@vellumai/assistant 0.6.6 → 0.7.0

package/src/__tests__/list-messages-tool-merge.test.ts

@@ -26,7 +26,8 @@ mock.module("../config/loader.js", () => ({
 }));
 
 import { addMessage, createConversation } from "../memory/conversation-crud.js";
-import { getDb, initializeDb } from "../memory/db.js";
+import { getDb } from "../memory/db-connection.js";
+import { initializeDb } from "../memory/db-init.js";
 import { handleListMessages } from "../runtime/routes/conversation-routes.js";
 
 initializeDb();
@@ -39,10 +40,8 @@ function resetTables() {
   db.run("DELETE FROM conversations");
 }
 
-function createTestUrl(conversationId: string): URL {
-  return new URL(
-    `http://localhost/v1/messages?conversationId=${conversationId}`,
-  );
+function createTestArgs(conversationId: string) {
+  return { queryParams: { conversationId } };
 }
 
 interface ToolCallPayload {
@@ -92,8 +91,8 @@ describe("handleListMessages tool_result merging", () => {
       ]),
     );
 
-    const response = handleListMessages(createTestUrl(conv.id), null);
-    const body = (await response.json()) as { messages: MessagePayload[] };
+    const response = handleListMessages(createTestArgs(conv.id), null);
+    const body = response as { messages: MessagePayload[] };
 
     // Should be 2 messages: user prompt + assistant (tool_result user msg suppressed)
     expect(body.messages).toHaveLength(2);
@@ -138,8 +137,8 @@ describe("handleListMessages tool_result merging", () => {
       ]),
     );
 
-    const response = handleListMessages(createTestUrl(conv.id), null);
-    const body = (await response.json()) as { messages: MessagePayload[] };
+    const response = handleListMessages(createTestArgs(conv.id), null);
+    const body = response as { messages: MessagePayload[] };
 
     expect(body.messages).toHaveLength(2);
     const toolCalls = body.messages[1].toolCalls!;
@@ -168,8 +167,8 @@ describe("handleListMessages tool_result merging", () => {
       JSON.stringify([{ type: "text", text: "how are you?" }]),
     );
 
-    const response = handleListMessages(createTestUrl(conv.id), null);
-    const body = (await response.json()) as { messages: MessagePayload[] };
+    const response = handleListMessages(createTestArgs(conv.id), null);
+    const body = response as { messages: MessagePayload[] };
 
     expect(body.messages).toHaveLength(3);
     expect(body.messages[2].role).toBe("user");
@@ -198,8 +197,8 @@ describe("handleListMessages tool_result merging", () => {
       JSON.stringify([{ type: "text", text: "response" }]),
     );
 
-    const response = handleListMessages(createTestUrl(conv.id), null);
-    const body = (await response.json()) as { messages: MessagePayload[] };
+    const response = handleListMessages(createTestArgs(conv.id), null);
+    const body = response as { messages: MessagePayload[] };
 
     // Orphan tool_result is preserved (not suppressed) to avoid data loss
     expect(body.messages).toHaveLength(2);
@@ -263,8 +262,8 @@ describe("handleListMessages tool_result merging", () => {
       JSON.stringify([{ type: "text", text: "thanks" }]),
     );
 
-    const response = handleListMessages(createTestUrl(conv.id), null);
-    const body = (await response.json()) as { messages: MessagePayload[] };
+    const response = handleListMessages(createTestArgs(conv.id), null);
+    const body = response as { messages: MessagePayload[] };
 
     // Consecutive assistant messages are merged at query time so the client
     // sees one grouped message (matching the streaming path behavior).
@@ -313,8 +312,8 @@ describe("handleListMessages tool_result merging", () => {
       ]),
     );
 
-    const response = handleListMessages(createTestUrl(conv.id), null);
-    const body = (await response.json()) as { messages: MessagePayload[] };
+    const response = handleListMessages(createTestArgs(conv.id), null);
+    const body = response as { messages: MessagePayload[] };
 
     expect(body.messages).toHaveLength(2);
     const tc = body.messages[1].toolCalls![0];

package/src/__tests__/llm-call-pipeline.test.ts

@@ -13,7 +13,7 @@
 
 import { afterAll, beforeEach, describe, expect, test } from "bun:test";
 
