gochat/backend/internal/llm/fake_provider.go

// Package llm provides FakeLLMProvider — a zero-dependency LLM provider that
// returns canned responses for ChatCompletion, embeddings, and streaming.
//
// Usage: set FAKE_AI_ENABLED=true in the environment to activate; it overrides
// the standard provider manager so Captain/Copilot endpoints respond without a
// real LLM API key.
package llm

import (
	"context"
	"fmt"
	"os"
	"sync/atomic"
)

// FakeLLMProvider implements Provider with canned responses.
// Useful for integration testing, staging, and demo environments.
type FakeLLMProvider struct {
	// CallCount tracks how many ChatCompletion calls were made.
	callCount atomic.Int64
}

// NewFakeLLMProvider creates a new FakeLLMProvider.
// It is only active when the FAKE_AI_ENABLED env var is set to "true".
func NewFakeLLMProvider() *FakeLLMProvider {
	return &FakeLLMProvider{}
}

// IsFakeAIEnabled returns true when the environment variable FAKE_AI_ENABLED=true.
func IsFakeAIEnabled() bool {
	return os.Getenv("FAKE_AI_ENABLED") == "true"
}

// ChatCompletion returns a canned response echoing the last user message.
func (f *FakeLLMProvider) ChatCompletion(_ context.Context, req ChatRequest) (*ChatResponse, error) {
	n := f.callCount.Add(1)

	// Build a reply from the last user message.
	userMsg := ""
	for i := len(req.Messages) - 1; i >= 0; i-- {
		if req.Messages[i].Role == "user" {
			userMsg = req.Messages[i].Content
			break
		}
	}
	if userMsg == "" {
		userMsg = "Hello! I'm a fake AI assistant. How can I help you today?"
	}

	reply := fmt.Sprintf("[FakeAI #%d] Received your message. Here is a simulated response: %s", n, userMsg)

	return &ChatResponse{
		ID:      "fake-" + fmt.Sprint(n),
		Object:  "chat.completion",
		Created: 1700000000,
		Model:   "fake-ai-model",
		Choices: []ChatChoice{
			{
				Index: 0,
				Message: ChatMessage{
					Role:    "assistant",
					Content: reply,
				},
				FinishReason: "stop",
			},
		},
		Usage: TokenUsage{
			PromptTokens:     10,
			CompletionTokens: 5,
			TotalTokens:      15,
		},
	}, nil
}

// CreateEmbedding returns a zero-vector embedding of dimension 384.
func (f *FakeLLMProvider) CreateEmbedding(_ context.Context, req EmbeddingRequest) (*EmbeddingResponse, error) {
	if len(req.Input) == 0 {
		return &EmbeddingResponse{
			Object: "list",
			Data:   []EmbeddingData{},
			Model:  "fake-embedding-model",
		}, nil
	}

	dim := req.Dimensions
	if dim <= 0 {
		dim = 384
	}

	embeddings := make([]EmbeddingData, len(req.Input))
	for i := range req.Input {
		embeddings[i] = EmbeddingData{
			Object:    "embedding",
			Index:     i,
			Embedding: make([]float64, dim),
		}
	}

	return &EmbeddingResponse{
		Object: "list",
		Data:   embeddings,
		Model:  "fake-embedding-model",
		Usage: TokenUsage{
			PromptTokens:     len(req.Input),
			CompletionTokens: 0,
			TotalTokens:      len(req.Input),
		},
	}, nil
}

// ChatCompletionStream simulates streaming by delivering a single chunk.
func (f *FakeLLMProvider) ChatCompletionStream(ctx context.Context, req ChatRequest, onChunk func(StreamChunk) error) error {
	resp, err := f.ChatCompletion(ctx, req)
	if err != nil {
		return err
	}
	if len(resp.Choices) == 0 {
		return nil
	}

	chunk := StreamChunk{
		ID:      resp.ID,
		Object:  "chat.completion.chunk",
		Created: resp.Created,
		Model:   resp.Model,
		Choices: []StreamChoice{
			{
				Index: 0,
				Delta: StreamDelta{
					Role:    "assistant",
					Content: resp.Choices[0].Message.Content,
				},
				FinishReason: "stop",
			},
		},
	}

	return onChunk(chunk)
}