diff --git a/cmd/olliesrv/internal/agent/cancellation_test.go b/cmd/olliesrv/internal/agent/cancellation_test.go
new file mode 100644
index 0000000..8c77b5f
--- /dev/null
+++ b/cmd/olliesrv/internal/agent/cancellation_test.go
@@ -0,0 +1,142 @@
+package agent
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"ollie/cmd/olliesrv/internal/backend"
+	olog "ollie/log"
+)
+
+// noOpLogger is a logger that discards all output.
+var noOpLogger = olog.NewWriter("test", olog.LevelError+1, nil, nil)
+
+// blockingBackend implements backend.Backend but blocks on ChatStream until ctx is cancelled.
+type blockingBackend struct {
+	chatStreamBlocked atomic.Bool
+	model             string
+	ctxLength         int
+}
+
+func (b *blockingBackend) ChatStream(ctx context.Context, messages []backend.Message, tools []backend.Tool, params backend.GenerationParams) (<-chan backend.StreamEvent, error) {
+	b.chatStreamBlocked.Store(true)
+	ch := make(chan backend.StreamEvent, 1)
+	go func() {
+		defer close(ch)
+		select {
+		case <-ctx.Done():
+			return
+		}
+	}()
+	return ch, nil
+}
+func (b *blockingBackend) Name() string                        { return "test" }
+func (b *blockingBackend) DefaultModel() string                { return "test-model" }
+func (b *blockingBackend) Model() string                       { return b.model }
+func (b *blockingBackend) SetModel(model string)               { b.model = model }
+func (b *blockingBackend) ContextLength(_ context.Context) int { return b.ctxLength }
+func (b *blockingBackend) Models(_ context.Context) []string   { return []string{"test-model"} }
+
+// TestStripColdCancellation verifies that /stop (Interrupt) cancels stripCold's
+// blocking LLM call promptly instead of waiting for it to complete.
+func TestStripColdCancellation(t *testing.T) {
+	b := &blockingBackend{model: "test", ctxLength: 128000}
+	h := newHistory("test input")
+
+	// Add enough messages so stripCold has work to do.
+	// stripCold processes messages before index len(messages)-hotTailSize (8).
+	// Need >8 cold-zone tool results (>200 chars each).
+	for i := 0; i < 10; i++ {
+		h.messages = append(h.messages, backend.Message{
+			Role:    "tool",
+			Content: fmt.Sprintf("tool-%d: %s", i, strings.Repeat("x", 300)),
+		})
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan struct{})
+
+	go func() {
+		h.stripCold(ctx, b)
+		close(done)
+	}()
+
+	// Wait for the backend to be blocked.
+	timeout := time.After(2 * time.Second)
+	for !b.chatStreamBlocked.Load() {
+		select {
+		case <-done:
+			t.Fatal("stripCold exited before backend was blocked")
+		case <-timeout:
+			t.Fatal("timed out waiting for backend to block")
+		case <-time.After(10 * time.Millisecond):
+		}
+	}
+
+	// Cancel via the same mechanism as Interrupt — this should unblock stripCold.
+	cancel()
+
+	select {
+	case <-done:
+		// Good — stripCold returned promptly after cancellation.
+	case <-time.After(1 * time.Second):
+		t.Fatal("stripCold did not return within 1s of context cancellation")
+	}
+}
+
+// TestPreRunCompactionCancellation verifies that /stop cancels pre-run auto-compaction
+// which uses runCompact → history.compact → ChatStream.
+func TestPreRunCompactionCancellation(t *testing.T) {
+	b := &blockingBackend{model: "test", ctxLength: 128000}
+
+	// Create an agent with enough history to trigger compaction (>hotTailSize+warmIndexSize=18).
+	ag := &Agent{
+		runtime: &Runtime{
+			Backend: b,
+		},
+		history: newHistory("test"),
+		log: noOpLogger,
+	}
+
+	// Add enough messages to exceed the compact threshold.
+	for i := 0; i < 30; i++ {
+		ag.history.appendUserMessage(strings.Repeat("x", 500))
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan struct{})
+
+	go func() {
+		_, err := ag.runCompact(ctx, "test")
+		if err != nil && !strings.Contains(err.Error(), "auto-compact") {
+			t.Logf("runCompact error (expected): %v", err)
+		}
+		close(done)
+	}()
+
+	// Wait for the backend to be blocked.
+	timeout := time.After(2 * time.Second)
+	for !b.chatStreamBlocked.Load() {
+		select {
+		case <-done:
+			t.Fatal("runCompact exited before backend was blocked")
+		case <-timeout:
+			t.Fatal("timed out waiting for backend to block")
+		case <-time.After(10 * time.Millisecond):
+		}
+	}
+
+	// Cancel — this should unblock the compaction.
+	cancel()
+
+	select {
+	case <-done:
+		// Good — runCompact returned promptly after cancellation.
+	case <-time.After(1 * time.Second):
+		t.Fatal("runCompact did not return within 1s of context cancellation")
+	}
+}
diff --git a/cmd/olliesrv/internal/agent/history.go b/cmd/olliesrv/internal/agent/history.go
index 63c2628..6f1653c 100644
--- a/cmd/olliesrv/internal/agent/history.go
+++ b/cmd/olliesrv/internal/agent/history.go
@@ -9,8 +9,6 @@ import (
 