-import type { TrustContext } from "../daemon/conversation-runtime-assembly.js";
+import type { TrustContext } from "../daemon/trust-context.js";
 import { defaultLlmCallPlugin } from "../plugins/defaults/llm-call.js";
 import { DEFAULT_TIMEOUTS, runPipeline } from "../plugins/pipeline.js";
 import {
@@ -198,13 +198,12 @@ describe("llmCall pipeline", () => {
   });
 
   test("default registered first does not shadow later-registered user middleware", async () => {
-    // Regression for a shadowing bug where `defaultLlmCallPlugin` called
-    // `provider.sendMessage` directly instead of `next(args)`. Because the
-    // default registers at module load (before `bootstrapPlugins()` loads
-    // user plugins), it sat at the outermost layer in production — any
-    // user-registered `llmCall` middleware would have been silently skipped.
-    // This test locks in the fix by registering the default FIRST (matching
-    // production ordering) and asserting a user-registered spy still runs.
+    // The default plugin registers at module load (before `bootstrapPlugins()`
+    // loads user plugins), so it sits at the outermost layer in the onion.
+    // This test registers the default FIRST (matching production ordering)
+    // and asserts that a user-registered spy still runs — confirming that
+    // the outermost middleware forwards via `next(args)` rather than
+    // short-circuiting the chain.
     const observed: LLMCallArgs[] = [];
     const spyPlugin: Plugin = {
       manifest: {

package/src/__tests__/llm-context-normalization.test.ts

@@ -774,7 +774,7 @@ describe("normalizeLlmContextPayloads", () => {
     const normalized = normalizeLlmContextPayloads({
       createdAt: 1_742_400_000_002,
       requestPayload: {
-        model: "gemini-3-flash",
+        model: "gemini-3-flash-preview",
         contents: [
           {
             role: "user",
@@ -818,7 +818,7 @@ describe("normalizeLlmContextPayloads", () => {
         },
       },
       responsePayload: {
-        model: "gemini-3-flash-001",
+        model: "gemini-3-flash-preview",
         text: "Here is the summary.",
         functionCalls: [
           {
@@ -837,7 +837,7 @@ describe("normalizeLlmContextPayloads", () => {
 
     expect(normalized.summary).toEqual({
       provider: "gemini",
-      model: "gemini-3-flash-001",
+      model: "gemini-3-flash-preview",
       inputTokens: 200,
       outputTokens: 31,
       stopReason: "STOP",
@@ -902,7 +902,7 @@ describe("normalizeLlmContextPayloads", () => {
         kind: "settings",
         label: "Generation config",
         data: {
-          model: "gemini-3-flash",
+          model: "gemini-3-flash-preview",
           config: {
             temperature: 0.4,
             responseMimeType: "application/json",
@@ -1521,6 +1521,71 @@ describe("normalizeLlmContextPayloads", () => {
     expect(normalized.summary?.stopReason).toBe("incomplete");
   });
 
+  test("extracts cached input tokens from Responses API input_tokens_details", () => {
+    const normalized = normalizeLlmContextPayloads({
+      createdAt: 1_742_400_000_027,
+      requestPayload: {
+        model: "gpt-5.4",
+        instructions: "Be brief.",
+        input: [
+          {
+            role: "user",
+            content: [{ type: "input_text", text: "Hi" }],
+            type: "message",
+          },
+        ],
+      },
+      responsePayload: {
+        model: "gpt-5.4",
+        output: [
+          {
+            type: "message",
+            role: "assistant",
+            content: [{ type: "output_text", text: "Hey!" }],
+          },
+        ],
+        usage: {
+          input_tokens: 100,
+          output_tokens: 5,
+          input_tokens_details: { cached_tokens: 60 },
+        },
+        status: "completed",
+      },
+    });
+
+    expect(normalized.summary?.cacheReadInputTokens).toBe(60);
+    expect(normalized.summary?.cacheCreationInputTokens).toBeUndefined();
+    expect(normalized.summary?.inputTokens).toBe(100);
+  });
+
+  test("extracts cached input tokens from Chat Completions prompt_tokens_details", () => {
+    const normalized = normalizeLlmContextPayloads({
+      createdAt: 1_742_400_000_028,
+      requestPayload: {
+        model: "gpt-4.1",
+        messages: [{ role: "user", content: "Hi" }],
+      },
+      responsePayload: {
+        model: "gpt-4.1",
+        choices: [
+          {
+            finish_reason: "stop",
+            message: { role: "assistant", content: "Hey!" },
+          },
+        ],
+        usage: {
+          prompt_tokens: 80,
+          completion_tokens: 4,
+          prompt_tokens_details: { cached_tokens: 40 },
+        },
+      },
+    });
+
+    expect(normalized.summary?.cacheReadInputTokens).toBe(40);
+    expect(normalized.summary?.cacheCreationInputTokens).toBeUndefined();
+    expect(normalized.summary?.inputTokens).toBe(80);
+  });
+
   test("legacy chat-completions payloads are still normalized correctly alongside Responses tests", () => {
     // This test verifies backward compatibility: existing chat-completions
     // logs stored in the database must continue to normalize correctly even

package/src/__tests__/llm-context-route-provider.test.ts

@@ -7,48 +7,28 @@ mock.module("../util/logger.js", () => ({
     }),
 }));
 
