move ai to v0 routes

.vscode/settings.json

@@ -3,6 +3,7 @@
   "cSpell.words": [
     "actix",
     "autoclean",
+    "autoincrement",
     "awscli",
     "awscliv",
     "ayush",
@@ -25,6 +26,7 @@
     "dpkg",
     "elif",
     "endregion",
+    "expressjwt",
     "finitize",
     "Fuzzysort",
     "GETCELL",

quadratic-api/src/routes/ai/anthropic.ts → quadratic-api/src/ai/handler/anthropic.ts

@@ -1,17 +1,21 @@
 import Anthropic from '@anthropic-ai/sdk';
 import type { Response } from 'express';
+import {
+  getAnthropicApiArgs,
+  parseAnthropicResponse,
+  parseAnthropicStream,
+} from 'quadratic-api/src/ai/helpers/anthropic.helper';
+import { ANTHROPIC_API_KEY } from 'quadratic-api/src/env-vars';
 import { getModelOptions } from 'quadratic-shared/ai/helpers/model.helper';
-import type { AIAutoCompleteRequestBody, AIMessagePrompt, AnthropicModel } from 'quadratic-shared/typesAndSchemasAI';
-import { ANTHROPIC_API_KEY } from '../../env-vars';
-import { getAnthropicApiArgs, parseAnthropicResponse, parseAnthropicStream } from './helpers/anthropic.helper';
+import type { AIMessagePrompt, AIRequestBody, AnthropicModel } from 'quadratic-shared/typesAndSchemasAI';
 
 const anthropic = new Anthropic({
   apiKey: ANTHROPIC_API_KEY,
 });
 
 export const handleAnthropicRequest = async (
   model: AnthropicModel,
-  args: Omit<AIAutoCompleteRequestBody, 'model'>,
+  args: Omit<AIRequestBody, 'model'>,
   response: Response
 ): Promise<AIMessagePrompt | undefined> => {
   const { system, messages, tools, tool_choice } = getAnthropicApiArgs(args);
@@ -34,7 +38,7 @@ export const handleAnthropicRequest = async (
       response.setHeader('Cache-Control', 'no-cache');
       response.setHeader('Connection', 'keep-alive');
 
-      const responseMessage = await parseAnthropicStream(chunks, response);
+      const responseMessage = await parseAnthropicStream(chunks, response, model);
       return responseMessage;
     } catch (error: any) {
       if (!response.headersSent) {
@@ -61,7 +65,7 @@ export const handleAnthropicRequest = async (
         tools,
         tool_choice,
       });
-      const responseMessage = parseAnthropicResponse(result, response);
+      const responseMessage = parseAnthropicResponse(result, response, model);
       return responseMessage;
     } catch (error: any) {
       if (error instanceof Anthropic.APIError) {

quadratic-api/src/routes/ai/bedrock.ts → quadratic-api/src/ai/handler/bedrock.ts

@@ -2,15 +2,19 @@ import { AnthropicBedrock } from '@anthropic-ai/bedrock-sdk';
 import Anthropic from '@anthropic-ai/sdk';
 import { BedrockRuntimeClient, ConverseCommand, ConverseStreamCommand } from '@aws-sdk/client-bedrock-runtime';
 import { type Response } from 'express';
-import { getModelOptions, isBedrockAnthropicModel } from 'quadratic-shared/ai/helpers/model.helper';
 import {
-  type AIAutoCompleteRequestBody,
-  type AIMessagePrompt,
-  type BedrockModel,
-} from 'quadratic-shared/typesAndSchemasAI';
-import { AWS_S3_ACCESS_KEY_ID, AWS_S3_REGION, AWS_S3_SECRET_ACCESS_KEY } from '../../env-vars';
-import { getAnthropicApiArgs, parseAnthropicResponse, parseAnthropicStream } from './helpers/anthropic.helper';
-import { getBedrockApiArgs, parseBedrockResponse, parseBedrockStream } from './helpers/bedrock.helper';
+  getAnthropicApiArgs,
+  parseAnthropicResponse,
+  parseAnthropicStream,
+} from 'quadratic-api/src/ai/helpers/anthropic.helper';
+import {
+  getBedrockApiArgs,
+  parseBedrockResponse,
+  parseBedrockStream,
+} from 'quadratic-api/src/ai/helpers/bedrock.helper';
+import { AWS_S3_ACCESS_KEY_ID, AWS_S3_REGION, AWS_S3_SECRET_ACCESS_KEY } from 'quadratic-api/src/env-vars';
+import { getModelOptions, isBedrockAnthropicModel } from 'quadratic-shared/ai/helpers/model.helper';
+import type { AIMessagePrompt, AIRequestBody, BedrockModel } from 'quadratic-shared/typesAndSchemasAI';
 
