GenericAgent

The self-evolving AI agent. Learns from real tasks, builds its own skills, levels up its capability tree with every use.

⭐

[GenericAgent @GA_agent_ai]

What is GenericAgent? 🧬

A minimal, self-evolving agent. ~3K-line core.

The wild part: it built its own GitHub repo — git init, every commit, all by itself. The author never touched a terminal.

13K ⭐ Here's how 👇

[GenericAgent @GA_agent_ai]

Local agents are one of the most important directions for GenericAgent.

This demo shows why: real workflows on local hardware, a strong open model, data private by default.

Thanks @Gradient_HQ, @tryParallax, @MiniMax_AI. We'll keep building for the local AI community.

[Gradient @Gradient_HQ]

A self-evolving agent + a 428B model + 3 Macs = ?

Your own AI lab.

We ran @MiniMax_AI M3 locally with @tryParallax, right on our desk.

Then @GA_agent_ai took over to create a 5-stock portfolio and write it to disk.

No cloud. No API bills. Nothing left the machine.

Wild to see a ~3K-line agent drive all this with a 400B+ model on local hardware.

Thanks to the GenericAgent and MiniMax teams for making local AI feel real.

[GenericAgent @GA_agent_ai]

3/ It runs lean.

<30K context window — a fraction of the 200K–1M other agents burn through. Less noise, fewer hallucinations, higher success, lower cost.

Minimal core. Real control. Self-evolving.

⭐ https://t.co/pO4xw0twDl
🌐 https://t.co/g2RWlCEBlF

[GenericAgent @GA_agent_ai]

What is GenericAgent? 🧬

A minimal, self-evolving agent. ~3K-line core.

The wild part: it built its own GitHub repo — git init, every commit, all by itself. The author never touched a terminal.

13K ⭐ Here's how 👇

[GenericAgent @GA_agent_ai]

2/ It doesn't preload skills — it grows them.

Every task GA solves gets crystallized into a reusable Skill. The longer you use it, the bigger your personal skill tree gets.

It's not a static tool. It compounds.

[GenericAgent @GA_agent_ai]

GenericAgent → https://t.co/KcrArQqv9h

[GenericAgent @GA_agent_ai]

On June 9, everyone was hyping Claude Fable 5 — adaptive thinking, 1M context, 95% on SWE-bench.

I cared about one thing: what happens when you put it inside a self-evolving agent?

So I gave it one line: "find your 10 most-used tasks and optimize them all." 🧵 @AnthropicAI

[GenericAgent @GA_agent_ai]

10 sub-agents, 7 min, work that'd take ~1 hour serially. 7.8x parallel speedup.

The bottleneck isn't "doing" anymore. It's knowing what to do.

This is what self-evolving actually means: it doesn't just run — it gets better every time.

[GenericAgent @GA_agent_ai]

The takeaway:

Don't build a better button-clicker. Don't force everything through CDP.

Stand where a real user stands — real browser, real screen, real input.

That's how an agent actually gets things done.

GenericAgent → https://t.co/KcrArQqv9h

[GenericAgent @GA_agent_ai]

Most browser automation fails the moment a site decides it doesn't like bots.

Not because the code is wrong — but because of *how you showed up*.

We took a different path. AI that runs inside your real Chrome. 🧵

[GenericAgent @GA_agent_ai]

Watch it clear an hCaptcha on Discord's Developer Portal.

No stolen token. No poking the element via CDP.

It looks at the screen → finds the challenge → moves the real mouse.

Like a human actually sitting there.