-import { getDb, initializeDb } from "../memory/db.js";
+import { getDb } from "../memory/db-connection.js";
+import { initializeDb } from "../memory/db-init.js";
 import { llmRequestLogs } from "../memory/schema.js";
-import { conversationQueryRouteDefinitions } from "../runtime/routes/conversation-query-routes.js";
+import { ROUTES } from "../runtime/routes/conversation-query-routes.js";
+import { NotFoundError } from "../runtime/routes/errors.js";
 
 initializeDb();
 
-const routes = conversationQueryRouteDefinitions();
-
-function dispatchLlmContext(messageId: string): Promise<Response> | Response {
-  const url = new URL(`http://localhost/v1/messages/${messageId}/llm-context`);
-  const route = routes.find(
-    (r) => r.method === "GET" && r.endpoint === "messages/:id/llm-context",
-  );
-  if (!route) {
-    throw new Error("No llm-context route found");
-  }
-
-  return route.handler({
-    req: new Request(url.toString(), { method: "GET" }),
-    url,
-    server: null as never,
-    authContext: {} as never,
-    params: { id: messageId },
-  });
+const llmContextRoute = ROUTES.find(
+  (r) => r.method === "GET" && r.endpoint === "messages/:id/llm-context",
+)!;
+
+const logPayloadRoute = ROUTES.find(
+  (r) => r.method === "GET" && r.endpoint === "llm-request-logs/:id/payload",
+)!;
+
+function dispatchLlmContext(messageId: string) {
+  return llmContextRoute.handler({ pathParams: { id: messageId } });
 }
 
-function dispatchLogPayload(logId: string): Promise<Response> | Response {
-  const url = new URL(`http://localhost/v1/llm-request-logs/${logId}/payload`);
-  const route = routes.find(
-    (r) => r.method === "GET" && r.endpoint === "llm-request-logs/:id/payload",
-  );
-  if (!route) {
-    throw new Error("No llm-request-logs payload route found");
-  }
-
-  return route.handler({
-    req: new Request(url.toString(), { method: "GET" }),
-    url,
-    server: null as never,
-    authContext: {} as never,
-    params: { id: logId },
-  });
+function dispatchLogPayload(logId: string) {
+  return logPayloadRoute.handler({ pathParams: { id: logId } });
 }
 
 function clearRequestLogs(): void {
@@ -117,20 +97,13 @@ describe("GET /v1/messages/:id/llm-context provider preference", () => {
       responsePayload: openAiResponsePayload,
     });
 
-    const response = await dispatchLlmContext("msg-openrouter");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
-      logs: Array<{
-        summary?: { provider: string };
-      }>;
+    const body = (await dispatchLlmContext("msg-openrouter")) as {
+      logs: Array<{ summary?: { provider: string } }>;
     };
 
     expect(body.logs).toHaveLength(1);
     expect(body.logs[0]?.summary).toEqual(
-      expect.objectContaining({
-        provider: "openrouter",
-      }),
+      expect.objectContaining({ provider: "openrouter" }),
     );
   });
 
@@ -143,20 +116,13 @@ describe("GET /v1/messages/:id/llm-context provider preference", () => {
       responsePayload: openAiResponsePayload,
     });
 
-    const response = await dispatchLlmContext("msg-fireworks");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
-      logs: Array<{
-        summary?: { provider: string };
-      }>;
+    const body = (await dispatchLlmContext("msg-fireworks")) as {
+      logs: Array<{ summary?: { provider: string } }>;
     };
 
     expect(body.logs).toHaveLength(1);
     expect(body.logs[0]?.summary).toEqual(
-      expect.objectContaining({
-        provider: "fireworks",
-      }),
+      expect.objectContaining({ provider: "fireworks" }),
     );
   });
 
@@ -169,20 +135,13 @@ describe("GET /v1/messages/:id/llm-context provider preference", () => {
       responsePayload: openAiResponsePayload,
     });
 
-    const response = await dispatchLlmContext("msg-legacy");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
-      logs: Array<{
-        summary?: { provider: string };
-      }>;
+    const body = (await dispatchLlmContext("msg-legacy")) as {
+      logs: Array<{ summary?: { provider: string } }>;
     };
 
     expect(body.logs).toHaveLength(1);
     expect(body.logs[0]?.summary).toEqual(
-      expect.objectContaining({
-        provider: "openai",
-      }),
+      expect.objectContaining({ provider: "openai" }),
     );
   });
 
