feat(chat): stop active llm responses

backend/internal/api/generation.go

@@ -1,6 +1,7 @@
 package api
 
 import (
+	"context"
 	"errors"
 	"slices"
 	"sync"
@@ -19,6 +20,8 @@ type generationManager struct {
 
 type generation struct {
 	mu          sync.RWMutex
+	ctx         context.Context
+	cancel      context.CancelFunc
 	subscribers map[chan *MessageChunk]struct{}
 	done        chan struct{}
 	closed      bool
@@ -36,7 +39,10 @@ func (m *generationManager) start(chatID uuid.UUID, prepare func(*generation) er
 	}
 
 	// Reserve Generation
+	ctx, cancel := context.WithCancel(context.Background())
 	gen := &generation{
+		ctx:         ctx,
+		cancel:      cancel,
 		subscribers: make(map[chan *MessageChunk]struct{}),
 		done:        make(chan struct{}),
 	}
@@ -79,6 +85,19 @@ func (m *generationManager) subscribe(chatID uuid.UUID) (<-chan *MessageChunk, f
 	return ch, func() { gen.unsubscribe(ch) }, true
 }
 
+func (m *generationManager) stop(chatID uuid.UUID) bool {
+	m.mu.RLock()
+	gen, found := m.generations[chatID]
+	m.mu.RUnlock()
+	if !found {
+		return false
+	}
+
+	// Cancel Generation
+	gen.cancel()
+	return true
+}
+
 func (g *generation) subscribe() chan *MessageChunk {
 	ch := make(chan *MessageChunk, 64)
 
@@ -126,6 +145,7 @@ func (g *generation) close() {
 		return
 	}
 	g.closed = true
+	g.cancel()
 	close(g.done)
 	for subscriber := range g.subscribers {
 		close(subscriber)