StreamChat Implementation - Todo List

1. Add Genkit Dependency

• [x] Add github.com/firebase/genkit/go to go.mod
• [x] Run go mod tidy to fetch dependencies

2. Create internal/llm Package Structure

• [x] Create internal/llm/client.go with LLMClient interface
• [x] Create internal/llm/message.go with message types
• [x] Create internal/llm/models.go with provider/model configuration
• [x] Create internal/llm/genkit.go with Genkit client implementation

3. Define Core Interfaces and Types

• [x] Define LLMClient interface with StreamChat(messages []Message) (<-chan StreamEvent, error) method
• [x] Define Message struct with Role and Content fields
• [x] Define StreamEvent struct with Type (chunk/done/error) and Content fields
• [x] Define Role type constants (System, User, Assistant)

4. Implement Genkit Client Wrapper

• [x] Create GenkitClient struct implementing LLMClient
• [x] Implement provider-specific initialization (Anthropic, OpenAI, Gemini)
• [x] Implement StreamChat method using genkit's streaming API
• [x] Handle streaming response chunks
• [x] Test: Write unit tests for StreamChat with mock genkit responses

5. Create LLM Client Factory

• [x] Create NewClient(config *config.ResolvedConfig) (LLMClient, error) factory function
• [x] Support Anthropic provider with API key and model
• [x] Support OpenAI provider with API key and model
• [x] Support Gemini provider with API key and model
• [x] Test: Write unit tests for factory with different provider configurations

6. Update REPL for Multi-line Input

• [x] Replace bufio.Scanner with bubbletea textarea
• [x] Implement Ctrl+J to insert new line in message
• [x] Implement Enter to send message to LLM
• [x] Update tips to show keybindings

7. Integrate LLM Client with REPL

• [ ] Add llmClient field to replState struct
• [ ] Initialize LLM client in RunREPL with current config
• [ ] Update handleInput to send non-command input to LLM
• [ ] Create method to handle LLM streaming response

8. Implement Streaming Response Display

• [ ] Create styled output area for LLM responses using lipgloss
• [ ] Stream response chunks to the output in real-time
• [ ] Add visual indicator while streaming (e.g., spinner or "...")
• [ ] Handle final formatting of complete response

9. Handle LLM Client Updates

• [ ] Create updateLLMClient method on replState
• [ ] Call updateLLMClient when /model command changes config
• [ ] Ensure old client connections are properly closed

10. Error Handling

• [ ] Handle API key errors with clear user message
• [ ] Handle network errors gracefully
• [ ] Handle rate limit errors with retry suggestion
• [ ] Display error messages in REPL without breaking the loop
• [ ] Log errors for debugging
• [ ] Test: Write unit tests for error scenarios (invalid API key, network failure, rate limits)