Hi ! My name is Jérémy, and I'm one of those late-night builders, juggling with 6 claude code terminals.
TLDR; I have built a "Harness Engineering" layer, and it's open source. Now you can spin a Hermes/Deepseek agent from Claude Code, Codex or any CLI.
Open to contributions and feedbacks.
Why did I build this ?
I was very tired of having to babysit my agents every few minutes to make sure that I was not stuck on a "Should I proceed ?" question or anything that would block the dev for nothing ("YEAH I WANT YOU TO PROCEED !!").
So I had to find a solution that would allow me to get better autonomy for long-running tasks.
What is AgentProto CLI ?
So, I haven't built only a "Claude Code" Orchestration layer, but an harness agnostic layer.
Meaning, you can spawn claude-code, but also Hermès, Codex or the Claw. So, fear not the "vendor lock-in" anymore, as you can just use ANY harness on top/bottom of it, and any models/providers. Anthropic, OpenAI, OpenRouter, Kimi, Zai, Xai and so on.
Well... It works "on my computer" but I'm still making the process smooth for you. So, if you're brave enough to go at the edge of CLI-agnostic Orchestration, this is a project that will motivate you. You might have that "Cursor" feeling, for those that remembers how drastically it changed their way.
I have been pouring so much work, and happily, into it and I don't want Internet to miss that, as I'm confident this "agnostic" layer will allow much greater use cases than locked-in CLI.
Can I help ?
Oh dear ! Yes, but testing out, leaving your feedback, opening issues etc.
This can get reaaaally big, and all contributors welcome.
I have been dogfooding AgentProto and I believe that there is, as of today, nothing that compares to it for developer productivity. Happy to be proved wrong !
If you want to be at the forefront of Harness Engineering, I'll be happy to chat with you. DM or jeremy#agentik#net
3 comments
[ 7.3 ms ] story [ 23.3 ms ] threadTLDR; I have built a "Harness Engineering" layer, and it's open source. Now you can spin a Hermes/Deepseek agent from Claude Code, Codex or any CLI.
Open to contributions and feedbacks. Why did I build this ? I was very tired of having to babysit my agents every few minutes to make sure that I was not stuck on a "Should I proceed ?" question or anything that would block the dev for nothing ("YEAH I WANT YOU TO PROCEED !!").
So I had to find a solution that would allow me to get better autonomy for long-running tasks.
What is AgentProto CLI ? So, I haven't built only a "Claude Code" Orchestration layer, but an harness agnostic layer.
Meaning, you can spawn claude-code, but also Hermès, Codex or the Claw. So, fear not the "vendor lock-in" anymore, as you can just use ANY harness on top/bottom of it, and any models/providers. Anthropic, OpenAI, OpenRouter, Kimi, Zai, Xai and so on.
Read more : https://cli.agentproto.sh
Well... It works "on my computer" but I'm still making the process smooth for you. So, if you're brave enough to go at the edge of CLI-agnostic Orchestration, this is a project that will motivate you. You might have that "Cursor" feeling, for those that remembers how drastically it changed their way.
Can I read the code ?
Everything is open source -and very active- available here https://github.com/agentproto/ts
I have been pouring so much work, and happily, into it and I don't want Internet to miss that, as I'm confident this "agnostic" layer will allow much greater use cases than locked-in CLI.
Can I help ?
Oh dear ! Yes, but testing out, leaving your feedback, opening issues etc.
This can get reaaaally big, and all contributors welcome.
I have been dogfooding AgentProto and I believe that there is, as of today, nothing that compares to it for developer productivity. Happy to be proved wrong !
If you want to be at the forefront of Harness Engineering, I'll be happy to chat with you. DM or jeremy#agentik#net
To learn more about AgentProto :
A bit of reading : https://dev.to/agentiknet
Thanks for reading, please share to your fellow agents to get their thoughts !