ContinuePRConversation reuses HandleQuickAssist for continuing a normal (persistent) PR conversation.

GenerateDraftPR uses the LLM's chat endpoint to generate a PR draft (stateless).

GeneratePreDraftPR obtains the latest commit message and diff, then returns them for usage in PR creation.

GenerateWhatSummary generates a summary of git diffs using the LLM (stateless).

HandleDCERequest handles ephemeral (DCE-driven) requests, returning the final text from a fresh ephemeral conversation, after running your DCE logic.

HandleQuickAssist returns the final LLM response for a persistent conversation, accumulating the streaming output behind-the-scenes into one string.

JSONHandler creates a handler for JSON requests/responses with unified error handling.

LoadDraftContext retrieves saved conversation context for a specific branch/commit.

SaveDraftContext saves conversation messages to disk for a specific branch/commit.

SetLLMClient allows injecting a different LLMClient (useful for testing or future extensions).

StartPRConversation initiates a new PR conversation with a commit message and diffs.

StartServer initializes and runs the HTTP server with full lifecycle management.

ServeCmd is the Cobra command to start the API server.

Request/Response types.

Request/Response types.

DefaultLLMClient implements the LLMClient interface using Ollama’s /api/chat.

Request/Response types.

Request/Response types.

LLMResponse represents the top-level structure from Ollama (non-streaming).

Request/Response types.

OllamaStreamChunk is used during streaming (partial response).

Request/Response types.

LLMClient defines the interface for interacting with the LLM (Ollama).