@aeye/openai 0.3.9 → 0.3.10
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/aeye-openai.md +2 -0
- package/package.json +1 -1
package/aeye-openai.md
CHANGED
|
@@ -115,6 +115,8 @@ Retry config can also be overridden per request via `request.extra.retry` / `req
|
|
|
115
115
|
|
|
116
116
|
The provider only emits OpenAI-shaped strict schemas (`supportedStrictFamilies = {'openai'}`). Strict mode for a tool or structured output engages only when the resolved model's strict format is `openai` **and** the model declares the `toolsStrict` / `structured` capability; otherwise it silently falls back to a lenient (non-strict) schema. A shared `SchemaBudget` is used so tools and the response format cooperate on per-request limits.
|
|
117
117
|
|
|
118
|
+
That default is correct for api.openai.com, which speaks exactly one dialect. **A subclass that fronts other providers' models must widen it** — `supportedStrictFamilies` is the gate on the whole family resolution, so leaving it at `{'openai'}` sends every other family's tool schemas out as LENIENT no matter what the model declares. `OpenRouterProvider` sets `{'openai', 'anthropic', 'google'}` for this reason; `AWSProvider` does the same.
|
|
119
|
+
|
|
118
120
|
## Usage examples
|
|
119
121
|
|
|
120
122
|
```typescript
|