Initial proxy artifact

This commit is contained in:
hwhung
2026-05-07 19:13:04 +08:00
commit 8c68a30044
20 changed files with 1309 additions and 0 deletions
+6
View File
@@ -0,0 +1,6 @@
dist/
node_modules/
.pnpm-store/
config.json
config.local.json
pnpm-lock.yaml
+2
View File
@@ -0,0 +1,2 @@
run = "PORT=${PORT:-3000} PROXY_API_KEY=${PROXY_API_KEY:-dev-proxy-key} pnpm prod"
onBoot = "pnpm install --ignore-workspace"
+23
View File
@@ -0,0 +1,23 @@
kind = "web"
previewPath = "/"
title = "Proxy Artifact"
version = "1.0.0"
id = "proxy-artifact"
router = "path"
[[services]]
name = "web"
paths = ["/"]
localPort = 3000
[services.development]
run = "PORT=${PORT:-3000} PROXY_API_KEY=${PROXY_API_KEY:-dev-proxy-key} pnpm prod"
[services.production]
build = ["sh", "-c", "pnpm install --ignore-workspace && pnpm build:fast"]
run = ["sh", "-c", "PORT=${PORT:-3000} PROXY_API_KEY=${PROXY_API_KEY:-dev-proxy-key} NODE_ENV=production node --enable-source-maps dist/index.mjs"]
[services.env]
PORT = "3000"
PROXY_API_KEY = "dev-proxy-key"
BASE_PATH = "/"
+128
View File
@@ -0,0 +1,128 @@
import { createRequire } from "node:module";
import path from "node:path";
import { fileURLToPath } from "node:url";
import { build as esbuild } from "esbuild";
import esbuildPluginPino from "esbuild-plugin-pino";
import { rm } from "node:fs/promises";
// Plugins (e.g. 'esbuild-plugin-pino') may use `require` to resolve dependencies
globalThis.require = createRequire(import.meta.url);
const artifactDir = path.dirname(fileURLToPath(import.meta.url));
async function buildAll() {
const distDir = path.resolve(artifactDir, "dist");
await rm(distDir, { recursive: true, force: true });
await esbuild({
entryPoints: [path.resolve(artifactDir, "src/index.ts")],
platform: "node",
bundle: true,
format: "esm",
outdir: distDir,
outExtension: { ".js": ".mjs" },
logLevel: "info",
// Some packages may not be bundleable, so we externalize them, we can add more here as needed.
// Some of the packages below may not be imported or installed, but we're adding them in case they are in the future.
// Examples of unbundleable packages:
// - uses native modules and loads them dynamically (e.g. sharp)
// - use path traversal to read files (e.g. @google-cloud/secret-manager loads sibling .proto files)
external: [
"*.node",
"openai",
"@anthropic-ai/sdk",
"sharp",
"better-sqlite3",
"sqlite3",
"canvas",
"bcrypt",
"argon2",
"fsevents",
"re2",
"farmhash",
"xxhash-addon",
"bufferutil",
"utf-8-validate",
"ssh2",
"cpu-features",
"dtrace-provider",
"isolated-vm",
"lightningcss",
"pg-native",
"oracledb",
"mongodb-client-encryption",
"nodemailer",
"handlebars",
"knex",
"typeorm",
"protobufjs",
"onnxruntime-node",
"@tensorflow/*",
"@prisma/client",
"@mikro-orm/*",
"@grpc/*",
"@swc/*",
"@aws-sdk/*",
"@azure/*",
"@opentelemetry/*",
"@google-cloud/*",
"@google/*",
"googleapis",
"firebase-admin",
"@parcel/watcher",
"@sentry/profiling-node",
"@tree-sitter/*",
"aws-sdk",
"classic-level",
"dd-trace",
"ffi-napi",
"grpc",
"hiredis",
"kerberos",
"leveldown",
"miniflare",
"mysql2",
"newrelic",
"odbc",
"piscina",
"realm",
"ref-napi",
"rocksdb",
"sass-embedded",
"sequelize",
"serialport",
"snappy",
"tinypool",
"usb",
"workerd",
"wrangler",
"zeromq",
"zeromq-prebuilt",
"playwright",
"puppeteer",
"puppeteer-core",
"electron",
],
sourcemap: "linked",
plugins: [
