feat: streaming chat, Gemini support, and embedding wrapper cleanup (PR 4 of 4)

cmd/pgedge-pg-mcp-svr/main.go

@@ -902,10 +902,18 @@ func main() {
 						Temperature: llm.Float(cfg.LLM.Temperature),
 					}
 				}
+				if cfg.LLM.GeminiAPIKey != "" {
+					providers["gemini"] = llm.Options{
+						APIKey:      cfg.LLM.GeminiAPIKey,
+						Model:       cfg.LLM.Model,
+						MaxTokens:   llm.Int(cfg.LLM.MaxTokens),
+						Temperature: llm.Float(cfg.LLM.Temperature),
+					}
+				}
 
 				if len(providers) == 0 {
 					return fmt.Errorf("LLM is enabled but no provider is configured; " +
-						"set at least one of anthropic_api_key, openai_api_key, or ollama_url")
+						"set at least one of anthropic_api_key, openai_api_key, gemini_api_key, or ollama_url")
 				}
 
 				p := proxy.New(proxy.Config{

docs/changelog.md

@@ -64,6 +64,21 @@ and this project adheres to
   exposed alongside the non-streaming endpoint. The non-streaming
   `/v1/chat` endpoint remains available for callers that prefer it.
 
+- The web chat interface now consumes the streaming endpoint
+  `/api/llm/v1/chat/stream` (Server-Sent Events) and renders the
+  assistant response incrementally as chunks arrive. The
+  non-streaming endpoint stays available for callers that prefer
+  it. A new `web/src/utils/sseChat.js` helper handles the SSE
+  parsing and assembles the final response into the same shape
+  the non-streaming endpoint returns, so the agentic chat loop is
+  unchanged.
+
+- The tools `search_knowledgebase`, `generate_embedding`, and
+  `similarity_search` now construct their embedding client
+  directly via `llm.NewClient` rather than going through the old
+  `embedding.NewProvider` wrapper. The `internal/embedding/`
+  package is deleted entirely.
+
 ### Fixed
 
 - Metadata loader now tolerates tables with zero columns
@@ -127,6 +142,12 @@ and this project adheres to
   token and needing access to multiple databases). The setting is
   honored by both the single-database and multi-database
   initialization paths. (#167)
+
+- Google Gemini is now a supported LLM provider. Configure via
+  `gemini_api_key` / `gemini_api_key_file` in the config file or
+  via the `PGEDGE_GEMINI_API_KEY` / `GEMINI_API_KEY` environment
+  variables.
+
 ### Fixed
 
 - Fixed port detection on Windows; the installer now reliably

go.mod

@@ -8,7 +8,7 @@ require (
 	github.com/fsnotify/fsnotify v1.9.0
 	github.com/google/uuid v1.6.0
 	github.com/jackc/pgx/v5 v5.7.6
-	github.com/pgEdge/pgedge-go-llm-lib v0.0.0-20260528080856-17b5f11c6044
+	github.com/pgEdge/pgedge-go-llm-lib v0.1.0
 	golang.org/x/crypto v0.49.0
 	golang.org/x/term v0.41.0
 	gopkg.in/yaml.v3 v3.0.1

go.sum

@@ -77,8 +77,8 @@ github.com/muesli/termenv v0.16.0 h1:S5AlUN9dENB57rsbnkPyfdGuWIlkmzJjbFf0Tf5FWUc
 github.com/muesli/termenv v0.16.0/go.mod h1:ZRfOIKPFDYQoDFF4Olj7/QJbW60Ol/kL1pU3VfY/Cnk=
 github.com/ncruces/go-strftime v0.1.9 h1:bY0MQC28UADQmHmaF5dgpLmImcShSi2kHU9XLdhx/f4=
 github.com/ncruces/go-strftime v0.1.9/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJmn9CehxcKcls=
-github.com/pgEdge/pgedge-go-llm-lib v0.0.0-20260528080856-17b5f11c6044 h1:VvcBNVFKWnVoLdJfZQhXkE42sefQWhNTeB8mPYU5U/k=
-github.com/pgEdge/pgedge-go-llm-lib v0.0.0-20260528080856-17b5f11c6044/go.mod h1:41rtSLcp/wwSUlBqetVHLQKisDZfzBmgSWt84WA+Eys=
+github.com/pgEdge/pgedge-go-llm-lib v0.1.0 h1:IiCWA99un19rwdB1hlDPOm2Ft+43LsCvp0oAhbBM/Nk=
+github.com/pgEdge/pgedge-go-llm-lib v0.1.0/go.mod h1:41rtSLcp/wwSUlBqetVHLQKisDZfzBmgSWt84WA+Eys=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94icq4NjY3clb7Lk8O1qJ8BdBEF8z0ibU0rE=

