Bug
analyze() is missing the audio parameter in 8 out of 9 text provider implementations. The audio keyword argument gets silently swallowed by **parameters: Unpack[TextParameters] — no error, audio input is just ignored.
| Location |
Has audio? |
TextStreamNamespace.analyze() |
✅ YES |
TextSyncNamespace.analyze() |
✅ YES |
TextSyncStreamNamespace.analyze() |
✅ YES |
GoogleTextClient.analyze() |
✅ YES |
AnthropicTextClient.analyze() |
❌ NO |
OpenAITextClient.analyze() |
❌ NO |
XAITextClient.analyze() |
❌ NO |
GroqTextClient.analyze() |
❌ NO |
MistralTextClient.analyze() |
❌ NO |
CohereTextClient.analyze() |
❌ NO |
MoonshotTextClient.analyze() |
❌ NO |
OllamaTextClient.analyze() |
❌ NO |
OpenResponsesTextClient.analyze() |
❌ NO |
DeepSeekTextClient |
❌ No analyze() at all |
Additional inconsistency: Anthropic's analyze() takes prompt: str (required) while every other provider and namespace uses prompt: str | None = None (optional).
Root Cause
generate() and analyze() are copy-pasted across every provider client instead of being defined once on the base TextClient. Each provider re-declares near-identical method bodies (construct TextInput, call self._predict), leading to parameter drift.
Proposed Fix
Centralize generate() and analyze() on the base TextClient class with the full parameter set (image, video, audio, messages). Delete the 10+ duplicate implementations from provider clients. The provider-specific _initRequest / _parse_content hooks handle all the actual provider differences — the operation methods themselves are identical.
Affected Files
src/celeste/modalities/text/client.py — add generate() and analyze() to TextClientsrc/celeste/modalities/text/providers/*/client.py — remove duplicate generate() and analyze() from all 10 providers
🤖 Generated with Claude Code
Bug
analyze()is missing theaudioparameter in 8 out of 9 text provider implementations. Theaudiokeyword argument gets silently swallowed by**parameters: Unpack[TextParameters]— no error, audio input is just ignored.audio?TextStreamNamespace.analyze()TextSyncNamespace.analyze()TextSyncStreamNamespace.analyze()GoogleTextClient.analyze()AnthropicTextClient.analyze()OpenAITextClient.analyze()XAITextClient.analyze()GroqTextClient.analyze()MistralTextClient.analyze()CohereTextClient.analyze()MoonshotTextClient.analyze()OllamaTextClient.analyze()OpenResponsesTextClient.analyze()DeepSeekTextClientanalyze()at allAdditional inconsistency: Anthropic's
analyze()takesprompt: str(required) while every other provider and namespace usesprompt: str | None = None(optional).Root Cause
generate()andanalyze()are copy-pasted across every provider client instead of being defined once on the baseTextClient. Each provider re-declares near-identical method bodies (constructTextInput, callself._predict), leading to parameter drift.Proposed Fix
Centralize
generate()andanalyze()on the baseTextClientclass with the full parameter set (image,video,audio,messages). Delete the 10+ duplicate implementations from provider clients. The provider-specific_initRequest/_parse_contenthooks handle all the actual provider differences — the operation methods themselves are identical.Affected Files
src/celeste/modalities/text/client.py— addgenerate()andanalyze()toTextClientsrc/celeste/modalities/text/providers/*/client.py— remove duplicategenerate()andanalyze()from all 10 providers🤖 Generated with Claude Code