// pino relies on workers to handle logging, instead of externalizing it we use a plugin to handle it
esbuildPluginPino({ transports: ["pino-pretty"] })
],
// Make sure packages that are cjs only (e.g. express) but are bundled continue to work in our esm output file
banner: {
js: `import { createRequire as __bannerCrReq } from 'node:module';
import __bannerPath from 'node:path';
import __bannerUrl from 'node:url';
globalThis.require = __bannerCrReq(import.meta.url);
globalThis.__filename = __bannerUrl.fileURLToPath(import.meta.url);
globalThis.__dirname = __bannerPath.dirname(globalThis.__filename);
`,
},
});
}
buildAll().catch((err) => {
console.error(err);
process.exit(1);
});
+35
View File
@@ -0,0 +1,35 @@
{
"name": "proxy-artifact",
"version": "0.0.0",
"license": "MIT",
"type": "module",
"scripts": {
"dev": "tsx src/index.ts",
"build": "tsc --noEmit && node build.mjs",
"build:fast": "node build.mjs",
"prod": "node build.mjs && node --enable-source-maps dist/index.mjs",
"start": "node --enable-source-maps dist/index.mjs",
"typecheck": "tsc --noEmit"
},
"private": true,
"dependencies": {
"@anthropic-ai/sdk": "^0.91.1",
"cors": "^2.8.6",
"express": "^5.2.1",
"openai": "^6.34.0",
"pino": "^9.14.0",
"pino-http": "^10.5.0"
},
"devDependencies": {
"@types/cors": "^2.8.19",
"@types/express": "^5.0.6",
"@types/node": "^25.3.5",
"esbuild": "0.27.3",
"esbuild-plugin-pino": "^2.3.3",
"pino-pretty": "^13.1.3",
"prettier": "^3.8.1",
"thread-stream": "3.1.0",
"tsx": "^4.21.0",
"typescript": "~5.9.2"
}
}
+2
View File
@@ -0,0 +1,2 @@
packages:
- "."
+53
View File
@@ -0,0 +1,53 @@
# Proxy Artifact
Standalone TypeScript API proxy packaged as a Replit `kind = "web"` artifact.
## Replit
The repo includes:
- `.replit` for the Run button
- `.replit-artifact/artifact.toml` for Replit Publish
Run command:
```bash
PORT=${PORT:-3000} PROXY_API_KEY=${PROXY_API_KEY:-dev-proxy-key} pnpm prod
```
Production publish build:
```bash
pnpm install --ignore-workspace && pnpm build:fast
```
Production publish run:
```bash
PORT=${PORT:-3000} PROXY_API_KEY=${PROXY_API_KEY:-dev-proxy-key} NODE_ENV=production node --enable-source-maps dist/index.mjs
```
## Config
The app reads Replit AI Integration variables automatically:
- `AI_INTEGRATIONS_OPENAI_BASE_URL`
- `AI_INTEGRATIONS_OPENAI_API_KEY`
- `AI_INTEGRATIONS_ANTHROPIC_BASE_URL`
- `AI_INTEGRATIONS_ANTHROPIC_API_KEY`
Server/proxy overrides:
- `HOST`
- `PORT`
- `PROXY_API_KEY`
CLI flags still work for manual runs, but mass deployments should use environment variables.
## Commands
- `pnpm install --ignore-workspace`
- `pnpm prod`
- `pnpm run build:fast`
- `pnpm run build`
- `pnpm run typecheck`
+59
View File
@@ -0,0 +1,59 @@
# Workspace
## Overview
Standalone TypeScript API server packaged as a Replit `kind = "web"` artifact.
## Stack
- **Node.js version**: 24
- **Package manager**: pnpm
- **TypeScript version**: 5.9
- **API framework**: Express 5
- **Build**: esbuild
## Provider Endpoints
Use Replit AI Integrations endpoints. The app reads these environment variables automatically:
- `AI_INTEGRATIONS_OPENAI_BASE_URL`
- `AI_INTEGRATIONS_OPENAI_API_KEY`
- `AI_INTEGRATIONS_ANTHROPIC_BASE_URL`
- `AI_INTEGRATIONS_ANTHROPIC_API_KEY`
CLI flags are still available for manual overrides:
- `--openai-base-url`
- `--openai-api-key`
- `--anthropic-base-url`
- `--anthropic-api-key`
Never use public official provider endpoints.