 	"ollie/cmd/olliesrv/internal/backend"
 	"ollie/paths"
-
-	"gopkg.in/yaml.v3"
 )
 
 const (
@@ -391,9 +389,9 @@ func (s *History) stripCold(ctx context.Context, b backend.Backend) {
 
 	// Collect cold-zone tool results exceeding 200 chars.
 	type item struct {
-		id     int
-		idx    int // index into s.messages
-		orig   string
+		id   int
+		idx  int // index into s.messages
+		orig string
 	}
 	var items []item
 	for i := range s.messages {
@@ -442,8 +440,7 @@ func (s *History) stripCold(ctx context.Context, b backend.Backend) {
 }
 
 // resolveCompactionModel returns the model to use for compaction.
-// Priority: agent config > OLLIE_COMPACTION_MODEL env > per-backend default from models.yaml > session's current model.
-// Returns "" if no override is configured (use the session's current model).
+// Priority: agent config > OLLIE_COMPACTION_MODEL > backends.conf compactionModel > backends.conf model > current model.
 func resolveCompactionModel(cfgModel string, b backend.Backend) string {
 	if cfgModel != "" {
 		return cfgModel
@@ -451,34 +448,11 @@ func resolveCompactionModel(cfgModel string, b backend.Backend) string {
 	if env := os.Getenv("OLLIE_COMPACTION_MODEL"); env != "" {
 		return env
 	}
-	cfg := loadModelsConfig()
 	if b != nil {
-		if m, ok := cfg.Compaction[b.Name()]; ok {
-			return m
+		if model := backend.CompactionModel(b.Name()); model != "" {
+			return model
 		}
+		return b.Model()
 	}
 	return ""
 }
-
-// modelsConfig holds the parsed contents of ~/.config/ollie/models.yaml.
-type modelsConfig struct {
-	Compaction map[string]string `yaml:"compaction"`
-	Completion struct {
-		Model   string `yaml:"model"`
-		Backend string `yaml:"backend"`
-	} `yaml:"completion"`
-}
-
-// loadModelsConfig reads and parses the models config file.
-// Returns zero value if the file doesn't exist or is invalid.
-func loadModelsConfig() modelsConfig {
-	data, err := os.ReadFile(paths.CfgDir() + "/models.yaml")
-	if err != nil {
-		return modelsConfig{}
-	}
-	var cfg modelsConfig
-	if err := yaml.Unmarshal(data, &cfg); err != nil {
-		return modelsConfig{}
-	}
-	return cfg
-}
diff --git a/cmd/olliesrv/internal/agent/loop.go b/cmd/olliesrv/internal/agent/loop.go
index 4a35392..3a09e14 100644
--- a/cmd/olliesrv/internal/agent/loop.go
+++ b/cmd/olliesrv/internal/agent/loop.go
@@ -201,7 +201,11 @@ func (ag *Agent) run(ctx context.Context) error {
 			}
 		}
 
