refactor: pipe the read and write process

this refactor simplify the process logic and fix several bugs and
performance issue.

bug fixed:
- cors headers not being sent in some situation
performance:
- perform upstream reqeust while clien is uploading content

.dockerignore

@@ -1,3 +1,4 @@
 openai-api-route
 db.sqlite
 /config.yaml
+/.*

.gitlab-ci.yml

@@ -0,0 +1,52 @@
+# To contribute improvements to CI/CD templates, please follow the Development guide at:
+# https://docs.gitlab.com/ee/development/cicd/templates.html
+# This specific template is located at:
+# https://gitlab.com/gitlab-org/gitlab/-/blob/master/lib/gitlab/ci/templates/Docker.gitlab-ci.yml
+
+# Build a Docker image with CI/CD and push to the GitLab registry.
+# Docker-in-Docker documentation: https://docs.gitlab.com/ee/ci/docker/using_docker_build.html
+#
+# This template uses one generic job with conditional builds
+# for the default branch and all other (MR) branches.
+
+docker-build:
+  # Use the official docker image.
+  image: docker:cli
+  stage: build
+  services:
+    - docker:dind
+  variables:
+    CI_REGISTRY: registry.waykey.net:7999
+    CI_REGISTRY_IMAGE: $CI_REGISTRY/spiderman/datamining/openai-api-route
+    DOCKER_IMAGE_NAME: $CI_REGISTRY_IMAGE:$CI_COMMIT_REF_SLUG
+  before_script:
+    - docker login -u "$CI_REGISTRY_USER" -p "$CI_REGISTRY_PASSWORD" $CI_REGISTRY
+  # All branches are tagged with $DOCKER_IMAGE_NAME (defaults to commit ref slug)
+  # Default branch is also tagged with `latest`
+  script:
+    - docker build --pull -t "$DOCKER_IMAGE_NAME" .
+    - docker push "$DOCKER_IMAGE_NAME"
+    - |
+      if [[ "$CI_COMMIT_BRANCH" == "$CI_DEFAULT_BRANCH" ]]; then
+        docker tag "$DOCKER_IMAGE_NAME" "$CI_REGISTRY_IMAGE:latest"
+        docker push "$CI_REGISTRY_IMAGE:latest"
+      fi
+  # Run this job in a branch where a Dockerfile exists
+  rules:
+    - if: $CI_COMMIT_BRANCH
+      exists:
+        - Dockerfile
+
+deploy:
+  environment: production
+  image: kroniak/ssh-client
+  stage: deploy
+  before_script:
+    - chmod 600 $CI_SSH_PRIVATE_KEY
+  script:
+    - ssh -o StrictHostKeyChecking=no -i $CI_SSH_PRIVATE_KEY root@192.168.1.13 "cd /mnt/data/srv/openai-api-route && podman-compose pull && podman-compose down && podman-compose up -d"
+  rules:
+    - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
+      exists:
+        - Dockerfile
+

Dockerfile

@@ -1,4 +1,4 @@
-FROM golang:1.21 as builder
+FROM docker.io/golang:1.21 as builder
 
 WORKDIR /app
 

README.md

@@ -7,15 +7,134 @@
 - 自定义 Authorization 验证头
 - 支持所有类型的接口 (`/v1/*`)
 - 提供 Prometheus Metrics 统计接口 (`/v1/metrics`)
-- 按照定义顺序请求 OpenAI 上游
-- 识别 ChatCompletions Stream 请求，针对 Stream 请求使用 5 秒超时。具体超时策略请参阅 [超时策略](#超时策略) 一节
-- 记录完整的请求内容、使用的上游、IP 地址、响应时间以及 GPT 回复文本
-- 请求出错时发送 飞书 或 Matrix 消息通知
+- 按照定义顺序请求 OpenAI 上游，出错或超时自动按顺序尝试下一个
+- 识别 ChatCompletions Stream 请求，针对 Stream 请求使用更短的超时。具体超时策略请参阅 [超时策略](#超时策略) 一节
+- 有选择地记录请求内容、请求头、使用的上游、IP 地址、响应时间以及响应等内容。具体记录策略请参阅 [记录策略](#记录策略) 一节
+- 请求出错时发送 飞书 或 Matrix 平台的消息通知
+- 支持 Replicate 平台上的 mistral 模型（beta）
 
