Feature Request: Auto-Adaptive Settings Based on Connected Model Size

[scottnzuk]

Feature Request: Auto-Adaptive Settings Based on Connected Model Size

Summary

Automatically adjust critical environment settings based on the size and type of the connected model.
The goal is to prevent context/token overflows and eliminate deadlocks caused by user misconfiguration — ensuring seamless operation without manual adjustment.

Details

Implement auto-adaptive rules triggered when connecting to a model, based on model size and model name properties:

---

Model Size Based Adjustments

16k Models or smaller

• Enable editing via diff → ❌ disabled
• Workspace file context limit → 20
• Open tab context limit → 20
• More aggressive output truncation.

32k Models

• Enable editing via diff (enable_editing_through_diff) → ✅ enabled
• Temperature (temperature) → set to 0.03
• Open tab context limit → 20
• Workspace file context limit → 100
• Terminal output line limit → 800
• File read auto-truncate threshold → 1000 lines

65k Models

• Enable editing via diff → ✅ enabled
• Temperature → 0.03
• Open tab context limit → 50
• Workspace file context limit → 200
• Terminal output line limit → 3000
• File read auto-truncate threshold → 2000 lines

1M+ Models

• Enable editing via diff → ✅ enabled
• Temperature → 0.03
• Open tab context limit → 200
• Workspace file context limit → 500 (or scaled accordingly)
• Terminal output line limit → 5000
• File read auto-truncate threshold → 2000 lines

---

Model Name Based Adjustments

Models with free in the name:

• Enforce rate limit → max 1 message every 5 seconds
  (to prevent API abuse throttling or bans)

---

Terminal Output and File Reading Defaults

Both Terminal Output Limit and File Read Auto-Truncate Threshold
should scale proportionally with the model context size:

Model Size	Terminal Output Limit	File Read Auto-Truncate
16k	1500 lines	1000 lines
32k	2500 lines	2000 lines
65k	3000 lines	2000 lines
1M	5000 lines	2000+ lines

---

Why This Matters

• Deadlock Prevention: Avoid the "over context limit" error that forces a hard reset of the environment.
• User Experience: Transparent auto-configuration without requiring manual tuning.
• Efficiency: Safer defaults help maximize token efficiency without surprises.
• Stability: Smooth, reliable behavior across different model sizes.

---

Bonus (Optional Future Extension)

• Allow users to override with a "manual mode" toggle if they wish to tune settings themselves.
• Provide a status banner when settings are auto-adjusted ("Settings tuned for 32k model").

---

Thank You

This would massively improve the developer experience, especially when switching frequently between models of different sizes and capabilities!

[BradKML]

Could be applied to OpenRouter first and foremost, since the same model can have different context size