-		ag.autoCompact(ctx)
+		_, compactErr := ag.autoCompact(ctx)
+		if compactErr != nil {
+			ag.emit(Event{Role: "error", Content: compactErr.Error()})
+			break
+		}
 
 		step++
 	}
@@ -210,20 +214,23 @@ func (ag *Agent) run(ctx context.Context) error {
 }
 
 // autoCompact triggers context compaction if above threshold.
-func (ag *Agent) autoCompact(ctx context.Context) {
+// Returns the number of messages compacted, or an error if compaction failed.
+func (ag *Agent) autoCompact(ctx context.Context) (int, error) {
 	if ctx.Err() != nil || ag.history == nil {
-		return
+		return 0, nil
 	}
 	limit := ag.autoCompactLimit(ctx)
 	if limit <= 0 || ag.history.estimateTokens() < limit {
-		return
+		return 0, nil
 	}
 	ag.emit(Event{Role: "info", Content: "auto-compacting context...\n"})
 	ag.SetState("compacting")
-	if _, err := ag.runCompact(ctx, "auto"); err != nil {
-		panic(fmt.Sprintf("mid-turn auto-compact: %v", err))
+	n, err := ag.runCompact(ctx, "auto")
+	if err != nil {
+		return n, fmt.Errorf("auto-compact: %w", err)
 	}
 	ag.SetState("thinking")
+	return n, nil
 }
 
 // ── streamResponse ──────────────────────────────────────────────────────────
diff --git a/cmd/olliesrv/internal/agent/turn.go b/cmd/olliesrv/internal/agent/turn.go
index 68ff578..d2097c0 100644
--- a/cmd/olliesrv/internal/agent/turn.go
+++ b/cmd/olliesrv/internal/agent/turn.go
@@ -180,14 +180,14 @@ func (ag *Agent) executeTurn(ctx context.Context, input string) string {
 	// Warn once when context usage crosses 60%; compact at 75%.
 	if ag.history != nil {
 		tokens := ag.history.estimateTokens()
-		if compactLimit := ag.autoCompactLimit(ctx); compactLimit > 0 && tokens >= compactLimit {
+		if compactLimit := ag.autoCompactLimit(actCtx); compactLimit > 0 && tokens >= compactLimit {
 			ag.emit(Event{Role: "info", Content: "auto-compacting context...\n"})
 			ag.SetState("compacting")
-			if _, err := ag.runCompact(ctx, "auto"); err != nil {
-				panic(fmt.Sprintf("auto-compact: %v", err))
+			if _, err := ag.runCompact(actCtx, "auto"); err != nil {
+				ag.log.Error("auto-compact failed: %v", err)
 			}
 			ag.SetState("thinking")
-		} else if warnLimit := ag.autoWarnLimit(ctx); warnLimit > 0 && tokens >= warnLimit && !ag.warnedContext {
+		} else if warnLimit := ag.autoWarnLimit(actCtx); warnLimit > 0 && tokens >= warnLimit && !ag.warnedContext {
 			ctxLen := ag.runtime.Backend.ContextLength(ctx)
 			if ctxLen <= 0 {
 				ctxLen = defaultContextLength
@@ -204,7 +204,7 @@ func (ag *Agent) executeTurn(ctx context.Context, input string) string {
 
 	// Strip cold-zone tool results (one batched LLM call per turn).
 	if ag.history != nil {
-		ag.history.stripCold(ctx, ag.runtime.Backend)
+		ag.history.stripCold(actCtx, ag.runtime.Backend)
 	}
 