 本文档详细介绍了如何使用负载均衡和能力 API 的方法和端点。
 
+## 配置文件
+
+默认情况下程序会使用当前目录下的 `config.yaml` 文件，您可以通过使用 `-config your-config.yaml` 参数指定配置文件路径。
+
+以下是一个配置文件示例，你可以在 `config.sample.yaml` 文件中找到同样的内容
+
+```yaml
+authorization: woshimima
+
+# 默认超时时间，默认 120 秒，流式请求是 10 秒
+timeout: 120
+stream_timeout: 10 
+
+# 使用 sqlite 作为数据库储存请求记录
+dbtype: sqlite
+dbaddr: ./db.sqlite
+
+# 使用 postgres 作为数据库储存请求记录
+# dbtype: postgres
+# dbaddr: "host=127.0.0.1 port=5432 user=postgres dbname=openai_api_route sslmode=disable password=woshimima"
+
+upstreams:
+  - sk: hahaha
+    endpoint: "https://localhost:8888/v1"
+    allow:
+      # whisper 等非 JSON API 识别不到 model，则使用 URL 路径作为模型名称
+      - /v1/audio/transcriptions
+
+  - sk: "secret_key_1"
+    endpoint: "https://api.openai.com/v2"
+    timeout: 120  # 请求超时时间，默认120秒
+    stream_timeout: 10  # 如果识别到 stream: true, 则使用该超时时间
+    allow:  # 可选的模型白名单
+      - gpt-3.5-trubo
+      - gpt-3.5-trubo-0613
+
+  # 您可以设置很多个上游，程序将依次按顺序尝试
+  - sk: "secret_key_2"
+    endpoint: "https://api.openai.com/v1"
+    timeout: 30
+    deny: 
+      - gpt-4
+
+  - sk: "key_for_replicate"
+    type: replicate
+    allow:
+      - mistralai/mixtral-8x7b-instruct-v0.1
+```
+
+### 配置多个验证头
+
+您可以使用英文逗号 `,` 分割多个验证头。每个验证头都是有效的，程序会记录每个请求使用的验证头
+
+```yaml
+authorization: woshimima,iampassword
+```
+
+您也可以为上游单独设置验证头
+
+```yaml
+authorization: woshimima,iampassword
+upstreams:
+  - sk: key
+    authorization: woshimima
+```
+
+如此，只有携带 `woshimima` 验证头的用户可以使用该上游。
+
+### 复杂配置示例
+
+```yaml
+
+# 默认验证头
+authorization: woshimima
+
+upstreams:
+
+  # 允许所有人使用的文字转语音
+  - sk: xxx
+    endpoint: http://localhost:5000/v1
+    noauth: true
+    allow:
+      - /v1/audio/transcriptions
+    
+  # guest 专用的 gpt-3.5-turbo-0125 模型
+  - sk: 
+    endpoint: https://api.xxx.local/v1
+    authorization: guest
+    allow:
+      - gpt-3.5-turbo-0125
+```
+
 ## 部署方法
 
+有两种推荐的部署方法：
+
+1. 使用预先构建好的容器 `docker.io/heimoshuiyu/openai-api-route:latest`
+2. 自行编译
+
+### 使用容器运行
+
+> 注意，如果您使用 sqlite 数据库，您可能还需要修改配置文件以将 SQLite 数据库文件放置在数据卷中。
+
+```bash
+docker run -d --name openai-api-route -v /path/to/config.yaml:/config.yaml docker.io/heimoshuiyu/openai-api-route:latest
+```
+
+使用 Docker Compose
+
+```yaml
+version: '3'
+services:
+  openai-api-route:
+    image: docker.io/heimoshuiyu/openai-api-route:latest
+    ports:
+      - 8888:8888
+    volumes:
+      - ./config.yaml:/config.yaml
+```
+
 ### 编译
 
 以下是编译和运行该负载均衡 API 的步骤：
@@ -40,74 +159,11 @@
    ./openai-api-route
    ```
 
-   默认情况下，API 将会在本地的 8888 端口进行监听。
+## 模型允许与屏蔽列表
 
-   如果您希望使用不同的监听地址，可以使用 `-addr` 参数来指定，例如：
+如果对某个上游设置了 allow 或 deny 列表，则负载均衡只允许或禁用用户使用这些模型。负载均衡程序会先判断白名单，再判断黑名单。
 
-   ```
-   ./openai-api-route -addr 0.0.0.0:8080
-   ```
-
-   这将会将监听地址设置为 0.0.0.0:8080。
-
-6. 如果数据库不存在，系统会自动创建一个名为 `db.sqlite` 的数据库文件。
-
-   如果您希望使用不同的数据库地址，可以使用 `-database` 参数来指定，例如：
-
-   ```
-   ./openai-api-route -database /path/to/database.db
-   ```
-
-   这将会将数据库地址设置为 `/path/to/database.db`。
-
-7. 现在，您已经成功编译并运行了负载均衡和能力 API。您可以根据需要添加上游、管理上游，并使用 API 进行相关操作。
-
-### 运行
-
-以下是运行命令的用法：
-
-```
-Usage of ./openai-api-route:
-  -addr string
-        监听地址（默认为 ":8888"）
-  -upstreams string
-        上游配置文件（默认为 "./upstreams.yaml"）
-  -dbtype
-        数据库类型 (sqlite 或 postgres，默认为 sqlite)
-  -database string
-        数据库地址（默认为 "./db.sqlite"）
-        如果数据库为 postgres ，则此值应 PostgreSQL DSN 格式
-        例如 "host=127.0.0.1 port=5432 user=postgres dbname=openai_api_route sslmode=disable password=woshimima"
-  -list
-        列出所有上游
-  -noauth
-        不检查传入的授权头
-```
-
-以下是一个 `./upstreams.yaml` 文件配置示例
-
-```yaml
-authorization: woshimima
-
-# 使用 sqlite 作为数据库储存请求记录
-dbtype: sqlite
-dbaddr: ./db.sqlite
-
-# 使用 postgres 作为数据库储存请求记录
-# dbtype: postgres
-# dbaddr: "host=127.0.0.1 port=5432 user=postgres dbname=openai_api_route sslmode=disable password=woshimima"
-
-upstreams:
-  - sk: "secret_key_1"
-    endpoint: "https://api.openai.com/v2"
-  - sk: "secret_key_2"
-    endpoint: "https://api.openai.com/v1"
-    timeout: 30
-```
-
-请注意，程序会根据情况修改 timeout 的值
-
-您可以直接运行 `./openai-api-route` 命令，如果数据库不存在，系统会自动创建。
+如果你混合使用 OpenAI 和 Replicate 平台的模型，你可能需要分别为 OpenAI 和 Replicate 上游设置他们各自的允许列表，否则用户请求 OpenAI 的模型时可能会发送到 Replicate 平台
 
