fix: multi-host fallback integration and accurate health status

CHANGELOG.md

@@ -7,6 +7,24 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.0.1] - 2026-02-05
+
+### Fixed
+
+- **Multi-host fallback now works in main chat loop** - HostPool is properly wired to Assistant for automatic
+  fallback when the primary host becomes unavailable
+- **Accurate host health status display** - `/hosts` now shows correct healthy/total count; hosts are only marked
+  healthy after connectivity verification via `DetectModels()`
+- **Fallback notification** - Users now see clear feedback when fallback occurs:
+  "Primary host unavailable, switched to: <host>"
+- **Fallback failure reporting** - When no fallback is available, the error is now properly reported to the user
+
+### Added
+
+- `isHostRetryableError()` helper for detecting connection errors that should trigger fallback
+- `switchToFallbackProvider()` method in Assistant for seamless host switching
+- `SetHostPool()` method to wire HostPool to Assistant
+
 ## [2.0.0] - 2026-02-04
 
 ### Added
@@ -26,8 +44,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
-- Version bump from 1.x to 2.x due to configuration schema changes
-- Agent registry now supports initialization from host pool
+- Version bumps from 1.x to 2.x due to configuration schema changes
+- Agent registry now supports initialization from the host pool
 - TaskBridge supports both single-host and multi-host operation modes
 
 ### Fixed
@@ -114,7 +132,9 @@ The project evolved through the following milestones before being open-sourced:
 - **v0.3.12** - File reference autocomplete, permissions system
 - **v0.3.8** - Native OpenAI function calling, security tools
 
-[Unreleased]: https://github.com/tara-vision/taracode/compare/v2.0.0...HEAD
+[Unreleased]: https://github.com/tara-vision/taracode/compare/v2.0.1...HEAD
+
+[2.0.1]: https://github.com/tara-vision/taracode/compare/v2.0.0...v2.0.1
 
 [2.0.0]: https://github.com/tara-vision/taracode/compare/v1.0.3...v2.0.0
 

cmd/repl.go

@@ -238,6 +238,9 @@ func startREPL() {
 			ui.SuccessStyle.Render(ui.IconSuccess),
 			hostPool.HealthyCount(),
 			hostPool.HostCount())
+
+		// Wire HostPool to Assistant for automatic fallback (v2.0)
+		asst.SetHostPool(hostPool)
 	}
 
 	// Initialize TaskBridge for multi-agent orchestration

internal/assistant/assistant.go

@@ -81,6 +81,22 @@ func isRetryable(err error) bool {
 		strings.Contains(errMsg, "temporary failure")
 }
 
+// isHostRetryableError checks if an error indicates a host connection failure
+// that should trigger a fallback to another host (v2.0 multi-host support)
+func isHostRetryableError(err error) bool {
+	if err == nil {
+		return false
+	}
+	errMsg := strings.ToLower(err.Error())
+	return strings.Contains(errMsg, "connection refused") ||
+		strings.Contains(errMsg, "host is down") ||
+		strings.Contains(errMsg, "no such host") ||
+		strings.Contains(errMsg, "i/o timeout") ||
+		strings.Contains(errMsg, "dial tcp") ||
+		strings.Contains(errMsg, "network is unreachable") ||
+		strings.Contains(errMsg, "connection reset")
+}
+
 // withRetry executes fn with exponential backoff retry for transient errors
 func withRetry[T any](ctx gocontext.Context, operation string, fn func() (T, error)) (T, error) {
 	var result T
@@ -142,6 +158,9 @@ type Assistant struct {
 
 	// Last AI response (for suggestion detection)
 	lastResponse string
+
+	// Multi-host fallback support (v2.0)
+	hostPool *provider.HostPool
 }
 
 // StreamFilter handles real-time filtering of think tags during streaming
@@ -923,6 +942,30 @@ func (a *Assistant) GetProvider() provider.Provider {
 	return a.provider
 }
 
+// SetHostPool sets the host pool for multi-host fallback support (v2.0)
+func (a *Assistant) SetHostPool(pool *provider.HostPool) {
+	a.hostPool = pool
+}
+
+// switchToFallbackProvider attempts to switch to a healthy fallback host (v2.0)
+// Returns the name of the host switched to, or an error if no fallback available
+func (a *Assistant) switchToFallbackProvider() (string, error) {
+	if a.hostPool == nil {
+		return "", fmt.Errorf("no host pool configured")
+	}
+
+	prov, hostName, err := a.hostPool.GetDefaultWithFallback()
+	if err != nil {
+		return "", err
+	}
+
+	// Update provider and client
+	a.provider = prov
+	a.client = prov.CreateClient()
+
+	return hostName, nil
+}
+
 // AddMCPToolDefinition adds an MCP tool definition for LLM function calling
 func (a *Assistant) AddMCPToolDefinition(tool openai.Tool) {
 	a.toolDefs = append(a.toolDefs, tool)
@@ -2220,6 +2263,17 @@ func (a *Assistant) processMessageStreamingWithImages(userMessage string, images
 			stream, err = a.client.CreateChatCompletionStream(ctx, req)
 		}
 
+		// If still failing with connection error, try host fallback (v2.0)
+		if err != nil && isHostRetryableError(err) && a.hostPool != nil {
+			if hostName, switchErr := a.switchToFallbackProvider(); switchErr == nil {
+				fmt.Printf("\n%s Primary host unavailable, switched to: %s\n", ui.IconWarning, hostName)
+				stream, err = a.client.CreateChatCompletionStream(ctx, req)
+			} else {
+				// Fallback also failed - log for visibility
+				fmt.Printf("\n%s Primary host unavailable, no fallback available: %v\n", ui.IconWarning, switchErr)
+			}
+		}
+
 		if err != nil {
 			if thinkingSpinner != nil {
 				thinkingSpinner.Stop()

internal/provider/hostpool.go

@@ -69,14 +69,17 @@ func (p *HostPool) Connect(ctx context.Context, name string) error {
 	}
 
 	conn.Provider = prov
-	conn.MarkHealthy(latency)
 
-	// Detect available models
+	// Detect available models to verify connectivity
 	models, err := prov.DetectModels(ctx)
-	if err == nil {
-		conn.Models = models
+	if err != nil {
+		conn.MarkUnavailable(err)
+		return fmt.Errorf("failed to verify host %q connectivity: %w", name, err)
 	}
 
+	conn.Models = models
+	conn.MarkHealthy(latency)
+
 	return nil
 }