 // aws-sdk for bedrock, generic for all models
 const bedrock = new BedrockRuntimeClient({
@@ -27,7 +31,7 @@ const bedrock_anthropic = new AnthropicBedrock({
 
 export const handleBedrockRequest = async (
   model: BedrockModel,
-  args: Omit<AIAutoCompleteRequestBody, 'model'>,
+  args: Omit<AIRequestBody, 'model'>,
   response: Response
 ): Promise<AIMessagePrompt | undefined> => {
   const { stream, temperature, max_tokens } = getModelOptions(model, args);
@@ -51,7 +55,7 @@ export const handleBedrockRequest = async (
         response.setHeader('Cache-Control', 'no-cache');
         response.setHeader('Connection', 'keep-alive');
 
-        const responseMessage = await parseAnthropicStream(chunks, response);
+        const responseMessage = await parseAnthropicStream(chunks, response, model);
         return responseMessage;
       } catch (error: any) {
         if (!response.headersSent) {
@@ -77,7 +81,7 @@ export const handleBedrockRequest = async (
           tools,
           tool_choice,
         });
-        const responseMessage = parseAnthropicResponse(result, response);
+        const responseMessage = parseAnthropicResponse(result, response, model);
         return responseMessage;
       } catch (error: any) {
         if (error instanceof Anthropic.APIError) {
@@ -111,7 +115,7 @@ export const handleBedrockRequest = async (
         response.setHeader('Cache-Control', 'no-cache');
         response.setHeader('Connection', 'keep-alive');
 
-        const responseMessage = await parseBedrockStream(chunks, response);
+        const responseMessage = await parseBedrockStream(chunks, response, model);
         return responseMessage;
       } catch (error: any) {
         if (!response.headersSent) {
@@ -141,7 +145,7 @@ export const handleBedrockRequest = async (
         });
 
         const result = await bedrock.send(command);
-        const responseMessage = parseBedrockResponse(result.output, response);
+        const responseMessage = parseBedrockResponse(result.output, response, model);
         return responseMessage;
       } catch (error: any) {
         if (error.response) {

quadratic-api/src/routes/ai/openai.ts → quadratic-api/src/ai/handler/openai.ts

@@ -1,21 +1,17 @@
 import { type Response } from 'express';
 import OpenAI from 'openai';
+import { getOpenAIApiArgs, parseOpenAIResponse, parseOpenAIStream } from 'quadratic-api/src/ai/helpers/openai.helper';
+import { OPENAI_API_KEY } from 'quadratic-api/src/env-vars';
 import { getModelOptions } from 'quadratic-shared/ai/helpers/model.helper';
-import {
-  type AIAutoCompleteRequestBody,
-  type AIMessagePrompt,
-  type OpenAIModel,
-} from 'quadratic-shared/typesAndSchemasAI';
-import { OPENAI_API_KEY } from '../../env-vars';
-import { getOpenAIApiArgs, parseOpenAIResponse, parseOpenAIStream } from './helpers/openai.helper';
+import type { AIMessagePrompt, AIRequestBody, OpenAIModel } from 'quadratic-shared/typesAndSchemasAI';
 