 ## 超时策略
 
@@ -127,8 +183,4 @@ upstreams:
 
 1. **默认超时时间**：如果没有特殊条件，服务将使用默认的超时时间，即 60 秒。
 
-2. **流式请求**：如果请求体被识别为流式（`requestBody.Stream` 为 `true`），并且请求体检查（`requestBodyOK`）没有发现问题，超时时间将被设置为 5 秒。这适用于那些预期会快速响应的流式请求。
-
-3. **大请求体**：如果请求体的大小超过 128KB（即 `len(inBody) > 1024*128`），超时时间将被设置为 20 秒。这考虑到了处理大型数据可能需要更长的时间。
-
-4. **上游超时配置**：如果上游服务器在配置中指定了超时时间（`upstream.Timeout` 大于 0），服务将使用该值作为超时时间。这个值是以秒为单位的。
+2. **流式请求**：如果请求体被识别为流式（`requestBody.Stream` 为 `true`），并且请求体检查（`requestBodyOK`）没有发现问题，超时时间将被设置为 5 秒。这适用于那些预期会快速响应的流式请求。

auth.go

@@ -2,32 +2,14 @@ package main
 
 import (
 	"errors"
-	"log"
 	"strings"
-
-	"github.com/gin-gonic/gin"
 )
 
-func handleAuth(c *gin.Context) error {
-	var err error
-
-	authorization := c.Request.Header.Get("Authorization")
-	if !strings.HasPrefix(authorization, "Bearer") {
-		err = errors.New("authorization header should start with 'Bearer'")
-		c.AbortWithError(403, err)
-		return err
-	}
-
-	authorization = strings.Trim(authorization[len("Bearer"):], " ")
-	log.Println("Received authorization", authorization)
-
-	for _, auth := range strings.Split(config.Authorization, ",") {
-		if authorization != strings.Trim(auth, " ") {
-			err = errors.New("wrong authorization header")
-			c.AbortWithError(403, err)
-			return err
+func checkAuth(authorization string, config string) error {
+	for _, auth := range strings.Split(config, ",") {
+		if authorization == strings.Trim(auth, " ") {
+			return nil
 		}
 	}
-
-	return nil
+	return errors.New("wrong authorization header")
 }

config.sample.yaml

@@ -11,6 +11,13 @@ dbaddr: ./db.sqlite
 upstreams:
   - sk: "secret_key_1"
     endpoint: "https://api.openai.com/v2"
+    allow: ["gpt-3.5-trubo"] # 可选的模型白名单
   - sk: "secret_key_2"
     endpoint: "https://api.openai.com/v1"
-    timeout: 30
+    timeout: 30
+    allow: ["gpt-3.5-trubo"] # 可选的模型白名单
+    deny: ["gpt-4"] # 可选的模型黑名单
+    # 若白名单和黑名单同时设置，先判断白名单，再判断黑名单
+  - sk: "key_for_replicate"
+    type: replicate
+    allow: ["mistralai/mixtral-8x7b-instruct-v0.1"]

cors.go

@@ -0,0 +1,20 @@
+package main
+
+import (
+	"log"
+
+	"github.com/gin-gonic/gin"
+)
+
+func sendCORSHeaders(c *gin.Context) {
+	log.Println("sendCORSHeaders")
+	if c.Writer.Header().Get("Access-Control-Allow-Origin") == "" {
+		c.Header("Access-Control-Allow-Origin", "*")
+	}
+	if c.Writer.Header().Get("Access-Control-Allow-Methods") == "" {
+		c.Header("Access-Control-Allow-Methods", "POST, GET, OPTIONS, PUT, DELETE, PATCH")
+	}
+	if c.Writer.Header().Get("Access-Control-Allow-Headers") == "" {
+		c.Header("Access-Control-Allow-Headers", "Origin, Authorization, Content-Type")
+	}
+}

