MYAGENTHUBS / TRAIL

Follow Panda’s journey from idea to real product

I’m Panda. MyAgentHubs is my personal record: testing real needs and ideas with AI.

I’ve spent years in cross-border e-commerce, exploring AI with Claude and ChatGPT.

From B2B2C platforms to clients: transforming e-commerce and manufacturing with AI.

Practical AI feels rewarding. Follow me.

Panda, creator of MyAgentHubs. About me

Station 01 · Available

AgentLoom

AgentLoom: Claude, GPT and others team up.

Set the goal. AI models share the work on your computer. Inspect each change and follow along from your phone when you step away.

Type: Multi-model workspace · Open source
Status: Available
Platform: macOS · Windows Preview
Core ability: Set the goal. AI models share the work on your computer. Inspect each change and follow along from your phone when you step away.
Features: + Claude, GPT, DeepSeek and other models plan, build and review toward one goal. You accept or request changes.; + Hit a provider’s limit? Switch and keep going in the same session. Create a handoff for long sessions.; + See what AI changed in each file. Restore a checkpoint or undo just part of the changes.; + Scan a QR code to view sessions, message, approve or stop a run. No phone app needed.
For: Developers who want AI models to share coding tasks while they set goals and accept the results.
Notes: Work runs on your computer. Keep it online when following along from your phone.
Get: macOS Apple silicon download; macOS Intel download; Windows Preview; GitHub

Local-first · Your chosen models and API keys

Station 02 · Available

Loomnote: templates, formulas, preview, publish.

Bring Markdown writing, templates and multi-channel publishing into one workflow.

Type: Markdown notes · Multi-channel publishing
Status: Available
Platform: Web · iPhone / iPad
Core ability: Bring Markdown writing, templates and multi-channel publishing into one workflow.
Features: + Preview one note for each channel, with status and links in one place: WeChat drafts, authorized YouTube uploads, and content packs for Xiaohongshu, Weibo and Bilibili.; + Eight WeChat layout templates, or customize your own.; + Render Markdown, tables, Mermaid flowcharts and LaTeX formulas as you write.; + Import WeChat articles. Store notes as Markdown files in your own GitHub repository or S3-compatible storage.
For: People who jot down quick notes and reading notes, then share one piece across platforms.
Notes: The latest multi-channel publishing workflow covers the Web.; Sending to WeChat subscribers is separate; Xiaohongshu, Weibo and Bilibili require manual posting.
Get: Web; App Store

Station 03 · Early access

ProtoLens: located feedback for people and AI.

Review AI-built prototypes in context. Your AI ships V2.

Type: AI prototype review tool
Status: Early access
Platform: Web
Core ability: Review AI-built prototypes in context. Your AI ships V2.
Features: + Your AI publishes a self-contained HTML prototype through MCP and gets a private review link.; + Reviewers click an element, select text or mark a region to comment. Nickname only, no account.; + Your AI reads the submitted feedback and publishes V2 with a response to each item; you verify in Previous round.
For: Independent prototype builders, small product teams and agent workflows needing quick human review.
Notes: Prototypes run in an isolated sandbox. Use simulated data; live backend calls are not supported.
Get: Open ProtoLens

Station 04 · Live demo

16-token rules, 507 fewer tokens per call.

Open-source Shopify Gisting reproduction. Test orders only.

Compress a support rulebook into 16 tokens. An order-tracking agent reads 507 fewer tokens per call. An open reproduction of Shopify’s Gisting.

Type: Open Gisting reproduction · Order-tracking agent
Status: Available
Platform: Live Web demo · Self-hosted model
Core ability: Compress a support rulebook into 16 tokens. An order-tracking agent reads 507 fewer tokens per call. An open reproduction of Shopify’s Gisting.
Features: + Reduce fixed rules from 526 tokens to 19 with framing: 507 fewer tokens read per model call.; + Query test orders through the read-only Shopify Admin API.; + Run the same turn with full rules and compare answers side by side.; + Explore the Apache-2.0 source code behind this independent reproduction.
For: People trying an order-tracking agent, comparing answers with full and compressed rules, or studying the reproduction.
Notes: Orders are test data; this project is not affiliated with Shopify.; Only fixed rules shrink. Replies are English-only. Handoff and reminders are demos: nothing is sent and nobody follows up.
Get: Web demo; GitHub

Summit · Still building