 const openai = new OpenAI({
   apiKey: OPENAI_API_KEY || '',
 });
 
 export const handleOpenAIRequest = async (
   model: OpenAIModel,
-  args: Omit<AIAutoCompleteRequestBody, 'model'>,
+  args: Omit<AIRequestBody, 'model'>,
   response: Response
 ): Promise<AIMessagePrompt | undefined> => {
   const { messages, tools, tool_choice } = getOpenAIApiArgs(args);
@@ -36,7 +32,7 @@ export const handleOpenAIRequest = async (
       response.setHeader('Cache-Control', 'no-cache');
       response.setHeader('Connection', 'keep-alive');
 
-      const responseMessage = await parseOpenAIStream(completion, response);
+      const responseMessage = await parseOpenAIStream(completion, response, model);
       return responseMessage;
     } catch (error: any) {
       if (!response.headersSent) {
@@ -60,7 +56,7 @@ export const handleOpenAIRequest = async (
         tools,
         tool_choice,
       });
-      const responseMessage = parseOpenAIResponse(result, response);
+      const responseMessage = parseOpenAIResponse(result, response, model);
       return responseMessage;
     } catch (error: any) {
       if (error instanceof OpenAI.APIError) {

quadratic-api/src/routes/ai/helpers/anthropic.helper.ts → quadratic-api/src/ai/helpers/anthropic.helper.ts

@@ -5,17 +5,17 @@ import { getSystemPromptMessages } from 'quadratic-shared/ai/helpers/message.hel
 import type { AITool } from 'quadratic-shared/ai/specs/aiToolsSpec';
 import { aiToolsSpec } from 'quadratic-shared/ai/specs/aiToolsSpec';
 import type {
-  AIAutoCompleteRequestBody,
   AIMessagePrompt,
-  AnthropicAutoCompleteRequestBody,
+  AIRequestBody,
+  AnthropicModel,
   AnthropicPromptMessage,
+  AnthropicRequestBody,
   AnthropicTool,
   AnthropicToolChoice,
+  BedrockAnthropicModel,
 } from 'quadratic-shared/typesAndSchemasAI';
 
-export function getAnthropicApiArgs(
-  args: Omit<AIAutoCompleteRequestBody, 'model'>
-): Omit<AnthropicAutoCompleteRequestBody, 'model'> {
+export function getAnthropicApiArgs(args: Omit<AIRequestBody, 'model'>): Omit<AnthropicRequestBody, 'model'> {
   const { messages: chatMessages, useTools, toolName } = args;
 
   const { systemMessages, promptMessages } = getSystemPromptMessages(chatMessages);
@@ -116,13 +116,15 @@ function getAnthropicToolChoice(useTools?: boolean, name?: AITool): AnthropicToo
 
 export async function parseAnthropicStream(
   chunks: Stream<Anthropic.Messages.RawMessageStreamEvent>,
-  response: Response
+  response: Response,
+  model: AnthropicModel | BedrockAnthropicModel
 ) {
   const responseMessage: AIMessagePrompt = {
     role: 'assistant',
     content: '',
     contextType: 'userPrompt',
     toolCalls: [],
+    model,
   };
 
   for await (const chunk of chunks) {
@@ -186,12 +188,17 @@ export async function parseAnthropicStream(
   return responseMessage;
 }
 
-export function parseAnthropicResponse(result: Anthropic.Messages.Message, response: Response): AIMessagePrompt {
+export function parseAnthropicResponse(
+  result: Anthropic.Messages.Message,
+  response: Response,
+  model: AnthropicModel | BedrockAnthropicModel
+): AIMessagePrompt {
   const responseMessage: AIMessagePrompt = {
     role: 'assistant',
     content: '',
     contextType: 'userPrompt',
     toolCalls: [],
+    model,
   };
 
   result.content?.forEach(

quadratic-api/src/routes/ai/helpers/bedrock.helper.ts → quadratic-api/src/ai/helpers/bedrock.helper.ts