## Key Commands
- `pnpm install --ignore-workspace` — install dependencies without joining a parent workspace
- `pnpm prod` — build and run API server locally on port `3000`
- `pnpm run typecheck` — typecheck the server
- `pnpm run build` — typecheck and build `dist/index.mjs`
- `pnpm run build:fast` — build `dist/index.mjs` without typechecking
- `pnpm start` — run the built server
## Replit Configuration
- `.replit` `run` uses `pnpm prod`
- `.replit` `onBoot` uses `pnpm install --ignore-workspace`
- `.replit-artifact/artifact.toml` marks this repo as a publishable `kind = "web"` artifact
## Server Overrides
- `--host` or `HOST`
- `--port` or `PORT`
- `--proxy-api-key` or `PROXY_API_KEY`
## Routes
- `/` and `/dashboard` — static model dashboard
- `/api/healthz` — health check
- `/v1/models` — OpenAI-compatible model list
- `/v1/chat/completions` — OpenAI-compatible chat proxy
+55
View File
@@ -0,0 +1,55 @@
Clone and run https://git.komnenos.me/springfw/proxy-artifact in ./proxy-artifact.
Goal: get the publishable web artifact running once on port 3000 with OpenAI and Anthropic
Replit AI Integrations wired.
Bearer token for this deployment: DEPLOYMENT_GENERATED_PROXY_TOKEN
Important: wire OpenAI and Anthropic via the ai-integrations-openai and
ai-integrations-anthropic skills. Call setupReplitAIIntegrations directly. Do not call
searchIntegrations. OpenAI and Anthropic are Replit AI Integrations that provide
AI_INTEGRATIONS_* variables; they are not OAuth connectors.
Do not:
- add it to a parent workspace
- move it into artifacts/
- modify source files
- edit package.json
- create config.json or /tmp config files
- inspect/free unrelated ports
- use nohup/setsid/background shell tricks
- call searchIntegrations
- search the Replit integrations catalog
- create provider API secrets manually
- use public OpenAI or Anthropic endpoints
- invent, rotate, print, store, or replace the bearer token
- run a foreground server command directly in the shell
- run verification commands
- run chat-completion smoke tests unless asked
Steps:
1. Clone into ./proxy-artifact.
2. Wire OpenAI and Anthropic via the ai-integrations-openai and ai-integrations-anthropic
skills:
- call setupReplitAIIntegrations directly
- do not call searchIntegrations
3. From ./proxy-artifact, run:
pnpm install --ignore-workspace
4. Ensure ./proxy-artifact/.replit-artifact/artifact.toml remains kind = "web".
5. Use the existing artifact run command, but set PROXY_API_KEY to DEPLOYMENT_GENERATED_PROXY_TOKEN
for this deployment.
6. Start/restart the artifact/workflow once.
7. Do not run health, dashboard, model-list, or chat-completion verification commands.
Report only:
- local URL
- whether setupReplitAIIntegrations wired OpenAI and Anthropic
- whether the kind = "web" artifact was used
- that no verification commands were run
+38
View File
@@ -0,0 +1,38 @@
import express, { type Express } from "express";
import cors from "cors";
import pinoHttp from "pino-http";
import dashboardRouter from "./routes/dashboard";
import router from "./routes";
import proxyRouter from "./routes/proxy";
import { logger } from "./lib/logger";
const app: Express = express();
app.use(
pinoHttp({
logger,
serializers: {
req(req) {
return {
id: req.id,
method: req.method,
url: req.url?.split("?")[0],
};
},
res(res) {
return {
statusCode: res.statusCode,
};
},
},
}),
);
app.use(cors());
app.use(express.json({ limit: "50mb" }));
app.use(express.urlencoded({ extended: true, limit: "50mb" }));
app.use(dashboardRouter);
app.use("/api", router);
app.use("/v1", proxyRouter);
export default app;
+15
View File
@@ -0,0 +1,15 @@
import app from "./app";
import { config } from "./lib/config";
import { logger } from "./lib/logger";
const port = config.server.port;
const host = config.server.host;
app.listen(port, host, (err) => {
if (err) {
logger.error({ err }, "Error listening on port");
process.exit(1);
}
logger.info({ host, port }, "Server listening");
});
+110
View File
@@ -0,0 +1,110 @@
export interface ProviderConfig {
baseURL?: string;
