incrementalize cold summaries

cmd/olliesrv/internal/agent/history.go

@@ -413,6 +413,8 @@ func (s *History) estimateTokens() int {
 	return chars / 4
 }
 
+const maxColdSummariesPerCall = 8
+
 // stripCold summarizes large tool-result messages outside the hot tail
 // using a single batched LLM call. Messages in the last hotTailSize slots
 // are left verbatim.
@@ -455,6 +457,9 @@ func (s *History) stripCold(ctx context.Context, b backend.Backend) (backend.Usa
 					continue
 				}
 				items = append(items, item{idx: i, orig: m.Content, messageID: id, hash: toolSummaryHash(*m)})
+				if len(items) >= maxColdSummariesPerCall {
+					break
+				}
 			}
 		}
 	}

cmd/olliesrv/internal/agent/history_incremental_test.go

@@ -0,0 +1,48 @@
+package agent
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	"ollie/cmd/olliesrv/internal/backend"
+)
+
+func TestStripColdLimitsEachBatch(t *testing.T) {
+	b := &batchSummaryBackend{}
+	h := newHistory("test")
+	for i := 0; i < hotTailSize+maxColdSummariesPerCall+1; i++ {
+		m := backend.Message{Role: "user", Content: "hot"}
+		if i < maxColdSummariesPerCall+1 {
+			m = backend.Message{Role: "tool", ToolCallID: "tc-" + string(rune('0'+i)), Content: strings.Repeat("result ", 40)}
+		}
+		h.messages = append(h.messages, m)
+	}
+
+	if _, _ = h.stripCold(context.Background(), b); b.calls != 1 {
+		t.Fatalf("first stripCold calls = %d, want 1", b.calls)
+	}
+	if b.items != maxColdSummariesPerCall {
+		t.Fatalf("first batch items = %d, want %d", b.items, maxColdSummariesPerCall)
+	}
+}
+
+type batchSummaryBackend struct {
+	calls int
+	items int
+}
+
+func (b *batchSummaryBackend) ChatStream(_ context.Context, messages []backend.Message, _ []backend.Tool, _ backend.GenerationParams) (<-chan backend.StreamEvent, error) {
+	b.calls++
+	b.items = strings.Count(messages[0].Content, "id=tool:")
+	out := make(chan backend.StreamEvent, 1)
+	out <- backend.StreamEvent{Content: `{}`, Done: true}
+	close(out)
+	return out, nil
+}
+func (b *batchSummaryBackend) Name() string                      { return "test" }
+func (b *batchSummaryBackend) DefaultModel() string              { return "test" }
+func (b *batchSummaryBackend) Model() string                     { return "test" }
+func (b *batchSummaryBackend) SetModel(string)                   {}
+func (b *batchSummaryBackend) ContextLength(context.Context) int { return 128000 }
+func (b *batchSummaryBackend) Models(context.Context) []string   { return []string{"test"} }