COMPARISON

Why Aether has no substitute

If you're coming from .cursorrules, the next three sections are all you need. The deeper comparisons below are for engineers already using LangSmith / CLAUDE.md / etc.

vs .cursorrules · 8 dimensions

Using .cursorrules today? Here's where Aether is strictly better.

Dimension .cursorrules Aether Winner
Activation model Static rules · project-wide Dynamic activation · on demand ⟁ Aether
Weighting on / off [-1, 1] floating-point dials ⟁ Aether
Negative concentration impossible linkedin=-0.8 actively repels ⟁ Aether
Multi-persona composition rules fight each other orthogonal dimensions stack ⟁ Aether
Measurable effect unverifiable fingerprint.py returns a number ⟁ Aether
Self-critique / evolution manual edits forever critic → evolve → promote ⟁ Aether
Cross-session memory none collapses/ + mirror/ ⟁ Aether
Learning curve zero · one file one command · 30s = tie
CODE · side by side

Same goal · two ways of writing it

You want the AI to review code — be direct, be strict, drop the LinkedIn tone.

Writing .cursorrules Hardcoded · rules interfere · no way to say "less of X" 
# .cursorrules
- Be direct and terse.
- Always mention security risks first.
- Use bullet points.
- Don't hedge ("maybe", "consider").
- Avoid LinkedIn-style marketing.
- When reviewing: severity matters.
- Don't offer optional suggestions.
- ...(add 20 more lines)
Rules pile up · conflicts resolved by hand
→
Activating Aether fields One line = one persona · can negate · can stack · can measure 
// In your Cursor chat
activate linus-torvalds=0.9
activate engineering-rigor=0.9
activate linkedin=-0.8

// Verify it fired
python tools/fingerprint.py
3 lines ≈ the 20+ rules on the left, and provably effective
WHEN TO UPGRADE

Three situations · time to leave .cursorrules behind

Any one of these hitting home = you've outgrown it.

Scenario 01

Rules fight each other

Pain

You write "concise AND thorough" in .cursorrules — the model picks one every time, and which one depends on its mood.

Aether fix

Aether splits them into orthogonal dimensions: concision=0.8, thoroughness=0.7 both fire, no conflict.

activate concision=0.8, thoroughness=0.7
Scenario 02

Style drift

Pain

Same ruleset — today the AI sounds like Linus, tomorrow like a LinkedIn copywriter. You have no idea whether the model drifted or your rule stopped firing.

Aether fix

Aether runs fingerprint.py and gives you a mathematical distance. How much did it shift? Numbers, not vibes.

python tools/fingerprint.py --last 10
Scenario 03

Different domains need different voices

Pain

Writing frontend? You want Ive's restraint. Writing backend? You want Linus's bluntness. One .cursorrules file means editing it every single time.

Aether fix

Aether lets you switch personas inside one project — one line flips the weights. No file editing.

activate ive=0.8   # frontend mode
activate linus=0.9  # backend mode

Wider view · 8-tool matrix

Feature.cursorrulesCLAUDE.mdLangSmithPromptfooCLAUDE ProjectsLettaAnthropic SkillsAether
Weighted activation❌❌❌❌❌❌❌✅
Negative concentration❌❌❌❌❌❌❌✅
Field composition❌❌❌❌❌❌❌✅
Fingerprint eval❌❌⚠️ quality✅❌❌❌✅ style
Self-critique❌❌❌❌❌❌❌✅
Self-evolution❌❌❌❌❌❌❌✅
Persona import/export❌⚠️❌❌⚠️✅⚠️✅
Zero-dep deploy✅✅❌⚠️❌❌✅✅
Open source✅✅partial✅❌✅✅✅ MIT

Deep Dive · Per Competitor

Honest analysis · we acknowledge what each does well and what it can't do

.cursorrules

Cursor project-level hardcoded rules

✓ Strengths

  • Zero setup · one file
  • Deep Cursor integration

✗ Limits

  • No weights · no stacking
  • No negative repulsion
  • No quantified evaluation
  • Edit once = edit once
vs Aether

Aether extends cursorrules into a 16-dim stackable field space

CLAUDE.md

Anthropic single-file convention

✓ Strengths

  • Standardized · cross-session
  • Natively read by Claude ecosystem

✗ Limits

  • One file, one style
  • No multi-persona composition
  • No feedback loop
vs Aether

Aether fields coexist with CLAUDE.md, adding a multi-dim weight layer

LangSmith

LangChain prompt observability

✓ Strengths

  • Mature product · enterprise
  • Full trace/dataset/eval stack

✗ Limits

  • Evaluates quality/latency/cost, not style
  • Doesn't handle field stacking
  • Requires SaaS account
vs Aether

Aether does style alignment, LangSmith does quality eval. Complementary, not competing

Promptfoo

Open-source prompt testing framework

✓ Strengths

  • CLI-friendly · CI-ready
  • Multi-model comparison

✗ Limits

  • Static tests only, no dynamic field activation
  • No persona abstraction
vs Aether

Aether's fingerprint.py is similar in spirit but targets field fingerprints, not general quality

CLAUDE Projects

Anthropic session context

✓ Strengths

  • Official support · cross-session knowledge
  • Zero engineering cost

✗ Limits

  • No multi-dim personas
  • No negation
  • Closed, non-portable
vs Aether

Aether goes with you. Projects are Anthropic-locked.

Letta (MemGPT)

AI long-term memory framework

✓ Strengths

  • Flat memory pool · vector retrieval
  • Research-oriented

✗ Limits

  • Doesn't handle style/fields
  • Vector-dependent
  • Steep learning curve
vs Aether

Letta solves "remember more", Aether solves "style alignment". Theoretically composable.

Anthropic Skills

Anthropic official skill spec

✓ Strengths

  • Standardized · officially blessed
  • Ecosystem starting point

✗ Limits

  • Dead files · no evolution
  • No weighted composition
  • Aether MANIFESTO explicitly says "post-skill"
vs Aether

Aether actively challenges the skill paradigm · architects the next abstraction (fields · species · collapse)

⟁

One sentence

Everyone else is building "a better prompt tool".
Aether is building "the next thing after prompts".

Fields + fingerprint + self-evolution — this combination exists exactly once in the world.