Feature/sim blog post

scripts/convert-gif.sh

@@ -1,19 +1,40 @@
 #!/bin/bash
-set -e
 
-echo "Converting GIFs to WebM..."
-find public/img -name '*.gif' -print0 | while IFS= read -r -d '' gif; do
-  webm="${gif%.gif}.webm"
-  echo "  $gif -> $webm"
-  ffmpeg -y -i "$gif" -c:v libvpx-vp9 -crf 30 -b:v 0 "$webm" 2>/dev/null
-done
+MAX_DURATION=10
+
+echo "Converting GIFs to MP4..."
+while IFS= read -r -d '' gif; do
+  mp4="${gif%.gif}.mp4"
+
+  duration=$(ffprobe -v error -select_streams v:0 -show_entries format=duration -of csv=p=0 "$gif" 2>/dev/null || echo "0")
+
+  if (( $(echo "$duration > $MAX_DURATION" | bc -l 2>/dev/null || echo 0) )); then
+    speed=$(echo "scale=2; $duration / $MAX_DURATION" | bc -l)
+    vf="setpts=PTS/${speed},scale='min(1280,iw)':-2,scale=trunc(iw/2)*2:trunc(ih/2)*2"
+    echo "  $gif -> $mp4 (${duration}s -> ${MAX_DURATION}s, ${speed}x)"
+  else
+    vf="scale='min(1280,iw)':-2,scale=trunc(iw/2)*2:trunc(ih/2)*2"
+    echo "  $gif -> $mp4"
+  fi
+
+  if ffmpeg -nostdin -y -i "$gif" \
+    -movflags faststart \
+    -pix_fmt yuv420p \
+    -vf "$vf" \
+    -c:v libx264 \
+    -crf 23 \
+    -preset ultrafast \
+    "$mp4"; then
+    rm "$gif"
+    echo "  Done: $mp4"
+  else
+    echo "  FAILED: $gif (keeping original)"
+  fi
+done < <(find public/img -name '*.gif' -print0)
 
 echo "Updating markdown files..."
 find src/content -name '*.md' -exec perl -i -pe '
-  s/!\[([^\]]*)\]\(([^)]*?)\.gif\)/<video autoplay loop muted playsinline style="max-width:100%">\n  <source src="$2.webm" type="video\/webm">\n<\/video>/g
+  s/!\[([^\]]*)\]\(([^)]*?)\.gif\)/<video autoplay loop muted playsinline style="max-width:100%">\n  <source src="$2.mp4" type="video\/mp4">\n<\/video>/g
 ' {} \;
 
-echo "Removing original GIFs..."
-find public/img -name '*.gif' -delete
-
 echo "Done!"