apiKey?: string;
}
export interface AppConfig {
server: {
host: string;
port: number;
};
proxy: {
apiKey: string;
};
providers: {
openai: ProviderConfig;
anthropic: ProviderConfig;
};
logging: {
level: string;
pretty: boolean;
};
}
const DEFAULT_CONFIG: AppConfig = {
server: {
host: "0.0.0.0",
port: 3000,
},
proxy: {
apiKey: "dev-proxy-key",
},
providers: {
openai: {},
anthropic: {},
},
logging: {
level: "info",
pretty: true,
},
};
function argValue(name: string): string | undefined {
const longName = `--${name}`;
const index = process.argv.findIndex((arg) => arg === longName);
if (index >= 0) {
const value = process.argv[index + 1];
return value && !value.startsWith("--") ? value : undefined;
}
const inlineArg = process.argv.find((arg) => arg.startsWith(`${longName}=`));
return inlineArg ? inlineArg.slice(longName.length + 1) : undefined;
}
function argFlag(name: string): boolean {
return process.argv.includes(`--${name}`);
}
function envValue(name: string): string | undefined {
const value = process.env[name];
return value && value.length > 0 ? value : undefined;
}
function applyArgOverrides(base: AppConfig): AppConfig {
const rawPort = argValue("port") ?? envValue("PORT");
const port = rawPort === undefined ? base.server.port : Number(rawPort);
if (!Number.isInteger(port) || port <= 0) {
throw new Error(`Invalid port value: ${rawPort}`);
}
const host = argValue("host") ?? envValue("HOST") ?? base.server.host;
const proxyApiKey = argValue("proxy-api-key") ?? envValue("PROXY_API_KEY") ?? base.proxy.apiKey;
const openaiBaseURL = argValue("openai-base-url") ?? envValue("AI_INTEGRATIONS_OPENAI_BASE_URL");
const openaiApiKey = argValue("openai-api-key") ?? envValue("AI_INTEGRATIONS_OPENAI_API_KEY");
const anthropicBaseURL =
argValue("anthropic-base-url") ?? envValue("AI_INTEGRATIONS_ANTHROPIC_BASE_URL");
const anthropicApiKey =
argValue("anthropic-api-key") ?? envValue("AI_INTEGRATIONS_ANTHROPIC_API_KEY");
return {
server: {
host,
port,
},
proxy: {
apiKey: proxyApiKey,
},
providers: {
openai: {
...base.providers.openai,
...(openaiBaseURL ? { baseURL: openaiBaseURL } : {}),
...(openaiApiKey ? { apiKey: openaiApiKey } : {}),
},
anthropic: {
...base.providers.anthropic,
...(anthropicBaseURL ? { baseURL: anthropicBaseURL } : {}),
...(anthropicApiKey ? { apiKey: anthropicApiKey } : {}),
},
},
logging: {
level: argValue("log-level") ?? base.logging.level,
pretty: argFlag("pretty-logs")
? true
: argFlag("no-pretty-logs")
? false
: base.logging.pretty,
},
};
}
export const config = applyArgOverrides(DEFAULT_CONFIG);
+19
View File
@@ -0,0 +1,19 @@
import pino from "pino";
import { config } from "./config";
export const logger = pino({
level: config.logging.level,
redact: [
"req.headers.authorization",
"req.headers.cookie",
"res.headers['set-cookie']",
],
...(config.logging.pretty
? {
transport: {
target: "pino-pretty",
options: { colorize: true },
},
}
: {}),
});
+38
View File
@@ -0,0 +1,38 @@
export const REASONING_BASE_MODELS = [
{ id: "gpt-5.2", provider: "openai" as const },
{ id: "gpt-5-mini", provider: "openai" as const },
{ id: "gpt-5-nano", provider: "openai" as const },
{ id: "o4-mini", provider: "openai" as const },
{ id: "o3", provider: "openai" as const },
{ id: "claude-opus-4-7", provider: "anthropic" as const },
{ id: "claude-opus-4-6", provider: "anthropic" as const },
{ id: "claude-sonnet-4-6", provider: "anthropic" as const },
{ id: "claude-haiku-4-5", provider: "anthropic" as const },
];
export const EFFORT_LEVELS = ["minimal", "low", "medium", "high"] as const;
export type EffortLevel = (typeof EFFORT_LEVELS)[number];
export interface ModelEntry {
id: string;
provider: "openai" | "anthropic";
baseModel: string;
effort: EffortLevel | null;
}
export const MODELS: ModelEntry[] = REASONING_BASE_MODELS.flatMap((m) => {
const variants: ModelEntry[] = [
{ id: m.id, provider: m.provider, baseModel: m.id, effort: null },
];
for (const effort of EFFORT_LEVELS) {
variants.push({
id: `${m.id}-${effort}`,
provider: m.provider,
baseModel: m.id,
effort,
});
}
return variants;