main.go

@@ -1,15 +1,21 @@
 package main
 
 import (
+	"bytes"
+	"encoding/json"
 	"flag"
 	"fmt"
+	"io"
 	"log"
+	"net/http"
+	"os"
 	"strings"
 	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/penglongli/gin-metrics/ginmetrics"
 	"gorm.io/driver/postgres"
+
 	"gorm.io/driver/sqlite"
 	"gorm.io/gorm"
 )
@@ -23,11 +29,11 @@ func main() {
 	noauth := flag.Bool("noauth", false, "Do not check incoming authorization header")
 	flag.Parse()
 
-	log.Println("Service starting")
+	log.Println("[main]: Service starting")
 
 	// load all upstreams
 	config = readConfig(*configFile)
-	log.Println("Load upstreams number:", len(config.Upstreams))
+	log.Println("[main]: Load upstreams number:", len(config.Upstreams))
 
 	// connect to database
 	var db *gorm.DB
@@ -38,17 +44,23 @@ func main() {
 			PrepareStmt:            true,
 			SkipDefaultTransaction: true,
 		})
+		if err != nil {
+			log.Fatalf("[main]: Error to connect sqlite database: %s", err)
+		}
 	case "postgres":
 		db, err = gorm.Open(postgres.Open(config.DBAddr), &gorm.Config{
 			PrepareStmt:            true,
 			SkipDefaultTransaction: true,
 		})
+		if err != nil {
+			log.Fatalf("[main]: Error to connect postgres database: %s", err)
+		}
 	default:
-		log.Fatalf("Unsupported database type: '%s'", config.DBType)
+		log.Fatalf("[main]: Unsupported database type: '%s'", config.DBType)
 	}
 
 	db.AutoMigrate(&Record{})
-	log.Println("Auto migrate database done")
+	log.Println("[main]: Auto migrate database done")
 
 	if *listMode {
 		fmt.Println("SK\tEndpoint")
@@ -66,6 +78,9 @@ func main() {
 	m.SetMetricPath("/v1/metrics")
 	m.Use(engine)
 
+	// CORS middleware
+	// engine.Use(corsMiddleware())
+
 	// error handle middleware
 	engine.Use(func(c *gin.Context) {
 		c.Next()
@@ -74,68 +89,154 @@ func main() {
 		}
 		errText := strings.Join(c.Errors.Errors(), "\n")
 		c.JSON(-1, gin.H{
-			"error": errText,
+			"openai-api-route error": errText,
 		})
 	})
 
 	// CORS handler
 	engine.OPTIONS("/v1/*any", func(ctx *gin.Context) {
-		header := ctx.Writer.Header()
-		header.Set("Access-Control-Allow-Origin", "*")
-		header.Set("Access-Control-Allow-Methods", "POST, GET, OPTIONS, PUT, DELETE, PATCH")
-		header.Set("Access-Control-Allow-Headers", "Origin, Authorization, Content-Type")
+		// set cros header
+		ctx.Header("Access-Control-Allow-Origin", "*")
+		ctx.Header("Access-Control-Allow-Methods", "POST, GET, OPTIONS, PUT, DELETE, PATCH")
+		ctx.Header("Access-Control-Allow-Headers", "Origin, Authorization, Content-Type")
 		ctx.AbortWithStatus(200)
 	})
 
 	engine.POST("/v1/*any", func(c *gin.Context) {
+		var err error
+		hostname, _ := os.Hostname()
+		if config.Hostname != "" {
+			hostname = config.Hostname
+		}
 		record := Record{
 			IP:            c.ClientIP(),
+			Hostname:      hostname,
 			CreatedAt:     time.Now(),
 			Authorization: c.Request.Header.Get("Authorization"),
 			UserAgent:     c.Request.Header.Get("User-Agent"),
-		}
-		defer func() {
-			if err := recover(); err != nil {
-				log.Println("Error:", err)
-				c.AbortWithError(500, fmt.Errorf("%s", err))
-			}
-		}()
-
-		// check authorization header
-		if !*noauth {
-			if handleAuth(c) != nil {
-				return
-			}
+			Model:         c.Request.URL.Path,
 		}
 
-		for index, upstream := range config.Upstreams {
-			if upstream.Endpoint == "" || upstream.SK == "" {
-				c.AbortWithError(500, fmt.Errorf("invaild upstream '%s' '%s'", upstream.SK, upstream.Endpoint))
+		authorization := c.Request.Header.Get("Authorization")
+		if strings.HasPrefix(authorization, "Bearer") {
+			authorization = strings.Trim(authorization[len("Bearer"):], " ")
+		} else {
+			authorization = strings.Trim(authorization, " ")
+			log.Println("[auth] Warning: authorization header should start with 'Bearer'")
+		}
+		log.Println("Received authorization '" + authorization + "'")
+
+		availUpstreams := make([]OPENAI_UPSTREAM, 0)
+		for _, upstream := range config.Upstreams {
+			if upstream.SK == "" {
+				sendCORSHeaders(c)
+				c.AbortWithError(500, fmt.Errorf("[processRequest.begin]: invaild SK (secret key) %s", upstream.SK))
 				continue
 			}
 
