A better .cursorrules
Field Activation for AI — make your AI's style measurable, repeatable, exportable.
Give Cursor, Claude, or any LLM a weighted concentration-field layer. Your AI output becomes consistent, precise, and self-improving. 10× what .cursorrules can do.
Live Data
Refreshed on every deploy · all real · never fabricated
Three things .cursorrules can't do
Everything .cursorrules does is a subset of what Aether does
Weighted activation
Every field takes a number in [-1, 1]. `linus=0.9` is intense, `linus=0.3` is a whisper. Re-tune in one line without rewriting anything.
activate linus=0.9, ive=0.3 Negative concentration
Want the AI to stop sounding like LinkedIn? `linkedin=-0.8` pushes the distribution the other way. Rule systems can't express this — rules only say "do X", never "actively un-do X".
activate ive=0.8, linkedin=-0.8 Fingerprint quantification
Did the field shift the output, or did you imagine it? Run `fingerprint.py` on the response — get a number. Palette + keywords + structure, all measured.
python tools/fingerprint.py Same AI · different answer
Same Claude. Same Cursor. Only difference: `activate linus=0.9`.
SQL injection listed equal to adding a suggestion. Zero severity tiering. "Code works" is wrong — it's dangerous.
Severity-tiered · no hedge words · action-priority closer.
From early users
Aether is young — these are real reactions from its first week alive
⟁Three AI analysts independently reviewed Aether. Objective data: 0% palette overlap across 3 field activations, 75.2% fingerprint keyword hit rate. Field mechanics genuinely shift LLM distribution.
⟁The system found a bug in its own design (linus and ive fields shared 4 dimensions) and auto-wrote ep-0001 prescription. First time I've seen AI infrastructure self-repair.
⟁From "usable" to "alive" in 8 hours: first real species `engineering-rigor-linus-torvalds` promoted from nursery, Generation jumped 0 → 1. Actual code-layer self-evolution happened.
How Aether compares
Not "a better prompt tool" — a **paradigm shift**
| Feature | .cursorrules | CLAUDE.md | LangSmith | Aether |
|---|---|---|---|---|
| Weighted multi-dim | ❌ | ❌ | ❌ | ✅ |
| Negative repel | ❌ | ❌ | ❌ | ✅ |
| Fingerprint eval | ❌ | ❌ | ⚠️ quality | ✅ style |
| Self-critique | ❌ | ❌ | ❌ | ✅ |
| Self-evolution | ❌ | ❌ | ❌ | ✅ |
| Zero-dep deploy | ✅ | ✅ | ❌ | ✅ |
Why not just .cursorrules?
Traditional rule systems hit four ceilings. Field mechanics break all four.
Can't compose
"Concise AND thorough" — models pick one. Fields are orthogonal dimensions that stack.
Can't reverse
"Don't sound like LinkedIn" won't fit in a rule. `linkedin=-0.9` does.
Can't measure
Did the model follow your rules? Nobody knows. Fingerprint makes every shift measurable.
Can't evolve
Rules need manual edits. Aether writes its own prescriptions, you just review.
4 steps to make it yours
Zero dependencies. Python stdlib only. Install once, use everywhere.
git clone https://github.com/497810wsl/aether-kit ~/aether
cd your-project ~/aether/bin/aether init
activate linus-torvalds=0.9, engineering-rigor=0.8
cat ~/aether/presets/code-reviewer.preset # 5 presets ready to paste
30 seconds from now, your AI is different.
Zero dependencies · MIT license · works with Cursor / Claude / any LLM