});
export const MODEL_INDEX: Map<string, ModelEntry> = new Map(MODELS.map((m) => [m.id, m]));
+151
View File
@@ -0,0 +1,151 @@
import { Router, type IRouter } from "express";
import { MODELS } from "../lib/models";
const router: IRouter = Router();
function escapeHtml(value: string): string {
return value
.replaceAll("&", "&amp;")
.replaceAll("<", "&lt;")
.replaceAll(">", "&gt;")
.replaceAll('"', "&quot;");
}
function modelRows(): string {
return MODELS.map((model) => {
const effort = model.effort ?? "default";
return `<tr>
<td><code>${escapeHtml(model.id)}</code></td>
<td>${escapeHtml(model.provider)}</td>
<td><code>${escapeHtml(model.baseModel)}</code></td>
<td>${escapeHtml(effort)}</td>
</tr>`;
}).join("");
}
function dashboardHtml(): string {
return `<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>API Server Models</title>
<style>
:root {
color-scheme: light dark;
font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
background: #f7f8fa;
color: #1b1f24;
}
body {
margin: 0;
min-height: 100vh;
background: #f7f8fa;
}
main {
max-width: 1080px;
margin: 0 auto;
padding: 32px 20px 48px;
}
header {
display: flex;
align-items: end;
justify-content: space-between;
gap: 16px;
margin-bottom: 20px;
}
h1 {
margin: 0;
font-size: 28px;
line-height: 1.2;
font-weight: 650;
}
.meta {
color: #59616c;
font-size: 14px;
}
.table-wrap {
overflow-x: auto;
border: 1px solid #d9dee7;
border-radius: 8px;
background: #ffffff;
}
table {
width: 100%;
border-collapse: collapse;
min-width: 720px;
}
th, td {
padding: 12px 14px;
border-bottom: 1px solid #e8ebf0;
text-align: left;
font-size: 14px;
white-space: nowrap;
}
th {
background: #eef1f5;
color: #3c434d;
font-size: 12px;
letter-spacing: 0;
text-transform: uppercase;
}
tr:last-child td {
border-bottom: 0;
}
code {
font-family: "SFMono-Regular", Consolas, "Liberation Mono", monospace;
font-size: 13px;
}
@media (prefers-color-scheme: dark) {
:root, body {
background: #101316;
color: #f1f4f8;
}
.meta {
color: #aab3bf;
}
.table-wrap {
background: #171b20;
border-color: #303741;
}
th {
background: #20262d;
color: #c1cad6;
}
th, td {
border-bottom-color: #29313a;
}
}
</style>
</head>
<body>
<main>
<header>
<h1>Model List</h1>
<div class="meta">${MODELS.length} models</div>
</header>
<div class="table-wrap">
<table>
<thead>
<tr>
<th>Model</th>
<th>Provider</th>
<th>Base Model</th>
<th>Effort</th>
</tr>
</thead>
<tbody>
${modelRows()}
</tbody>
</table>
</div>
</main>
</body>
</html>`;
}
router.get(["/", "/dashboard"], (_req, res) => {
res.type("html").send(dashboardHtml());
});
export default router;
+9
View File
@@ -0,0 +1,9 @@
import { Router, type IRouter } from "express";
const router: IRouter = Router();
router.get("/healthz", (_req, res) => {
res.json({ status: "ok" });
});
export default router;
+8
View File
@@ -0,0 +1,8 @@
import { Router, type IRouter } from "express";
import healthRouter from "./health";
const router: IRouter = Router();
router.use(healthRouter);
export default router;
+525
View File
@@ -0,0 +1,525 @@
import { Router, type IRouter, type Request, type Response } from "express";
import OpenAI from "openai";
import Anthropic from "@anthropic-ai/sdk";
import type { MessageCreateParamsNonStreaming } from "@anthropic-ai/sdk/resources/messages";
import { config } from "../lib/config";
import { MODEL_INDEX, MODELS } from "../lib/models";
const router: IRouter = Router();
function openAIClient(): OpenAI | null {
const provider = config.providers.openai;
if (!provider.apiKey || !provider.baseURL) return null;
return new OpenAI({
baseURL: provider.baseURL,
apiKey: provider.apiKey,
});
}
function anthropicClient(): Anthropic | null {
const provider = config.providers.anthropic;
if (!provider.apiKey || !provider.baseURL) return null;
return new Anthropic({
baseURL: provider.baseURL,
apiKey: provider.apiKey,
});
}
function flush(res: Response): void {
(res as Response & { flush?: () => void }).flush?.();
}
function authorize(req: Request, res: Response): boolean {
const expected = config.proxy.apiKey;
