fix: compact cleanup, ingestBatch parallelism, trimToBudget early exit (#5)

Co-authored-by: Jason Woltje <jason@diversecanvas.com>
Co-committed-by: Jason Woltje <jason@diversecanvas.com>

tests/engine.test.ts

@@ -131,6 +131,54 @@ describe("OpenBrainContextEngine", () => {
     expect(mockClient.createThought).toHaveBeenCalledTimes(2);
   });
 
+  it("ingests batches in parallel chunks of five", async () => {
+    const mockClient = makeMockClient();
+    let inFlight = 0;
+    let maxInFlight = 0;
+    let createdCount = 0;
+
+    vi.mocked(mockClient.createThought).mockImplementation(async (input: OpenBrainThoughtInput) => {
+      inFlight += 1;
+      maxInFlight = Math.max(maxInFlight, inFlight);
+      await new Promise((resolve) => {
+        setTimeout(resolve, 20);
+      });
+      inFlight -= 1;
+      createdCount += 1;
+      return {
+        id: `thought-${createdCount}`,
+        content: input.content,
+        source: input.source,
+        metadata: input.metadata,
+        createdAt: new Date().toISOString(),
+        updatedAt: undefined,
+        score: undefined,
+      };
+    });
+
+    const engine = new OpenBrainContextEngine(
+      {
+        baseUrl: "https://brain.example.com",
+        apiKey: "secret",
+      },
+      { createClient: () => mockClient },
+    );
+
+    await engine.bootstrap({ sessionId, sessionFile: "/tmp/session.json" });
+    const result = await engine.ingestBatch({
+      sessionId,
+      messages: Array.from({ length: 10 }, (_, index) => ({
+        role: index % 2 === 0 ? "user" : "assistant",
+        content: `message-${index + 1}`,
+        timestamp: index + 1,
+      })),
+    });
+
+    expect(result.ingestedCount).toBe(10);
+    expect(maxInFlight).toBe(5);
+    expect(mockClient.createThought).toHaveBeenCalledTimes(10);
+  });
+
   it("assembles context from recent + semantic search, deduped and budget-aware", async () => {
     const mockClient = makeMockClient();
     vi.mocked(mockClient.listRecent).mockResolvedValue([
@@ -192,12 +240,19 @@ describe("OpenBrainContextEngine", () => {
     ]);
   });
 
-  it("compact archives a summary thought", async () => {
+  it("compact archives a summary thought and deletes summarized inputs", async () => {
     const mockClient = makeMockClient();
-    vi.mocked(mockClient.listRecent).mockResolvedValue([
-      makeThought("t1", "first message", sessionId, "user", "2026-03-06T12:00:00.000Z"),
-      makeThought("t2", "second message", sessionId, "assistant", "2026-03-06T12:01:00.000Z"),
-    ]);
+    vi.mocked(mockClient.listRecent).mockResolvedValue(
+      Array.from({ length: 12 }, (_, index) => {
+        return makeThought(
+          `t${index + 1}`,
+          `message ${index + 1}`,
+          sessionId,
+          index % 2 === 0 ? "user" : "assistant",
+          `2026-03-06T12:${String(index).padStart(2, "0")}:00.000Z`,
+        );
+      }),
+    );
 
     const engine = new OpenBrainContextEngine(
       {
@@ -226,6 +281,54 @@ describe("OpenBrainContextEngine", () => {
         }),
       }),
     );
+    const deletedIds = vi
+      .mocked(mockClient.deleteThought)
+      .mock.calls.map(([id]) => id)
+      .sort((left, right) => left.localeCompare(right));
+    expect(deletedIds).toEqual([
+      "t10",
+      "t11",
+      "t12",
+      "t3",
+      "t4",
+      "t5",
+      "t6",
+      "t7",
+      "t8",
+      "t9",
+    ]);
+  });
+
+  it("stops trimming once the newest message exceeds budget", async () => {
+    const mockClient = makeMockClient();
+    const oversizedNewest = "z".repeat(400);
+    vi.mocked(mockClient.listRecent).mockResolvedValue([
+      makeThought("t1", "small older message", sessionId, "assistant", "2026-03-06T12:00:00.000Z"),
+      makeThought("t2", oversizedNewest, sessionId, "assistant", "2026-03-06T12:01:00.000Z"),
+    ]);
+
+    const engine = new OpenBrainContextEngine(
+      {
+        baseUrl: "https://brain.example.com",
+        apiKey: "secret",
+      },
+      { createClient: () => mockClient },
+    );
+
+    await engine.bootstrap({ sessionId, sessionFile: "/tmp/session.json" });
+    const result = await engine.assemble({
+      sessionId,
+      messages: [
+        {
+          role: "user",
+          content: "query",
+          timestamp: Date.now(),
+        },
+      ],
+      tokenBudget: 12,
+    });
+
+    expect(result.messages.map((message) => String(message.content))).toEqual([oversizedNewest]);
   });
 
   it("prepares subagent spawn and rollback deletes seeded context", async () => {