@@ -4,17 +4,16 @@ import { getSystemPromptMessages } from 'quadratic-shared/ai/helpers/message.hel
 import type { AITool } from 'quadratic-shared/ai/specs/aiToolsSpec';
 import { aiToolsSpec } from 'quadratic-shared/ai/specs/aiToolsSpec';
 import type {
-  AIAutoCompleteRequestBody,
   AIMessagePrompt,
-  BedrockAutoCompleteRequestBody,
+  AIRequestBody,
+  BedrockModel,
   BedrockPromptMessage,
+  BedrockRequestBody,
   BedrockTool,
   BedrockToolChoice,
 } from 'quadratic-shared/typesAndSchemasAI';
 
-export function getBedrockApiArgs(
-  args: Omit<AIAutoCompleteRequestBody, 'model'>
-): Omit<BedrockAutoCompleteRequestBody, 'model'> {
+export function getBedrockApiArgs(args: Omit<AIRequestBody, 'model'>): Omit<BedrockRequestBody, 'model'> {
   const { messages: chatMessages, useTools, toolName } = args;
 
   const { systemMessages, promptMessages } = getSystemPromptMessages(chatMessages);
@@ -114,12 +113,17 @@ function getBedrockToolChoice(useTools?: boolean, name?: AITool): BedrockToolCho
   return toolChoice;
 }
 
-export async function parseBedrockStream(chunks: AsyncIterable<ConverseStreamOutput> | never[], response: Response) {
+export async function parseBedrockStream(
+  chunks: AsyncIterable<ConverseStreamOutput> | never[],
+  response: Response,
+  model: BedrockModel
+) {
   const responseMessage: AIMessagePrompt = {
     role: 'assistant',
     content: '',
     contextType: 'userPrompt',
     toolCalls: [],
+    model,
   };
 
   for await (const chunk of chunks) {
@@ -185,12 +189,17 @@ export async function parseBedrockStream(chunks: AsyncIterable<ConverseStreamOut
   return responseMessage;
 }
 
-export function parseBedrockResponse(result: ConverseOutput | undefined, response: Response): AIMessagePrompt {
+export function parseBedrockResponse(
+  result: ConverseOutput | undefined,
+  response: Response,
+  model: BedrockModel
+): AIMessagePrompt {
   const responseMessage: AIMessagePrompt = {
     role: 'assistant',
     content: '',
     contextType: 'userPrompt',
     toolCalls: [],
+    model,
   };
 
   result?.message?.content?.forEach((contentBlock) => {

quadratic-api/src/routes/ai/helpers/context.helper.ts → quadratic-api/src/ai/helpers/context.helper.ts

@@ -1,10 +1,10 @@
+import { ConnectionDocs } from 'quadratic-api/src/ai/docs/ConnectionDocs';
+import { FormulaDocs } from 'quadratic-api/src/ai/docs/FormulaDocs';
+import { JavascriptDocs } from 'quadratic-api/src/ai/docs/JavascriptDocs';
+import { PythonDocs } from 'quadratic-api/src/ai/docs/PythonDocs';
+import { QuadraticDocs } from 'quadratic-api/src/ai/docs/QuadraticDocs';
 import { aiToolsSpec } from 'quadratic-shared/ai/specs/aiToolsSpec';
 import type { ChatMessage, CodeCellType } from 'quadratic-shared/typesAndSchemasAI';
-import { ConnectionDocs } from '../docs/ConnectionDocs';
-import { FormulaDocs } from '../docs/FormulaDocs';
-import { JavascriptDocs } from '../docs/JavascriptDocs';
-import { PythonDocs } from '../docs/PythonDocs';
-import { QuadraticDocs } from '../docs/QuadraticDocs';
 
 export const getQuadraticContext = (language?: CodeCellType): ChatMessage[] => [
   {

quadratic-api/src/routes/ai/helpers/openai.helper.ts → quadratic-api/src/ai/helpers/openai.helper.ts