const header = req.header("authorization") ?? "";
const match = header.match(/^Bearer\s+(.+)$/i);
if (!match || match[1] !== expected) {
res.status(401).json({ error: { message: "Invalid API key", type: "invalid_request_error" } });
return false;
}
return true;
}
router.get("/models", (req, res) => {
if (!authorize(req, res)) return;
const created = Math.floor(Date.now() / 1000);
res.json({
object: "list",
data: MODELS.map((m) => ({
id: m.id,
object: "model",
created,
owned_by: m.provider,
})),
});
});
interface ChatMessage {
role: "system" | "user" | "assistant" | "developer" | "tool";
content: unknown;
name?: string;
}
interface ChatCompletionRequest {
model: string;
messages: ChatMessage[];
stream?: boolean;
temperature?: number;
top_p?: number;
max_tokens?: number;
max_completion_tokens?: number;
stop?: string | string[];
frequency_penalty?: number;
presence_penalty?: number;
user?: string;
reasoning_effort?: "minimal" | "low" | "medium" | "high";
thinking?: { type?: string; budget_tokens?: number } | null;
[key: string]: unknown;
}
function effortToBudget(effort: string | undefined): number | null {
switch (effort) {
case "minimal":
return 1024;
case "low":
return 2048;
case "medium":
return 8000;
case "high":
return 16000;
default:
return null;
}
}
function isOpenAIModel(model: string): boolean {
return /^(gpt|o\d|chatgpt)/i.test(model);
}
function isAnthropicModel(model: string): boolean {
return /^claude/i.test(model);
}
function newId(): string {
return "chatcmpl-" + Math.random().toString(36).slice(2, 14);
}
router.post("/chat/completions", async (req, res) => {
if (!authorize(req, res)) return;
const body = req.body as ChatCompletionRequest;
if (!body || typeof body !== "object" || !body.model || !Array.isArray(body.messages)) {
res.status(400).json({
error: { message: "model and messages are required", type: "invalid_request_error" },
});
return;
}
const requestedId = body.model;
const entry = MODEL_INDEX.get(requestedId);
let provider: "openai" | "anthropic";
if (entry) {
body.model = entry.baseModel;
if (entry.effort && !body.reasoning_effort) {
body.reasoning_effort = entry.effort;
}
provider = entry.provider;
} else if (isOpenAIModel(requestedId)) {
provider = "openai";
} else if (isAnthropicModel(requestedId)) {
provider = "anthropic";
} else {
res.status(400).json({
error: {
message: `Unsupported model: ${requestedId}`,
type: "invalid_request_error",
},
});
return;
}
const stream = body.stream === true;
if (provider === "openai") {
await handleOpenAI(req, res, body, stream);
} else {
await handleAnthropic(req, res, body, stream);
}
});
function setupSse(res: Response): NodeJS.Timeout {
res.setHeader("Content-Type", "text/event-stream");
res.setHeader("Cache-Control", "no-cache, no-transform");
res.setHeader("Connection", "keep-alive");
res.setHeader("X-Accel-Buffering", "no");
res.flushHeaders();
const keepalive = setInterval(() => {
try {
res.write(": keepalive\n\n");
flush(res);
} catch {
// ignore
}
}, 5000);
return keepalive;
}
async function handleOpenAI(
req: Request,
res: Response,
body: ChatCompletionRequest,
stream: boolean,
): Promise<void> {
const client = openAIClient();
if (!client) {
providerNotConfigured(res, "openai");
return;
}
const { stream: _ignored, ...rest } = body;
if (!stream) {
try {
const completion = await client.chat.completions.create({
...rest,
stream: false,
} as Parameters<typeof client.chat.completions.create>[0]);
res.json(completion);
} catch (err) {
handleProviderError(req, res, err, "openai");
}
return;
}
let keepalive: NodeJS.Timeout | undefined;
let clientGone = false;
res.on("close", () => {
if (!res.writableEnded) clientGone = true;
if (keepalive) clearInterval(keepalive);
});
try {
const upstream = await client.chat.completions.create({
...rest,
stream: true,
} as Parameters<typeof client.chat.completions.create>[0]);
keepalive = setupSse(res);
for await (const chunk of upstream as AsyncIterable<unknown>) {
if (clientGone) break;
res.write(`data: ${JSON.stringify(chunk)}\n\n`);
flush(res);
}
if (!clientGone) {
res.write("data: [DONE]\n\n");
