2 things shipped at Simular in 2 weeks. Together, they’re the architecture - not the parts.
1️⃣ Sai can now run Claude Code.
Claude Code handles code generation - planner, coder, reviewer, tester. Sai opens a browser on its cloud desktop, navigates to your actual staging environment, and visually verifies the change works.
Cursor, Claude Code, Codex - none of them open your deployed product after git push to confirm it works. Sai closes that loop.
2️⃣ Simulang is live.
A JavaScript scripting framework for desktop automation. Deterministic, replayable, no LLM call per step - a 50-step browser flow costs roughly 1 model call instead of 50.
Goal-oriented abstraction, not step-by-step recording. Workflows survive UI changes. Ships with a translator that turns natural-language goals into Simulang.
Readable Simulang ⬇️
Here’s the part that actually matters:
Pure LLM-only agents don’t fail because the model is dumb. They fail because they’re non-deterministic. Drift, hallucination, breakage on the second run.
The fix isn’t smarter models. It’s specialised layers. Reasoning at the top > deterministic execution where determinism matters > visual verification where outcomes matter.
That’s the architecture we keep finding ourselves building.
If there’s interest, we’ll put together a starter set of real workflows. What do you want to see?
➡️ https://t.co/cG0SGI9NBc
@SimularAI@angli_ai
@cybernowxyz@Gradient_HQ that’s not from GN! kyc is not required and has not been mentioned at the moment
do keep yourself safe from scams. follow discord announcements for any updates
@GloriousKing225@Gradient_HQ that’s not us! kyc is not required and has not been mentioned at the moment
do keep yourself safe from scams. follow discord announcements for any updates