@@ -5,17 +5,16 @@ import { getSystemPromptMessages } from 'quadratic-shared/ai/helpers/message.hel
 import type { AITool } from 'quadratic-shared/ai/specs/aiToolsSpec';
 import { aiToolsSpec } from 'quadratic-shared/ai/specs/aiToolsSpec';
 import type {
-  AIAutoCompleteRequestBody,
   AIMessagePrompt,
-  OpenAIAutoCompleteRequestBody,
+  AIRequestBody,
+  OpenAIModel,
   OpenAIPromptMessage,
+  OpenAIRequestBody,
   OpenAITool,
   OpenAIToolChoice,
 } from 'quadratic-shared/typesAndSchemasAI';
 
-export function getOpenAIApiArgs(
-  args: Omit<AIAutoCompleteRequestBody, 'model'>
-): Omit<OpenAIAutoCompleteRequestBody, 'model'> {
+export function getOpenAIApiArgs(args: Omit<AIRequestBody, 'model'>): Omit<OpenAIRequestBody, 'model'> {
   const { messages: chatMessages, useTools, toolName } = args;
 
   const { systemMessages, promptMessages } = getSystemPromptMessages(chatMessages);
@@ -108,13 +107,15 @@ function getOpenAIToolChoice(useTools?: boolean, name?: AITool): OpenAIToolChoic
 
 export async function parseOpenAIStream(
   chunks: Stream<OpenAI.Chat.Completions.ChatCompletionChunk>,
-  response: Response
+  response: Response,
+  model: OpenAIModel
 ) {
   const responseMessage: AIMessagePrompt = {
     role: 'assistant',
     content: '',
     contextType: 'userPrompt',
     toolCalls: [],
+    model,
   };
 
   for await (const chunk of chunks) {
@@ -193,13 +194,15 @@ export async function parseOpenAIStream(
 
 export function parseOpenAIResponse(
   result: OpenAI.Chat.Completions.ChatCompletion,
-  response: Response
+  response: Response,
+  model: OpenAIModel
 ): AIMessagePrompt {
   const responseMessage: AIMessagePrompt = {
     role: 'assistant',
     content: '',
     contextType: 'userPrompt',
     toolCalls: [],
+    model,
   };
 
   const message = result.choices[0].message;

quadratic-api/src/routes/ai/aiRateLimiter.ts → quadratic-api/src/ai/middleware/aiRateLimiter.ts

@@ -1,6 +1,6 @@
 import rateLimit from 'express-rate-limit';
-import { RATE_LIMIT_AI_REQUESTS_MAX, RATE_LIMIT_AI_WINDOW_MS } from '../../env-vars';
-import type { Request } from '../../types/Request';
+import { RATE_LIMIT_AI_REQUESTS_MAX, RATE_LIMIT_AI_WINDOW_MS } from 'quadratic-api/src/env-vars';
+import type { Request } from 'quadratic-api/src/types/Request';
 
 export const ai_rate_limiter = rateLimit({
   windowMs: Number(RATE_LIMIT_AI_WINDOW_MS) || 3 * 60 * 60 * 1000, // 3 hours

quadratic-api/src/app.ts

@@ -7,10 +7,9 @@ import 'express-async-errors';
 import fs from 'fs';
 import helmet from 'helmet';
 import path from 'path';
+import internal_router from 'quadratic-api/src/routes/internal';
+import { ApiError } from 'quadratic-api/src/utils/ApiError';
 import { CORS, NODE_ENV, SENTRY_DSN } from './env-vars';
-import ai_router from './routes/ai/ai';
-import internal_router from './routes/internal';
-import { ApiError } from './utils/ApiError';
 export const app = express();
 
 // Configure Sentry
@@ -68,8 +67,6 @@ app.get('/', (req, res) => {
 });
 
 // App routes
-// TODO: eventually move all of these into the `v0` directory and register them dynamically
-app.use('/ai', ai_router);
 // Internal routes
 app.use('/v0/internal', internal_router);