113K Stars In 2 Days: DeepSeek Harness vs Hermes

DeepSeek Harness collected around 113,000 GitHub stars in two days and has since passed 136,000.

Hermes took six months to reach roughly 226,000, and that already made it the fastest growing open source agent framework of the year.

So the deepseek harness vs hermes fight is not a challenger against an incumbent.

It is two of the biggest open source agent projects in existence, both free, both wanting to run your business.

Here is what each one is actually good at.

The short version

DeepSeek Harness wins on building, speed and how much you can change.

Hermes wins on memory, learning and reaching you wherever you already are.

Run both, give each a lane, and stop treating it as a fight.

What happened at launch

DeepSeek shipped two things at once: the V4 Pro model and a developer preview of DeepSeek Harness.

The model got attention. The harness is the bigger story.

Two days in it was near 113,000 stars.

One day after launch, 421 community plugins had already been published.

That is not normal, and it tells you the design landed with developers before the marketing did.

The design that caused it

"Everything is a plugin" is the official slogan and they mean it literally.

Model: plugin. Tools: plugin. Memory, session locks, search, sub-agents, scheduling: plugins.

Even the main loop that makes the agent think and act is a plugin.

Any piece can be pulled out, swapped or replaced without touching the core.

Most agents ship as a finished toy — great until you want to change one part, at which point you are stuck.

This ships as a box of bricks with instructions.

The plugin framework underneath is called Cordis, and it is not new. It has powered an open source chatbot project for years.

That is a large part of why a version 0.1 preview already feels solid rather than experimental.

What Hermes brings that nothing else does

Hermes came out of Nous Research on 25 February 2026.

It reaches you through Telegram, WhatsApp, Discord, Microsoft Teams and iMessage.

And it learns.

Hermes has a three-layer memory system and a self-improvement loop — when it solves a problem it saves what it learned as a skill, and next time the job goes faster because it remembers.

Most agents have no self-improvement loop at all.

That single feature is why Hermes stays installed on my machine no matter what launches next.

Three differences that decide it

One: which one breaks. Hermes is powerful and fiddly, and it breaks more. DeepSeek Harness breaks less, takes plain-text instructions, and makes setting the model easy. For a beginner that alone settles it.

Two: output and speed. For coding and building, DeepSeek Harness produces better results and replies noticeably faster — partly because DeepSeek V4 Pro is a frontier model built specifically for the harness. Hermes takes longer depending on the model you plug in, and on big coding tasks it times out. Hermes is not really a coding agent; it is for scheduled and smaller jobs.

Three: where it lives. This is the one nobody talks about. DeepSeek Harness is hosted locally as a website, so it fits the browser you already live in. Hermes is normally a desktop app, so it becomes another thing to click into next to Obsidian and your coding tool.

	DeepSeek Harness	Hermes
Stars	~113K in 2 days, 136K now	~226K in 6 months
Plugins on day one	421	Mature ecosystem
Best at	Coding, building, tooling	Scheduling, messaging, memory
Learns from past runs	No	Yes
Big coding jobs	Handles them	Times out
Customising the UI	Creator mode, live	Hours of coding
Runs in	Browser, locally	Desktop app
Maturity	v0.1 preview	6 months shipping

Creator mode versus hours of coding

In DeepSeek Harness you start a session, pick creator mode, and describe what you want changed — including the interface itself.

I asked for a corner panel with my three daily tasks. Built.

I asked for a mission control. Built.

I asked for a daily scheduler with times, days and repeatability, and watched it design the day semantics, plan the UI, validate options and write CSS live.

In Hermes I have built the equivalent — Hermes Astra, Hermes Muse, Hermes Oracle, which pulls trending news from Twitter — and every one cost me hours.

Hermes feels fixed. Harness feels moldable.

The gaps are closing faster than you think

People assume Hermes owns phone access. There is already a plugin pairing DeepSeek Harness with a mobile device.

People assume Hermes owns scheduling. The harness supports the same cron-style setup, and you can build the scheduler yourself in creator mode.

That is what 421 plugins in a single day actually buys — every "only Hermes can do this" claim has a shelf life.

What has not been replicated is the memory and the self-improvement loop.

🔥 Want both in one dashboard?

That is exactly what I build inside the AI Profit Boardroom — Hermes and DeepSeek Harness in one Agent OS, four coaching calls a week, 4,000+ members.

→ Get access here

Why I run both

I do not see this as a choice at all.

Both run inside my agent operating system and each gets the lane it is best at.

Hermes Oracle does daily research in the background. Hermes Muse analyses my content on a schedule. DeepSeek Harness builds.

The outputs feed the same workflows and I manage everything from one window instead of ten tabs.

The wider point is the one worth taking away.

Open source is winning here, the tools are free, and the models behind them can be free too — plug DeepSeek V4 Flash from OpenCode into either one and the cost goes to zero.

Which means it barely matters what harness launches next.

If your system is built to plug things in, every launch is an upgrade instead of a migration.

If you can only pick one

I would take DeepSeek Harness — more moldable, more intuitive, more fun, and it breaks less.

But it is v0.1, so expect it to change fast.

Hermes is the more mature system and the only one that genuinely learns.

Pick the harness to start, add Hermes when repetition starts costing you time.

FAQ

How many stars does DeepSeek Harness have? Around 113,000 within two days of launch, now past 136,000. Hermes has roughly 226,000 after six months.

Is DeepSeek Harness better than Hermes? For coding and building, yes. For memory, learning and messaging, Hermes is still ahead.

What is Cordis? The plugin framework underneath DeepSeek Harness. It has powered an open source chatbot project for years, which is why the preview already feels stable.

Can DeepSeek Harness message my phone? There is already a community plugin that pairs it with a mobile device, though Hermes does it natively.

Do I have to pick one? No. Run both behind one orchestrator and give each the work it is best at.

What the star count does and does not prove

A GitHub star is not a quality score.

Nobody starred DeepSeek Harness because they had benchmarked it against Hermes for a fortnight.

They starred it because the idea landed: a free, local, fully swappable agent with a frontier model built for it.

So read 113,000 in two days as a measure of appetite, not of finish.

Read 226,000 over six months as a measure of durability, because Hermes had to keep people through six months of real use to get there.

Those are different achievements and both are real.

The number that impressed me most is neither of them.

It is 421 plugins published one day after launch, because that is people building rather than bookmarking.

Where I would put your first hour

Do not spend it reading comparisons, including this one.

Install DeepSeek Harness, point it at a free brain, and give it one job you actually needed doing.

Then install Hermes and set one recurring task — a daily research brief is the obvious starter.

At the end of the week you will know which one you kept opening, and that is worth more than any scoreboard.

Both are free, so the only thing you are spending is attention.

About Julian

I'm Julian Goldie — AI entrepreneur, SEO expert, and founder of the AI Profit Boardroom (4,000+ members). I help business owners scale with AI agents, automation, and SEO.