Consolidate to GPT-5 only; upgrade deps; remove non-OpenAI paths

[shouryamaanjain]

Summary (why)

• Consolidates to a single, stable provider (OpenAI GPT-5) to reduce maintenance overhead and remove ambiguous behavior across multiple providers/paths.
• Modernizes the codebase to OpenAI SDK v1 with streaming + tool_calls for reliability and forward compatibility.
• Strips non-core UX and prompts to keep the tool focused on executing code via a single function tool.

Changes

• Dependencies: upgrade to latest stable
  • openai 1.105.x, rich 14.1.x, tiktoken 0.11.x, python-dotenv 1.1.x, pytest 8.4.x (dev)
  • Remove litellm, huggingface-hub, inquirer, requests, and all local/HF/Azure dependencies
  • poetry.lock updated
• Model/provider consolidation
  • Fixed model to "gpt-5" end-to-end
  • Removed Azure/custom api_base/LiteLLM/local model paths entirely
• OpenAI SDK v1 migration
  • Replace legacy/LiteLLM calls with OpenAI().chat.completions.create(...)
  • Implement streaming with tool_calls handling; accumulate streamed function.arguments into a JSON buffer and parse on finish
  • Fallback to non-streaming if the org/model disallows streaming
• CLI simplification
  • Only -y/--yes is supported; removed --fast, --local, --falcon, --model, --api_base, --use-azure and related env toggles
  • Console script entry updated to emplode.cli:cli_entry; python -m emplode works
• Token trimming
  • Replace tokentrim usage with tiktoken-based approximate trimming respecting context_window and max_tokens
• Files & cleanup
  • Delete emplode/get_hf_llm.py
  • Remove all Azure/custom base / local/HF logic and messaging
  • Minimize system_message.txt and README to core functionality only

Impact

• Breaking: non-OpenAI providers and local models are no longer supported; only OpenAI GPT-5 is available
• Requires OPENAI_API_KEY; Azure/custom api_base paths are removed
• Behavior is now deterministic around a single provider/model; fewer failure modes

Testing

• Locked with Poetry; built wheel; installed into a clean venv; CLI help shows only -y
• Basic run exercised both streaming and non-streaming paths (fallback when streaming not allowed)

Notes

• No remaining references to litellm, Azure, HF, or local model logic
• Documentation updated to reflect single-model policy and setup

₍ᐢ•(ܫ)•ᐢ₎ Generated by Capy (view task)