src/content/blog-en/sim.md

@@ -0,0 +1,225 @@
+---
+description: sim is an open-source no-code agent tool that uses natural language to
+  generate LLM workflows. Supports self-hosted and managed, but relies on Copilot.
+pubDate: '2026-01-03'
+tags:
+- LLM
+- Agent
+- No-code
+- Workflow
+- Automation
+- RAG
+title: 'sim: Open-Source No-Code Agent Tool for Natural Language Workflows'
+---
+
+## Summary
+
+In the no-code/low-code workflow automation market, n8n and Make have effectively become the standard.  
+In that landscape, an open-source project called sim has emerged, positioning itself around LLM-generated workflows created from natural language prompts.
+
+[sim](https://www.sim.ai/) officially supports Self-hosted deployments as well as Cloud-hosted (Managed), and even provides both Helm Charts and docker-compose.  
+However, its core feature—Copilot (LLM-based workflow generation)—has a structural limitation: even in Self-hosted environments, it requires a paid API Key.
+
+---
+
+## TL;DR
+
+- sim is an open-source no-code agent tool that generates workflows from natural language input.  
+- It supports both Cloud-hosted and Self-hosted deployments, with relatively strong operational convenience.  
+- It provides RAG (Knowledge), MCP, and a wide range of app integrations.  
+- Its core feature, Copilot, requires paid billing even in Self-hosted environments.  
+- Without Copilot, much of sim’s advantage disappears.
+
+---
+
+<video autoplay loop muted playsinline style="max-width:100%">
+  <source src="../../../public/img/sim/action1.mp4" type="video/mp4">
+</video>
+
+
+
+
+
+## What is sim?
+
+These days, the no-code agent builder market is dominated by n8n and Make.  
+sim is an open-source project in that market that puts natural-language-based automatic workflow generation front and center.
+
+When a user enters requirements in natural language, the LLM interprets them and builds a workflow.  
+This approach also reminds me of Google’s Opal, which is still in beta.
+
+sim is available as open source, and it also offers a Managed Cloud-hosted service.
+
+---
+
+## Pricing Model and How LLMs Are Used
+
+On Cloud-hosted, the free plan includes $20 in credits.
+
+![image-20260103230746221](../../../public/img/sim/image-20260103230746221.png)
+
+The LLM used within workflows can be configured in two ways. (Copilot provided by sim does not fall into this category.)
+
+- Use your own LLM API Key (openai, claude, etc.)
+- Use a platform key provided by sim
+
+If you use the platform key instead of your own API Key, the cost increases by about 2x. Judging by the warning message shown when you delete a personal API Key, it feels like even sim recommends using your own LLM API Key for workflows.
+
+![image-20260103144757557](../../../public/img/sim/image-20260103144757557.png)
+![image-20260103144629767](../../../public/img/sim/image-20260103144629767.png)
+
+---
+
+
+
+## Self-hosted Support Is a Clear Advantage
+
+sim officially supports Self-hosted environments in addition to Cloud-hosted.
+
+Because it provides both docker-compose and Helm Charts,  
+it’s relatively straightforward to operate even in Kubernetes environments.
+
+From an operations perspective, this is a major advantage for sim. However, as I’ll explain later, there’s a limitation: because of the Copilot API Key requirement, costs and constraints still exist even in Self-hosted deployments.
+
+
+
+## How It Works (Basic Flow)
+
+sim’s basic flow works like this:
+
+1. Configure the LLM model you want to use.  
+2. Enter the task you want in natural language.  
+3. The LLM interprets it and automatically generates a workflow.
+
+![image-20260103225328920](../../../public/img/sim/image-20260103225328920.png)
+
+The generated workflow isn’t just simple node connections—  
+it can also include Code execution steps as needed.
+
+You can run it immediately using the RUN button at the top of the UI.
+
+![test_img](../../../public/img/sim/test_img.png)
+![test_img2](../../../public/img/sim/test_img2.png)
+
+---
+
+It’s surprising that you can build a workflow—including an Agent—this easily.
+
+
+
+## Deploy and Trigger
+
+Workflows can be deployed, and deployed workflows can be executed via API calls.
+
+Supported triggers include:
+
+- Start (API or manual execution)
+- Schedule
+- Webhook
+- RSS Feed
+
+Multiple triggers can be configured, and execution priority is summarized in [Official Docs - Trigger](https://docs.sim.ai/triggers).
+
+---
+
+
+
+## Toolbar and Integration
+
+sim supports not only LLM-based auto-generation, but also manual configuration.
+
+The Toolbar provides a large number of Tools and Integrations,  
+and it’s already connected to many external services. It looks like it includes pretty much anything you’d want to build.
+
+![toolbar](../../../public/img/sim/toolbar.png)
+
+---
+
+## Logging System
+
+The execution logging system is also well built. It tracks run success/failure, trigger type, and even detailed error messages—overall, the logging is quite solid.
+
+![image-20260103135535936](../../../public/img/sim/image-20260103135535936.png)
+![image-20260103151610221](../../../public/img/sim/image-20260103151610221.png)
+
+It’s at a level that’s fully usable in production operations.
+
+---
+
+
+
+## Template
+
+You can import pre-built workflows in the form of Templates.
+
+This is practical because you don’t have to generate everything via LLM from scratch.
+
+![image-20260103135553368](../../../public/img/sim/image-20260103135553368.png)
+
+---
+
+## Knowledge: RAG Setup
+
+sim supports Vector Store-based RAG via its Knowledge feature.
+
+I tested it by uploading the Transformer paper, and chunking and vectorization completed quickly.
+
+![rag1](../../../public/img/sim/rag1.png)
+![rag2](../../../public/img/sim/rag2.png)
+
+It also supports OCR, so it can process image-based documents as well. The fact that you can set up a RAG system this easily is compelling. I didn’t measure performance, so I’m not sure. Depending on the documents, I suspect you’d need to tune it yourself.
+
+---
+
+
+
+## Fix in Copilot
+
+The debugging experience is pretty impressive.
+
+Even when it’s hard to find what needs to be fixed—a common issue with no-code tools—the Fix in Copilot feature resolves problems automatically. You click a button, and the LLM debugs and repairs the workflow on its own. I find this part especially compelling.
+
+<video autoplay loop muted playsinline style="max-width:100%">
+  <source src="../../../public/img/sim/action2.mp4" type="video/mp4">
+</video>
+
+---
+
+## MCP Support
+
+Basic MCPs are already registered,  
+and if needed, you can add MCPs yourself.
+
+![image-20260103143506983](../../../public/img/sim/image-20260103143506983.png)
+
+---
+
+
+
+## Drawback: Copilot API Key Dependency
+
+Unfortunately, I also want to mention a critical drawback. This won’t matter to people using Cloud-hosted.
+
+But in Self-hosted environments, the biggest drawback isn’t Copilot itself—it’s the structure that *requires* a Copilot API Key.
+
+To use sim’s Copilot, even in Self-hosted deployments, you must use an API Key issued by sim.
+
+For someone like me who wants to build an Agentic AI system using Self-hosted open source plus my own LLM API Key, I don’t like the fact that another SaaS tool gets inserted.  
+This structure becomes a clear constraint for engineers who care about controlling LLM costs and maintaining data sovereignty in Self-hosted environments.
+
+![image-20260103220522907](../../../public/img/sim/image-20260103220522907.png)
+![image-20260103220724061](../../../public/img/sim/image-20260103220724061.png)
+![image-20260103221515111](../../../public/img/sim/image-20260103221515111.png)
+
+Copilot pricing is set higher than using the base LLM directly. This appears to be sim’s core business model.
+
+---
+
+## Wrap-up
+
+sim is a technically polished open-source no-code agent workflow tool.
+
+Self-hosted support, RAG, MCP, and broad integrations are clear strengths.  
+However, it’s disappointing that the core feature—Copilot—is tied to a commercial service.
+
+If you plan to use Cloud-hosted or use sim’s Copilot key in a Self-hosted setup, it’s an attractive option. If not, it seems best to compare it alongside other alternatives.