 	// Inject the preamble system message into history before running.
@@ -234,7 +234,7 @@ func (ag *Agent) executeTurn(ctx context.Context, input string) string {
 			ag.history.messages = ag.history.messages[:preTurnLen]
 			ag.emit(Event{Role: "info", Content: "context overflow — compacting and retrying...\n"})
 			ag.SetState("compacting")
-			if _, cerr := ag.runCompact(ctx, "overflow"); cerr != nil {
+			if _, cerr := ag.runCompact(actCtx, "overflow"); cerr != nil {
 				break
 			}
 			ag.history.appendUserMessage(input)
diff --git a/cmd/olliesrv/internal/backend/config.go b/cmd/olliesrv/internal/backend/config.go
index 5c29d09..f696363 100644
--- a/cmd/olliesrv/internal/backend/config.go
+++ b/cmd/olliesrv/internal/backend/config.go
@@ -10,10 +10,11 @@ import (
 
 // BackendConfig holds the per-backend configuration from backends.conf.
 type BackendConfig struct {
-	Key   string // API key / token
-	URL   string // base URL override (empty = use backend's default)
-	Model string // default model override
-	Host  string // for ollama: host address
+	Key             string // API key / token
+	URL             string // base URL override (empty = use backend's default)
+	Model           string // default model override
+	CompactionModel string // model used for history compaction (empty = use Model)
+	Host            string // for ollama: host address
 }
 
 // configFile is the parsed contents of backends.conf: section name → key/value pairs.
@@ -28,6 +29,18 @@ func loadConfig() configFile {
 	return parseConfigFile(paths.CfgDir() + "/backends.conf")
 }
 
+// CompactionModel returns the configured compaction model for a backend.
+// It falls back to that backend's configured default model when no dedicated
+// compaction model is set.
+func CompactionModel(name string) string {
+	cfg := loadConfig()
+	bc := cfg.Backends[name]
+	if bc.CompactionModel != "" {
+		return bc.CompactionModel
+	}
+	return bc.Model
+}
+
 // parseConfigFile parses an INI-style config file.
 // Format:
 //
@@ -89,6 +102,8 @@ func parseConfigFile(path string) configFile {
 			bc.URL = v
 		case "model":
 			bc.Model = v
+		case "compactionModel", "compaction_model", "compaction-model":
+			bc.CompactionModel = v
 		case "host":
 			bc.Host = v
 		}
diff --git a/cmd/olliesrv/internal/backend/config_test.go b/cmd/olliesrv/internal/backend/config_test.go
new file mode 100644
index 0000000..a3daa8d
--- /dev/null
+++ b/cmd/olliesrv/internal/backend/config_test.go
@@ -0,0 +1,26 @@
+package backend
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func TestParseConfigFileCompactionModel(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "backends.conf")
+	content := "backend = openrouter\n\n[openrouter]\nmodel = primary\ncompactionModel = compact\n\n[ollama]\nmodel = local\n"
+	if err := os.WriteFile(path, []byte(content), 0600); err != nil {
+		t.Fatal(err)
+	}
+
+	cfg := parseConfigFile(path)
+	if got := cfg.Backends["openrouter"].CompactionModel; got != "compact" {
+		t.Fatalf("openrouter compactionModel = %q, want %q", got, "compact")
+	}
+	if got := cfg.Backends["ollama"].CompactionModel; got != "" {
+		t.Fatalf("ollama compactionModel = %q, want empty", got)
+	}
+	if got := cfg.Backends["ollama"].Model; got != "local" {
+		t.Fatalf("ollama model = %q, want %q", got, "local")
+	}
+}
diff --git a/data/backends.conf b/data/backends.conf
index 35a9be9..214b0be 100644
--- a/data/backends.conf
+++ b/data/backends.conf
@@ -9,6 +9,7 @@
 #   key   = API key or token
 #   url   = base URL override (optional; each backend has a sensible default)
 #   model = default model (optional; overridden by session model= parameter)
+#   compactionModel = model used for history compaction (optional; defaults to model)
 #   host  = host address (ollama only)
 
 backend = openrouter