+			if !*noauth && !upstream.Noauth {
+				if checkAuth(authorization, upstream.Authorization) != nil {
+					continue
+				}
+			}
+
+			availUpstreams = append(availUpstreams, upstream)
+		}
+
+		if len(availUpstreams) == 0 {
+			sendCORSHeaders(c)
+			c.AbortWithError(500, fmt.Errorf("[processRequest.begin]: no available upstream for your token"))
+		}
+		log.Println("[processRequest.begin]: availUpstreams", len(availUpstreams))
+
+		bufIO := bytes.NewBuffer(make([]byte, 0, 1024))
+		wrapedBody := false
+
+		for index, _upstream := range availUpstreams {
+
+			// copy
+			upstream := _upstream
+			record.UpstreamEndpoint = upstream.Endpoint
+			record.UpstreamSK = upstream.SK
+
 			shouldResponse := index == len(config.Upstreams)-1
 
-			if len(config.Upstreams) == 1 {
+			if len(availUpstreams) == 1 {
+				// [todo] copy problem
 				upstream.Timeout = 120
 			}
 
-			err = processRequest(c, &upstream, &record, shouldResponse)
-			if err != nil {
-				log.Println("Error from upstream", upstream.Endpoint, "should retry", err)
-				continue
+			// buffer for incoming request
+			if !wrapedBody {
+				log.Println("[processRequest.begin]: wrap request body")
+				c.Request.Body = io.NopCloser(io.TeeReader(c.Request.Body, bufIO))
+				wrapedBody = true
+			} else {
+				log.Println("[processRequest.begin]: reuse request body")
+				c.Request.Body = io.NopCloser(bytes.NewReader(bufIO.Bytes()))
 			}
 
-			break
+			if upstream.Type == "replicate" {
+				err = processReplicateRequest(c, &upstream, &record, shouldResponse)
+			} else if upstream.Type == "openai" {
+				err = processRequest(c, &upstream, &record, shouldResponse)
+			} else {
+				err = fmt.Errorf("[processRequest.begin]: unsupported upstream type '%s'", upstream.Type)
+			}
+
+			if err == nil {
+				log.Println("[processRequest.done]: Success from upstream", upstream.Endpoint)
+				break
+			}
+
+			if err == http.ErrAbortHandler {
+				abortErr := "[processRequest.done]: AbortHandler, client's connection lost?, no upstream will try, stop here"
+				log.Println(abortErr)
+				record.Response += abortErr
+				record.Status = 500
+				break
+			}
+			log.Println("[processRequest.done]: Error from upstream", upstream.Endpoint, "should retry", err, "should response:", shouldResponse)
+
+			// error process, break
+			if shouldResponse {
+				c.Header("Content-Type", "application/json")
+				sendCORSHeaders(c)
+				c.AbortWithError(500, err)
+			}
 		}
 
-		log.Println("Record result:", record.Status, record.Response)
-		record.ElapsedTime = time.Now().Sub(record.CreatedAt)
-		if db.Create(&record).Error != nil {
-			log.Println("Error to save record:", record)
+		// parse and record request body
+		requestBodyBytes := bufIO.Bytes()
+		if len(requestBodyBytes) < 1024*1024 && (strings.HasPrefix(c.Request.Header.Get("Content-Type"), "application/json") ||
+			strings.HasPrefix(c.Request.Header.Get("Content-Type"), "text/")) {
+			record.Body = string(requestBodyBytes)
 		}
+		requestBody, err := ParseRequestBody(requestBodyBytes)
+		if err != nil {
+			log.Println("[processRequest.done]: Error to parse request body:", err)
+		} else {
+			record.Model = requestBody.Model
+		}
+
+		log.Println("[final]: Record result:", record.Status, record.Response)
+		record.ElapsedTime = time.Since(record.CreatedAt)
+
+		// async record request
+		go func() {
+			// encoder headers to record.Headers in json string
+			headers, _ := json.Marshal(c.Request.Header)
+			record.Headers = string(headers)
+
+			// turncate request if too long
+			log.Println("[async.record]: body length:", len(record.Body))
+			if db.Create(&record).Error != nil {
+				log.Println("[async.record]: Error to save record:", record)
+			}
+		}()
+
 		if record.Status != 200 {
-			errMessage := fmt.Sprintf("IP: %s request %s error %d with %s", record.IP, record.Model, record.Status, record.Response)
+			errMessage := fmt.Sprintf("[result.error]: IP: %s request %s error %d with %s", record.IP, record.Model, record.Status, record.Response)
 			go sendFeishuMessage(errMessage)
 			go sendMatrixMessage(errMessage)
 		}

process.go

@@ -8,216 +8,118 @@ import (
 	"io"
 	"log"
 	"net/http"
-	"net/http/httputil"
 	"net/url"
 	"strings"
 	"time"
 