@@ -195,13 +154,8 @@ describe("GET /v1/messages/:id/llm-context provider preference", () => {
       responsePayload: "still-not-json",
     });
 
-    const response = await dispatchLlmContext("msg-raw-only");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
-      logs: Array<{
-        summary?: { provider: string };
-      }>;
+    const body = (await dispatchLlmContext("msg-raw-only")) as {
+      logs: Array<{ summary?: { provider: string } }>;
     };
 
     expect(body.logs).toHaveLength(1);
@@ -217,14 +171,8 @@ describe("GET /v1/messages/:id/llm-context provider preference", () => {
       responsePayload: openAiResponsePayload,
     });
 
-    const response = await dispatchLlmContext("msg-null-payload");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
-      logs: Array<{
-        requestPayload: unknown;
-        responsePayload: unknown;
-      }>;
+    const body = (await dispatchLlmContext("msg-null-payload")) as {
+      logs: Array<{ requestPayload: unknown; responsePayload: unknown }>;
     };
 
     expect(body.logs).toHaveLength(1);
@@ -272,21 +220,13 @@ describe("GET /v1/messages/:id/llm-context provider preference", () => {
       responsePayload: responsesApiResponsePayload,
     });
 
-    const response = await dispatchLlmContext("msg-responses-legacy");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
-      logs: Array<{
-        summary?: { provider: string; inputTokens?: number };
-      }>;
+    const body = (await dispatchLlmContext("msg-responses-legacy")) as {
+      logs: Array<{ summary?: { provider: string; inputTokens?: number } }>;
     };
 
     expect(body.logs).toHaveLength(1);
     expect(body.logs[0]?.summary).toEqual(
-      expect.objectContaining({
-        provider: "openai",
-        inputTokens: 11,
-      }),
+      expect.objectContaining({ provider: "openai", inputTokens: 11 }),
     );
   });
 
@@ -299,20 +239,13 @@ describe("GET /v1/messages/:id/llm-context provider preference", () => {
       responsePayload: responsesApiResponsePayload,
     });
 
-    const response = await dispatchLlmContext("msg-responses-openai");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
-      logs: Array<{
-        summary?: { provider: string };
-      }>;
+    const body = (await dispatchLlmContext("msg-responses-openai")) as {
+      logs: Array<{ summary?: { provider: string } }>;
     };
 
     expect(body.logs).toHaveLength(1);
     expect(body.logs[0]?.summary).toEqual(
-      expect.objectContaining({
-        provider: "openai",
-      }),
+      expect.objectContaining({ provider: "openai" }),
     );
   });
 });
@@ -331,10 +264,7 @@ describe("GET /v1/llm-request-logs/:id/payload", () => {
       responsePayload: resPayload,
     });
 
-    const response = await dispatchLogPayload("log-payload-ok");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
+    const body = (await dispatchLogPayload("log-payload-ok")) as {
       id: string;
       requestPayload: unknown;
       responsePayload: unknown;
@@ -345,9 +275,8 @@ describe("GET /v1/llm-request-logs/:id/payload", () => {
     expect(body.responsePayload).toEqual(JSON.parse(resPayload));
   });
 
-  test("returns 404 for a nonexistent log", async () => {
-    const response = await dispatchLogPayload("does-not-exist");
-    expect(response.status).toBe(404);
+  test("returns NotFoundError for a nonexistent log", () => {
+    expect(() => dispatchLogPayload("does-not-exist")).toThrow(NotFoundError);
   });
 
   test("falls back to string values for non-JSON payloads", async () => {
@@ -359,10 +288,7 @@ describe("GET /v1/llm-request-logs/:id/payload", () => {
       responsePayload: "raw-response-text",
     });
 
-    const response = await dispatchLogPayload("log-raw-strings");
-    expect(response.status).toBe(200);
-
-    const body = (await response.json()) as {
+    const body = (await dispatchLogPayload("log-raw-strings")) as {
       id: string;
       requestPayload: unknown;
       responsePayload: unknown;

package/src/__tests__/llm-request-log-turn-query.test.ts

@@ -25,7 +25,8 @@ import {
   createConversation,
   forkConversation,
 } from "../memory/conversation-crud.js";
-import { getDb, initializeDb } from "../memory/db.js";
+import { getDb } from "../memory/db-connection.js";
+import { initializeDb } from "../memory/db-init.js";
 import {
   backfillMessageIdOnLogs,
   getRequestLogsByMessageId,