Models & Temperature
Codality uses two models: a primary model for agent work and a fast model for lightweight checks.
Primary model
The primary model handles all agent work: reading code, reasoning about changes, writing code, and responding to feedback.
Default: gpt-5.2
Recommended models:
| Model | Best for | Notes |
|---|---|---|
gpt-5.2 |
General-purpose agent work | Default. Strong at coding, reasoning, and tool use. |
gpt-4.1 |
Cost-conscious usage | Capable but less expensive than 5.x series. |
o3 |
Complex reasoning tasks | Strong reasoning but slower and more expensive. |
Set the primary model in Admin → Prompts & Model → Model Configuration → Model ID.
Fast model
The fast model handles quick, cheap operations:
- Plan readiness checks
- Auto-generated plan titles
- AI description generation for priorities and kinds
Default: gpt-4.1-nano
Recommended fast models:
| Model | Notes |
|---|---|
gpt-4.1-nano |
Default. Extremely fast and cheap. |
gpt-4.1-mini |
Slightly more capable, still fast. |
Set the fast model in Admin → Prompts & Model → Model Configuration → Fast Model ID.
Temperature
Temperature controls randomness in model output. Lower values produce more deterministic output.
- Default: Model default (not set)
- Range: 0.0 to 2.0
You can set a global temperature in Admin → Prompts & Model → Model Configuration → Temperature.
Individual ticket kinds can override the global temperature. See Priorities & Kinds.
Tip: For coding work, lower temperatures (0.2–0.7) tend to produce more reliable code. Leave it at the model default unless you have a specific reason to change it.
Reasoning effort
Reasoning effort controls how much the model "thinks" before responding. This is set per priority level:
| Effort | Behavior |
|---|---|
low |
Quick responses, less deliberation. Good for urgent fixes. |
medium |
Balanced. Default for most work. |
high |
Deep reasoning. Good for complex bugs and low-priority thorough work. |
Reasonin effort is configured on each priority in Admin → Prompts & Model → Priorities. Individual ticket kinds can override the priority's reasoning effort.
API compatibility
Codality uses the OpenAI Responses API. Any API endpoint that implements this API works. Set GITHUB_API_BASE_URL if using a custom endpoint (note: this is a different env var than the OpenAI base URL — the OpenAI client uses the standard OPENAI_API_KEY).