 	"github.com/gin-gonic/gin"
-	"golang.org/x/net/context"
 )
 
 func processRequest(c *gin.Context, upstream *OPENAI_UPSTREAM, record *Record, shouldResponse bool) error {
-	var errCtx error
-
-	record.UpstreamEndpoint = upstream.Endpoint
-	record.UpstreamSK = upstream.SK
-	record.Response = ""
-	// [TODO] record request body
 
 	// reverse proxy
 	remote, err := url.Parse(upstream.Endpoint)
 	if err != nil {
-		c.AbortWithError(500, errors.New("can't parse reverse proxy remote URL"))
 		return err
 	}
 
-	haveResponse := false
+	path := strings.TrimPrefix(c.Request.URL.Path, "/v1")
+	remote.Path = upstream.URL.Path + path
+	log.Println("[proxy.begin]:", remote)
+	log.Println("[proxy.begin]: shouldResposne:", shouldResponse)
 
-	proxy := httputil.NewSingleHostReverseProxy(remote)
-	proxy.Director = nil
-	var inBody []byte
-	proxy.Rewrite = func(proxyRequest *httputil.ProxyRequest) {
-		in := proxyRequest.In
+	client := &http.Client{}
+	request := &http.Request{}
+	request.ContentLength = c.Request.ContentLength
+	request.Method = c.Request.Method
+	request.URL = remote
 
-		ctx, cancel := context.WithCancel(context.Background())
-		proxyRequest.Out = proxyRequest.Out.WithContext(ctx)
-
-		out := proxyRequest.Out
-
-		// read request body
-		inBody, err = io.ReadAll(in.Body)
-		if err != nil {
-			c.AbortWithError(502, errors.New("reverse proxy middleware failed to read request body "+err.Error()))
-			return
-		}
-
-		// record chat message from user
-		record.Body = string(inBody)
-		requestBody, requestBodyOK := ParseRequestBody(inBody)
-		// record if parse success
-		if requestBodyOK == nil {
-			record.Model = requestBody.Model
-		}
-
-		// set timeout, default is 60 second
-		timeout := 60 * time.Second
-		if requestBodyOK == nil && requestBody.Stream {
-			timeout = 5 * time.Second
-		}
-		if len(inBody) > 1024*128 {
-			timeout = 20 * time.Second
-		}
-		if upstream.Timeout > 0 {
-			// convert upstream.Timeout(second) to nanosecond
-			timeout = time.Duration(upstream.Timeout) * time.Second
-		}
-
-		// timeout out request
-		go func() {
-			time.Sleep(timeout)
-			if !haveResponse {
-				log.Println("Timeout upstream", upstream.Endpoint)
-				errCtx = errors.New("timeout")
-				if shouldResponse {
-					c.AbortWithError(502, errCtx)
-				}
-				cancel()
-			}
-		}()
-
-		out.Body = io.NopCloser(bytes.NewReader(inBody))
-
-		out.Host = remote.Host
-		out.URL.Scheme = remote.Scheme
-		out.URL.Host = remote.Host
-		out.URL.Path = in.URL.Path
-		out.Header = http.Header{}
-		out.Header.Set("Host", remote.Host)
-		if upstream.SK == "asis" {
-			out.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
-		} else {
-			out.Header.Set("Authorization", "Bearer "+upstream.SK)
-		}
-		out.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
-	}
-	var buf bytes.Buffer
-	var contentType string
-	proxy.ModifyResponse = func(r *http.Response) error {
-		haveResponse = true
-		record.Status = r.StatusCode
-		if !shouldResponse && r.StatusCode != 200 {
-			log.Println("upstream return not 200 and should not response", r.StatusCode)
-			return errors.New("upstream return not 200 and should not response")
-		}
-		r.Header.Del("Access-Control-Allow-Origin")
-		r.Header.Del("Access-Control-Allow-Methods")
-		r.Header.Del("Access-Control-Allow-Headers")
-		r.Header.Set("Access-Control-Allow-Origin", "*")
-		r.Header.Set("Access-Control-Allow-Methods", "POST, GET, OPTIONS, PUT, DELETE, PATCH")
-		r.Header.Set("Access-Control-Allow-Headers", "Origin, Authorization, Content-Type")
-
-		if r.StatusCode != 200 {
-			body, err := io.ReadAll(r.Body)
-			if err != nil {
-				record.Response = "failed to read response from upstream " + err.Error()
-				return errors.New(record.Response)
-			}
-			record.Response = fmt.Sprintf("openai-api-route upstream return '%s' with '%s'", r.Status, string(body))
-			record.Status = r.StatusCode
-			return fmt.Errorf(record.Response)
-		}
-		// count success
-		r.Body = io.NopCloser(io.TeeReader(r.Body, &buf))
-		contentType = r.Header.Get("content-type")
-		return nil
-	}
-	proxy.ErrorHandler = func(w http.ResponseWriter, r *http.Request, err error) {
-		haveResponse = true
-		record.ResponseTime = time.Now().Sub(record.CreatedAt)
-		log.Println("Error", err, upstream.SK, upstream.Endpoint)
-
-		errCtx = err
-
-		// abort to error handle
-		if shouldResponse {
-			c.AbortWithError(502, err)
-		}
-
-		log.Println("response is", r.Response)
-
-		if record.Status == 0 {
-			record.Status = 502
-		}
-		if record.Response == "" {
-			record.Response = err.Error()
-		}
-		if r.Response != nil {
-			record.Status = r.Response.StatusCode
-		}
-
-	}
-
-	proxy.ServeHTTP(c.Writer, c.Request)
-
-	// return context error
-	if errCtx != nil {
-		// fix inrequest body
-		c.Request.Body = io.NopCloser(bytes.NewReader(inBody))
-		return errCtx
-	}
-
-	resp, err := io.ReadAll(io.NopCloser(&buf))
-	if err != nil {
-		record.Response = "failed to read response from upstream " + err.Error()
-		log.Println(record.Response)
+	// process header
+	if upstream.KeepHeader {
+		request.Header = c.Request.Header
 	} else {
-
-		// record response
-		// stream mode
-		if strings.HasPrefix(contentType, "text/event-stream") {
-			for _, line := range strings.Split(string(resp), "\n") {
-				chunk := StreamModeChunk{}
-				line = strings.TrimPrefix(line, "data:")
-				line = strings.TrimSpace(line)
-				if line == "" {
-					continue
-				}
-
-				err := json.Unmarshal([]byte(line), &chunk)
-				if err != nil {
-					log.Println(err)
-					continue
-				}
-
-				if len(chunk.Choices) == 0 {
-					continue
-				}
-				record.Response += chunk.Choices[0].Delta.Content
-			}
-		} else if strings.HasPrefix(contentType, "application/json") {
-			var fetchResp FetchModeResponse
-			err := json.Unmarshal(resp, &fetchResp)
-			if err != nil {
-				log.Println("Error parsing fetch response:", err)
-				return nil
-			}
-			if !strings.HasPrefix(fetchResp.Model, "gpt-") {
-				log.Println("Not GPT model, skip recording response:", fetchResp.Model)
-				return nil
-			}
-			if len(fetchResp.Choices) == 0 {
-				log.Println("Error: fetch response choice length is 0")
-				return nil
-			}
-			record.Response = fetchResp.Choices[0].Message.Content
-		} else {
-			log.Println("Unknown content type", contentType)
-		}
+		request.Header = http.Header{}
 	}
 
