74f3nePt4
2. get founder free
3. done

hurry, free forever only kicks in if you're in before oct 15.

use this before the cutoff