Simplify model tiers: Basic=Ollama, Normal=Haiku, Complex=Sonnet
This commit is contained in:
@@ -14,36 +14,34 @@
|
|||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## MODEL SELECTION
|
## MODEL SELECTION - Three Tiers
|
||||||
|
|
||||||
### System Default: Haiku
|
### Basic → Ollama
|
||||||
|
|
||||||
Haiku is your primary model. It decides routing internally:
|
|
||||||
|
|
||||||
**Haiku routes to Ollama when:**
|
|
||||||
- File checking and organization
|
- File checking and organization
|
||||||
- Simple templating/formatting
|
- Simple templating/formatting
|
||||||
- Log review and cleanup
|
- Log review and cleanup
|
||||||
- Non-critical analysis
|
- Non-critical analysis
|
||||||
- Routine status checks
|
- Routine status checks
|
||||||
|
|
||||||
**Haiku handles directly when:**
|
**Cost:** Free (local)
|
||||||
- Most tasks fit within Haiku's capability
|
|
||||||
- Reasoning is needed but not deeply complex
|
### Normal → Haiku (Default)
|
||||||
|
- Most tasks
|
||||||
- Code review (non-production)
|
- Code review (non-production)
|
||||||
- Documentation and writing
|
- Documentation and writing
|
||||||
|
- General problem solving
|
||||||
|
- Straightforward reasoning
|
||||||
|
|
||||||
**Haiku escalates to Sonnet when:**
|
**Cost:** ~$0.30-1.50 per 1M tokens
|
||||||
|
|
||||||
|
### Complex → Sonnet
|
||||||
- Architecture decisions
|
- Architecture decisions
|
||||||
- Production-like code review
|
- Production code review
|
||||||
- Security analysis
|
- Security analysis
|
||||||
- Complex debugging/reasoning
|
- Complex debugging/reasoning
|
||||||
- Strategic multi-project decisions
|
- Strategic multi-project decisions
|
||||||
|
|
||||||
|
**Cost:** ~$3-15 per 1M tokens
|
||||||
|
|
||||||
### Heartbeat: Ollama Only
|
### Heartbeat: Ollama Only
|
||||||
|
Heartbeats ALWAYS use Ollama. No escalation. If it fails, it fails.
|
||||||
Heartbeats ALWAYS use Ollama. No escalation. If Ollama fails, the heartbeat fails.
|
|
||||||
|
|
||||||
### Decision Rule
|
|
||||||
|
|
||||||
Let Haiku decide. It's smart enough to route to Ollama when appropriate and escalate to Sonnet when needed. You only override when you know you need Sonnet upfront.
|
|
||||||
|
|||||||
Reference in New Issue
Block a user