Search for a command to run...
Open source AI coding agent with free models or connect your own, like Claude, GPT, Gemini.
Hi 👋, I'm Dung Huynh Duc A passionate engineer from Singapore
🔭 I’m currently working in AirCarbon
👨💻 All of my projects are available at https://productsway.com
📝 I regularly write articles on https://productsway.com
📫 How to reach me dung@productsway.com
📹 I often publish my video every Sunday on IT Man Channel
No comments yet. Be the first to comment.
In the previous Power-Up posts, I've explored AI coding assistants, AI workflows, agent-based development, and AI launchers. This week, I want to talk about something that feels like the next step: au
AI coding tools are everywhere now: Claude Code, Codex, OpenCode, Amp, Pi, etc. The problem is not “which AI is best” anymore. The real problem is: How do you actually integrate them into your daily w
AI is no longer just a chatbot. For engineers, it has become a team of specialists.
In the previous post, I shared about OpenCode works. Now, let’s go one level deeper. This is the exact workflow I use daily to ship code faster, combining GitHub issue, Copilot Agent, and multiple AI
I just returned from the meetup and wanted to share a few thoughts while they’re still fresh. To be honest, I came in expecting to see something new, maybe a beta feature, an experimental workflow, or
After experimenting with Amp and Plannotator, I’ve been spending more time with OpenCode, and it’s quickly becoming one of the most thoughtfully designed AI coding agents I’ve used.
This post focuses on why OpenCode feels different, and when you might want to go even further with its experimental fork, Shuvcode.
One detail that deserves more attention: OpenCode ships with genuinely usable free models.
At the time of writing, free options include:
opencode/glm-4.7-free
opencode/minimax-m2.1-free
opencode/big-pickle
These aren’t toy demos. They’re good enough for:
Exploration and planning
Lightweight refactors
Reviewing unfamiliar codebases
For many workflows, you can stay productive without paying anything, and only switch to premium models when needed. That’s a smart, developer-friendly default.
OpenCode doesn’t lock you into a single model provider.
You can:
Use built-in free models
Bring your own provider via OpenRouter
Plug in OpenAI, Claude, Gemini, GLM, Qwen, and more
Switch models based on task, cost, or latency
This flexibility keeps OpenCode future-proof and practical. You optimize for your constraints, not the tool’s.
If you enjoy being close to the frontier, Shuvcode is worth your attention.
Shuvcode is a fork of OpenCode that serves as a proving ground for new features before they are integrated into the main branch. If you want to try ideas early, accept some instability, and give feedback upstream, this is the version to run.
Think of it as:
OpenCode → stable, production-ready
Shuvcode → experimental, fast-moving, opinionated
For developers who enjoy testing what’s next, Shuvcode feels like the right playground.
One thing I really love about OpenCode is its two default agents:
Plan – read-only, safe, exploratory
Build – full access, can modify files
This separation sounds simple, but it fundamentally changes how you work with AI.
When you’re in Plan mode, you can:
Explore the codebase
Ask “what if” questions
Review ideas without fear of side effects
Once you switch to Build, changes are intentional and explicit. No accidental edits. No surprise diffs.
It encourages a mindset that mirrors how senior engineers actually work: understand first, change second.
Another standout feature is sub-agent visibility.
When OpenCode delegates work to sub-agents:
You can see which sub-agent is running
Navigate into it
Observe what it’s doing in real time
Once a sub-agent finishes, it sends a report back to the parent agent. You can then move between agents to understand how conclusions were reached, not just the final answer.
This transparency builds trust—and makes debugging AI behavior far easier than black-box systems.
This is a feature I already appreciated in Amp, and OpenCode does it well too.
You can:
Send it to teammates or collaborators
Toggle sessions between public and private at any time
That makes it easy to:
Review AI-assisted decisions asynchronously
Share debugging context
Collaborate without screen recording or copy-pasting logs
For team workflows, this is quietly powerful.
OpenCode isn’t trying to be flashy. It’s trying to be correct. The agent model, execution visibility, provider freedom, and thoughtful defaults all point in the same direction:
AI that fits real developer workflows, not demos.
If you want stability, start with OpenCode.
If you want to see what’s next, explore Shuvcode.
Either way, this is a strong signal of where AI coding tools are heading, and one worth paying attention to.