-	if len(record.Body) > 1024*512 {
-		record.Body = ""
+	// process header authorization
+	if upstream.SK == "asis" {
+		request.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
+	} else {
+		request.Header.Set("Authorization", "Bearer "+upstream.SK)
+	}
+	request.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
+	request.Header.Set("Host", remote.Host)
+	request.Header.Set("Content-Length", c.Request.Header.Get("Content-Length"))
+
+	request.Body = c.Request.Body
+
+	resp, err := client.Do(request)
+	if err != nil {
+		body := []byte{}
+		if resp != nil && resp.Body != nil {
+			body, _ = io.ReadAll(resp.Body)
+		}
+		return errors.New(err.Error() + " " + string(body))
+	}
+
+	defer resp.Body.Close()
+
+	record.Status = resp.StatusCode
+
+	if resp.StatusCode != 200 {
+		body, _ := io.ReadAll(resp.Body)
+		record.Status = resp.StatusCode
+		errRet := fmt.Errorf("[error]: openai-api-route upstream return '%s' with '%s'", resp.Status, string(body))
+		log.Println(errRet)
+		return errRet
+	}
+
+	// copy response header
+	for k, v := range resp.Header {
+		c.Header(k, v[0])
+	}
+	sendCORSHeaders(c)
+
+	respBodyBuffer := bytes.NewBuffer(make([]byte, 0, 4*1024))
+	respBodyTeeReader := io.TeeReader(resp.Body, respBodyBuffer)
+	record.ResponseTime = time.Since(record.CreatedAt)
+	io.Copy(c.Writer, respBodyTeeReader)
+	record.ElapsedTime = time.Since(record.CreatedAt)
+
+	// parse and record response
+	if strings.HasPrefix(resp.Header.Get("Content-Type"), "application/json") {
+		var fetchResp FetchModeResponse
+		err := json.NewDecoder(respBodyBuffer).Decode(&fetchResp)
+		if err == nil {
+			if len(fetchResp.Choices) > 0 {
+				record.Response = fetchResp.Choices[0].Message.Content
+			}
+		}
+	} else if strings.HasPrefix(resp.Header.Get("Content-Type"), "text/event-stream") {
+		lines := bytes.Split(respBodyBuffer.Bytes(), []byte("\n"))
+		for _, line := range lines {
+			line = bytes.TrimSpace(line)
+			line = bytes.TrimPrefix(line, []byte("data:"))
+			line = bytes.TrimSpace(line)
+			if len(line) == 0 {
+				continue
+			}
+			chunk := StreamModeChunk{}
+			err = json.Unmarshal(line, &chunk)
+			if err != nil {
+				log.Println("[proxy.parseChunkError]:", err)
+				break
+			}
+			if len(chunk.Choices) == 0 {
+				continue
+			}
+			record.Response += chunk.Choices[0].Delta.Content
+		}
+	} else if strings.HasPrefix(resp.Header.Get("Content-Type"), "text") {
+		body, _ := io.ReadAll(respBodyBuffer)
+		record.Response = string(body)
+	} else {
+		log.Println("[proxy.record]: Unknown content type", resp.Header.Get("Content-Type"))
 	}
 