flush(res);
}
} catch (err) {
if (!res.headersSent) {
handleProviderError(req, res, err, "openai");
return;
}
req.log?.error({ err }, "openai stream error");
try {
res.write(
`data: ${JSON.stringify({ error: { message: errorMessage(err), type: "upstream_error" } })}\n\n`,
);
flush(res);
} catch {
// ignore
}
} finally {
if (keepalive) clearInterval(keepalive);
if (!clientGone) {
try {
res.end();
} catch {
// ignore
}
}
}
}
interface AnthropicConvertedMessages {
system: string | undefined;
messages: Array<{ role: "user" | "assistant"; content: unknown }>;
}
function convertMessagesForAnthropic(messages: ChatMessage[]): AnthropicConvertedMessages {
const systemParts: string[] = [];
const out: Array<{ role: "user" | "assistant"; content: unknown }> = [];
for (const msg of messages) {
if (msg.role === "system" || msg.role === "developer") {
if (typeof msg.content === "string") {
systemParts.push(msg.content);
} else if (Array.isArray(msg.content)) {
for (const part of msg.content as Array<{ type?: string; text?: string }>) {
if (part?.type === "text" && typeof part.text === "string") {
systemParts.push(part.text);
}
}
}
} else if (msg.role === "user" || msg.role === "assistant") {
out.push({ role: msg.role, content: msg.content });
}
// ignore tool messages for this passthrough
}
return {
system: systemParts.length > 0 ? systemParts.join("\n\n") : undefined,
messages: out,
};
}
async function handleAnthropic(
req: Request,
res: Response,
body: ChatCompletionRequest,
stream: boolean,
): Promise<void> {
const client = anthropicClient();
if (!client) {
providerNotConfigured(res, "anthropic");
return;
}
const { system, messages } = convertMessagesForAnthropic(body.messages);
let maxTokens =
body.max_completion_tokens ??
body.max_tokens ??
8192;
let thinkingParam: { type: "enabled"; budget_tokens: number } | null = null;
if (body.thinking && (body.thinking.type === "enabled" || typeof body.thinking.budget_tokens === "number")) {
const budget = Math.max(1024, body.thinking.budget_tokens ?? 4096);
thinkingParam = { type: "enabled", budget_tokens: budget };
} else {
const budget = effortToBudget(body.reasoning_effort);
if (budget !== null) {
thinkingParam = { type: "enabled", budget_tokens: budget };
}
}
if (thinkingParam) {
if (maxTokens <= thinkingParam.budget_tokens) {
maxTokens = thinkingParam.budget_tokens + 4096;
}
}
const baseParams: Record<string, unknown> = {
model: body.model,
max_tokens: maxTokens,
messages,
};
if (system) baseParams["system"] = system;
if (thinkingParam) {
baseParams["thinking"] = thinkingParam;
// Anthropic requires temperature=1 (or unset) and rejects top_p when thinking is enabled.
} else {
if (typeof body.temperature === "number") baseParams["temperature"] = body.temperature;
if (typeof body.top_p === "number") baseParams["top_p"] = body.top_p;
}
if (body.stop !== undefined) {
baseParams["stop_sequences"] = Array.isArray(body.stop) ? body.stop : [body.stop];
}
if (!stream) {
try {
const message = await client.messages.create(
baseParams as unknown as MessageCreateParamsNonStreaming,
);
let text = "";
let reasoning = "";
for (const b of message.content ?? []) {
if (b.type === "text") text += b.text;
else if (b.type === "thinking") reasoning += (b as { thinking?: string }).thinking ?? "";
}
const finishReason = mapAnthropicStopReason(message.stop_reason ?? null);
const assistantMessage: Record<string, unknown> = { role: "assistant", content: text };
if (reasoning) assistantMessage["reasoning_content"] = reasoning;
res.json({
id: message.id,
object: "chat.completion",
created: Math.floor(Date.now() / 1000),
model: message.model,
choices: [
{
index: 0,
message: assistantMessage,
finish_reason: finishReason,
},
],
usage: {
prompt_tokens: message.usage?.input_tokens ?? 0,
completion_tokens: message.usage?.output_tokens ?? 0,
total_tokens:
(message.usage?.input_tokens ?? 0) + (message.usage?.output_tokens ?? 0),
},
});
} catch (err) {
handleProviderError(req, res, err, "anthropic");
}
return;
}
let keepalive: NodeJS.Timeout | undefined;
