AI analytics #1339
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
nams1570 marked this conversation as resolved.
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,21 +1,44 @@ | ||
| import { | ||
| assertProjectAccess, | ||
| handleGenerateMode, | ||
| handleStreamMode, | ||
| } from "@/lib/ai/ai-query-handlers"; | ||
| import type { CommonLogFields, ModeContext } from "@/lib/ai/types"; | ||
| import { selectModel } from "@/lib/ai/models"; | ||
| import { getFullSystemPrompt, type SystemPromptId } from "@/lib/ai/prompts"; | ||
| import { requestBodySchema } from "@/lib/ai/schema"; | ||
| import { getTools } from "@/lib/ai/tools"; | ||
| import { getVerifiedQaContext } from "@/lib/ai/qa/verified-qa"; | ||
| import { SmartResponse } from "@/route-handlers/smart-response"; | ||
| import { createSmartRouteHandler } from "@/route-handlers/smart-route-handler"; | ||
| import { validateImageAttachments } from "@stackframe/stack-shared/dist/ai/image-limits"; | ||
| import { KnownErrors } from "@stackframe/stack-shared/dist/known-errors"; | ||
| import { yupMixed, yupObject, yupString } from "@stackframe/stack-shared/dist/schema-fields"; | ||
| import { getEnvVariable } from "@stackframe/stack-shared/dist/utils/env"; | ||
| import type { ModelMessage } from "ai"; | ||
|
|
||
| function getStepLimit(systemPromptId: SystemPromptId, hasTools: boolean): number { | ||
| if (!hasTools) return 1; | ||
| if (systemPromptId === "docs-ask-ai" || systemPromptId === "command-center-ask-ai") return 50; | ||
| if (systemPromptId === "create-dashboard") return 12; | ||
| return 5; | ||
| } | ||
|
|
||
| async function buildSystemPrompt(systemPromptId: SystemPromptId): Promise<string> { | ||
| let systemPrompt = getFullSystemPrompt(systemPromptId); | ||
| const isDocsOrSearch = systemPromptId === "docs-ask-ai" || systemPromptId === "command-center-ask-ai"; | ||
| if (isDocsOrSearch) { | ||
| // Stuffing the entire verified QA corpus into the system prompt on every | ||
| // request is intentionally naive — it grows monotonically with each new | ||
| // QA pair and re-fetches/re-sends content that's unchanged across | ||
| // requests. Once the corpus is large enough to matter we should swap to | ||
| // a retriever based system (maybe something like an embedding-based retriever | ||
| // (top-k by query similarity)) and/or cache the assembled context, | ||
| // but for the current corpus size this is fine and lets the model see everything | ||
| systemPrompt += await getVerifiedQaContext(); | ||
| } | ||
| return systemPrompt; | ||
| } | ||
|
|
||
| export const POST = createSmartRouteHandler({ | ||
| metadata: { | ||
|
|
@@ -34,144 +57,81 @@ export const POST = createSmartRouteHandler({ | |
| const { quality, speed, systemPrompt: systemPromptId, tools: toolNames, messages, projectId } = body; | ||
|
|
||
| if (projectId != null) { | ||
| await assertProjectAccess(projectId, fullReq.auth); | ||
| } | ||
| const imageValidationResult = validateImageAttachments(messages); | ||
| if (!imageValidationResult.ok) { | ||
| const { failure } = imageValidationResult; | ||
| switch (failure.code) { | ||
| case "too_many": { | ||
| throw new KnownErrors.TooManyImageAttachments(failure.maxImages); | ||
| } | ||
| case "too_large": { | ||
| throw new KnownErrors.ImageAttachmentTooLarge(failure.maxBytes, failure.actualBytes); | ||
| } | ||
| } | ||
| } | ||
|
|
||
| const authenticatedApiKey = isAuthenticated | ||
| ? getEnvVariable("STACK_OPENROUTER_AUTHENTICATED_API_KEY", "") | ||
| : ""; | ||
| const model = selectModel(quality, speed, isAuthenticated, authenticatedApiKey || undefined); | ||
| const systemPrompt = await buildSystemPrompt(systemPromptId); | ||
| const tools = await getTools(toolNames, { auth: fullReq.auth, targetProjectId: projectId }); | ||
| const toolsArg = Object.keys(tools).length > 0 ? tools : undefined; | ||
| const stepLimit = getStepLimit(systemPromptId, toolsArg != null); | ||
|
|
||
| const correlationId = crypto.randomUUID(); | ||
| const conversationIdForLog = body.mcpCallMetadata | ||
| ? body.mcpCallMetadata.conversationId ?? crypto.randomUUID() | ||
| : undefined; | ||
| const common: CommonLogFields = { | ||
| correlationId, | ||
| mode, | ||
|
Comment on lines
+89
to
+90
Contributor
There was a problem hiding this comment.