internal/config/config.go

@@ -446,7 +446,7 @@ type EmbeddingConfig struct {
 // LLMConfig holds LLM configuration for web client chat proxy
 type LLMConfig struct {
 	Enabled             bool    `yaml:"enabled"`                // Whether LLM proxy is enabled (default: false)
-	Provider            string  `yaml:"provider"`               // "anthropic", "openai", or "ollama"
+	Provider            string  `yaml:"provider"`               // "anthropic", "openai", "ollama", or "gemini"
 	Model               string  `yaml:"model"`                  // Provider-specific model name
 	AnthropicAPIKey     string  `yaml:"anthropic_api_key"`      // API key for Anthropic (direct - discouraged, use api_key_file or env var instead)
 	AnthropicAPIKeyFile string  `yaml:"anthropic_api_key_file"` // Path to file containing Anthropic API key
@@ -455,6 +455,8 @@ type LLMConfig struct {
 	OpenAIAPIKeyFile    string  `yaml:"openai_api_key_file"`    // Path to file containing OpenAI API key
 	OpenAIBaseURL       string  `yaml:"openai_base_url"`        // Base URL for OpenAI API (default: https://api.openai.com)
 	OllamaURL           string  `yaml:"ollama_url"`             // URL for Ollama service (default: http://localhost:11434)
+	GeminiAPIKey        string  `yaml:"gemini_api_key"`         // API key for Google Gemini (direct - discouraged, use api_key_file or env var instead)
+	GeminiAPIKeyFile    string  `yaml:"gemini_api_key_file"`    // Path to file containing Gemini API key
 	MaxTokens           int     `yaml:"max_tokens"`             // Maximum tokens for LLM response (default: 4096)
 	Temperature         float64 `yaml:"temperature"`            // Temperature for LLM sampling (default: 0.7)
 }
@@ -750,6 +752,12 @@ func mergeConfig(dest, src *Config) {
 		if src.LLM.OllamaURL != "" {
 			dest.LLM.OllamaURL = src.LLM.OllamaURL
 		}
+		if src.LLM.GeminiAPIKey != "" {
+			dest.LLM.GeminiAPIKey = src.LLM.GeminiAPIKey
+		}
+		if src.LLM.GeminiAPIKeyFile != "" {
+			dest.LLM.GeminiAPIKeyFile = src.LLM.GeminiAPIKeyFile
+		}
 		if src.LLM.MaxTokens != 0 {
 			dest.LLM.MaxTokens = src.LLM.MaxTokens
 		}
@@ -1011,6 +1019,7 @@ func applyEnvironmentVariables(cfg *Config) {
 	// 1. Try environment variables first (PGEDGE_ prefixed, then standard)
 	setStringFromEnvWithFallback(&cfg.LLM.AnthropicAPIKey, "PGEDGE_ANTHROPIC_API_KEY", "ANTHROPIC_API_KEY")
 	setStringFromEnvWithFallback(&cfg.LLM.OpenAIAPIKey, "PGEDGE_OPENAI_API_KEY", "OPENAI_API_KEY")
+	setStringFromEnvWithFallback(&cfg.LLM.GeminiAPIKey, "PGEDGE_GEMINI_API_KEY", "GEMINI_API_KEY")
 	// 2. If env vars not set and api_key_file is specified, load from file
 	if cfg.LLM.AnthropicAPIKey == "" && cfg.LLM.AnthropicAPIKeyFile != "" {
 		if key, err := readAPIKeyFromFile(cfg.LLM.AnthropicAPIKeyFile); err == nil && key != "" {
@@ -1024,7 +1033,13 @@ func applyEnvironmentVariables(cfg *Config) {
 		}
 		// Note: errors are silently ignored - file may not exist and that's ok
 	}
-	// 3. Direct config value (if set) is already in cfg.LLM.AnthropicAPIKey/OpenAIAPIKey from mergeConfig
+	if cfg.LLM.GeminiAPIKey == "" && cfg.LLM.GeminiAPIKeyFile != "" {
+		if key, err := readAPIKeyFromFile(cfg.LLM.GeminiAPIKeyFile); err == nil && key != "" {
+			cfg.LLM.GeminiAPIKey = key
+		}
+		// Note: errors are silently ignored - file may not exist and that's ok
+	}
+	// 3. Direct config value (if set) is already in cfg.LLM.AnthropicAPIKey/OpenAIAPIKey/GeminiAPIKey from mergeConfig
 	setStringFromEnv(&cfg.LLM.OllamaURL, "PGEDGE_OLLAMA_URL")
 	// Base URL overrides for LLM providers (useful for proxies)
 	setStringFromEnv(&cfg.LLM.AnthropicBaseURL, "PGEDGE_ANTHROPIC_BASE_URL")