let clientGone = false;
res.on("close", () => {
if (!res.writableEnded) clientGone = true;
if (keepalive) clearInterval(keepalive);
});
const id = newId();
const created = Math.floor(Date.now() / 1000);
try {
const upstream = client.messages.stream(
baseParams as Parameters<typeof client.messages.stream>[0],
);
keepalive = setupSse(res);
res.write(
`data: ${JSON.stringify({
id,
object: "chat.completion.chunk",
created,
model: body.model,
choices: [{ index: 0, delta: { role: "assistant" }, finish_reason: null }],
})}\n\n`,
);
flush(res);
let stopReason: string | null = null;
for await (const event of upstream) {
if (clientGone) break;
if (event.type === "content_block_delta") {
let delta: Record<string, unknown> | null = null;
if (event.delta.type === "text_delta") {
delta = { content: event.delta.text };
} else if (event.delta.type === "thinking_delta") {
const td = event.delta as unknown as { thinking?: string };
if (td.thinking) delta = { reasoning_content: td.thinking };
}
if (delta) {
const chunk = {
id,
object: "chat.completion.chunk",
created,
model: body.model,
choices: [{ index: 0, delta, finish_reason: null }],
};
res.write(`data: ${JSON.stringify(chunk)}\n\n`);
flush(res);
}
} else if (event.type === "message_delta") {
const md = event as unknown as { delta?: { stop_reason?: string | null } };
if (md.delta?.stop_reason) {
stopReason = md.delta.stop_reason;
}
}
}
if (!clientGone) {
const finishChunk = {
id,
object: "chat.completion.chunk",
created,
model: body.model,
choices: [
{
index: 0,
delta: {},
finish_reason: mapAnthropicStopReason(stopReason),
},
],
};
res.write(`data: ${JSON.stringify(finishChunk)}\n\n`);
res.write("data: [DONE]\n\n");
flush(res);
}
} catch (err) {
if (!res.headersSent) {
handleProviderError(req, res, err, "anthropic");
return;
}
req.log?.error({ err }, "anthropic stream error");
try {
res.write(
`data: ${JSON.stringify({ error: { message: errorMessage(err), type: "upstream_error" } })}\n\n`,
);
flush(res);
} catch {
// ignore
}
} finally {
if (keepalive) clearInterval(keepalive);
if (!clientGone) {
try {
res.end();
} catch {
// ignore
}
}
}
}
function mapAnthropicStopReason(reason: string | null): string {
switch (reason) {
case "end_turn":
case "stop_sequence":
return "stop";
case "max_tokens":
return "length";
case "tool_use":
return "tool_calls";
default:
return "stop";
}
}
function errorMessage(err: unknown): string {
if (err instanceof Error) return err.message;
if (typeof err === "string") return err;
try {
return JSON.stringify(err);
} catch {
return "unknown error";
}
}
function providerNotConfigured(res: Response, provider: "openai" | "anthropic"): void {
res.status(503).json({
error: {
message: `${provider} provider is not configured. Pass the matching Replit AI Integrations CLI flags, not public provider endpoints.`,
type: "server_error",
},
});
}
function handleProviderError(
req: Request,
res: Response,
err: unknown,
provider: string,
): void {
req.log?.error({ err, provider }, "provider error");
const status =
typeof (err as { status?: number })?.status === "number"
? (err as { status: number }).status
: 500;
if (!res.headersSent) {
res.status(status).json({
error: {
message: errorMessage(err),
type: "upstream_error",
provider,
},
});
}
}
export default router;
+25
View File
@@ -0,0 +1,25 @@
{
"compilerOptions": {
"isolatedModules": true,
"lib": ["es2022"],
"module": "esnext",
"moduleResolution": "bundler",
"noEmitOnError": true,
"noFallthroughCasesInSwitch": true,
"noImplicitOverride": false,
"noImplicitReturns": true,
"noUnusedLocals": false,
"noImplicitAny": true,
"noImplicitThis": true,
"strictNullChecks": true,
"strictFunctionTypes": false,
"strictBindCallApply": true,
"strictPropertyInitialization": true,
"useUnknownInCatchVariables": true,
"alwaysStrict": true,
"skipLibCheck": true,
"target": "es2022",
"types": [],
"customConditions": ["workspace"]
}
}
+8
View File
@@ -0,0 +1,8 @@
{
"extends": "./tsconfig.base.json",
"compileOnSave": false,
"compilerOptions": {
"types": ["node"]
},
"include": ["src/**/*.ts"]
}