Prompt To Fix With AIThis is a comment left during a code review.
Path: apps/backend/src/app/api/latest/ai/query/[mode]/route.ts
Line: 89-90
Comment:
**`mcpCorrelationId` is always equal to `correlationId`**
`correlationId` is the UUID freshly minted for this AI query, and `mcpCorrelationId` is set to that same value when `mcpCallMetadata` is present. In `logIfMcpToolCall`, the MCP call log row is also written with the same `correlationId`. So `ai_query_log.mcpCorrelationId` and `ai_query_log.correlationId` carry identical values, making the field redundant. The field name implies a _different_ ID (the MCP log's own correlation key), which will confuse future maintainers. Either give the MCP call log its own distinct UUID and store that here, or remove the `mcpCorrelationId` column and rely on the shared `correlationId`.
How can I resolve this? If you propose a fix, please make it concise. |
||
| systemPromptId, | ||
| quality, | ||
| speed, | ||
| modelId: String(model.modelId), | ||
| isAuthenticated, | ||
| projectId: projectId ?? undefined, | ||
| userId: fullReq.auth?.user?.id, | ||
| requestedToolsJson: JSON.stringify(toolNames), | ||
| messagesJson: JSON.stringify(messages), | ||
| mcpCorrelationId: body.mcpCallMetadata ? correlationId : undefined, | ||
| conversationId: conversationIdForLog, | ||
|
aadesh18 marked this conversation as resolved.
|
||
| }; | ||
| const startedAt = performance.now(); | ||
|
|
||
| const isAnthropic = model.modelId.startsWith("anthropic/"); | ||
| // Can be optimized: only opt into prompt caching for routes that are hit | ||
| // frequently enough to amortize the write. | ||
| const systemMessage: ModelMessage = { | ||
| role: "system", | ||
| content: systemPrompt, | ||
| ...(isAnthropic && { | ||
| providerOptions: { | ||
| openrouter: { cacheControl: { type: "ephemeral" } }, | ||
| }, | ||
| }), | ||
| }; | ||
|
Comment on lines
+110
to
+116
Collaborator
There was a problem hiding this comment. I want to point out a few things. Here, we're caching for only 5 min (https://openrouter.ai/docs/guides/best-practices/prompt-caching#cache-ttl-options)
Collaborator
Author
There was a problem hiding this comment. As we discussed, most of the routes are being hit at least once in 5 mins. however in future we should think about not having the cache for the routes that are not being hit too often. added a comment explain the same |
||
| // Cast: the schema narrows role and leaves content as unknown, but the | ||
| // AI SDK accepts a superset (role: "system" etc.). We've intentionally | ||
| // excluded `system` at the schema layer to prevent prompt-injection via | ||
| // client-supplied system messages — see schema.ts. | ||
| const modelMessages = messages as unknown as ModelMessage[]; | ||
| const cachedMessages: ModelMessage[] = [systemMessage, ...modelMessages]; | ||
|
|
||
| const ctx: ModeContext = { model, cachedMessages, toolsArg, stepLimit, common, startedAt }; | ||
| const extras = { | ||
| messages, | ||
| mcpCallMetadata: body.mcpCallMetadata ?? undefined, | ||
| correlationId, | ||
| conversationIdForLog, | ||
| }; | ||
|
|
||
| if (mode === "stream") { | ||
| return handleStreamMode({ ...ctx, ...extras }); | ||
| } | ||
| return await handleGenerateMode({ ...ctx, ...extras }); | ||
| }, | ||
| }); | ||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Uh oh!
There was an error while loading. Please reload this page.