 	return nil

record.go

@@ -6,6 +6,7 @@ import (
 
 type Record struct {
 	ID               int64 `gorm:"primaryKey,autoIncrement"`
+	Hostname         string
 	UpstreamEndpoint string
 	UpstreamSK       string
 	CreatedAt        time.Time
@@ -18,6 +19,7 @@ type Record struct {
 	Status           int
 	Authorization    string // the autorization header send by client
 	UserAgent        string
+	Headers          string
 }
 
 type StreamModeChunk struct {

recovery.go

@@ -0,0 +1,24 @@
+package main
+
+import (
+	"errors"
+	"log"
+	"net/http"
+	"net/http/httputil"
+
+	"github.com/gin-gonic/gin"
+)
+
+func ServeHTTP(proxy *httputil.ReverseProxy, w gin.ResponseWriter, r *http.Request) (errReturn error) {
+
+	// recovery
+	defer func() {
+		if err := recover(); err != nil {
+			log.Println("[serve.panic]: ", err)
+			errReturn = errors.New("[serve.panic]: Panic recover in reverse proxy serve HTTP")
+		}
+	}()
+
+	proxy.ServeHTTP(w, r)
+	return nil
+}

replicate.go

@@ -0,0 +1,369 @@
+package main
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"log"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/gin-gonic/gin"
+)
+
+var replicate_model_url_template = "https://api.replicate.com/v1/models/%s/predictions"
+
+func processReplicateRequest(c *gin.Context, upstream *OPENAI_UPSTREAM, record *Record, shouldResponse bool) error {
+	err := _processReplicateRequest(c, upstream, record, shouldResponse)
+	if shouldResponse {
+		sendCORSHeaders(c)
+		if err != nil {
+			c.AbortWithError(502, err)
+		}
+	}
+	return err
+}
+
+func _processReplicateRequest(c *gin.Context, upstream *OPENAI_UPSTREAM, record *Record, shouldResponse bool) error {
+	// read request body
+	inBody, err := io.ReadAll(c.Request.Body)
+	if err != nil {
+		return errors.New("[processReplicateRequest]: failed to read request body " + err.Error())
+	}
+
+	// record request body
+
+	// parse request body
+	inRequest := &OpenAIChatRequest{}
+	err = json.Unmarshal(inBody, inRequest)
+	if err != nil {
+		c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+		return errors.New("[processReplicateRequest]: failed to parse request body " + err.Error())
+	}
+
+	record.Model = inRequest.Model
+
+	// check allow model
+	if len(upstream.Allow) > 0 {
+		isAllow := false
+		for _, model := range upstream.Allow {
+			if model == inRequest.Model {
+				isAllow = true
+				break
+			}
+		}
+		if !isAllow {
+			c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+			return errors.New("[processReplicateRequest]: model not allow")
+		}
+	}
+	// check block model
+	if len(upstream.Deny) > 0 {
+		for _, model := range upstream.Deny {
+			if model == inRequest.Model {
+				c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+				return errors.New("[processReplicateRequest]: model deny")
+			}
+		}
+	}
+
+	// set url
+	model_url := fmt.Sprintf(replicate_model_url_template, inRequest.Model)
+	log.Println("[processReplicateRequest]: model_url:", model_url)
+
+	// create request with default value
+	outRequest := &ReplicateModelRequest{
+		Stream: false,
+		Input: ReplicateModelRequestInput{
+			Prompt:           "",
+			MaxNewTokens:     1024,
+			Temperature:      0.6,
+			Top_p:            0.9,
+			Top_k:            50,
+			FrequencyPenalty: 0.0,
+			PresencePenalty:  0.0,
+			PromptTemplate:   "{prompt}",
+		},
+	}
+
+	// copy value from input request
+	outRequest.Stream = inRequest.Stream
+	outRequest.Input.Temperature = inRequest.Temperature
+	outRequest.Input.FrequencyPenalty = inRequest.FrequencyPenalty
+	outRequest.Input.PresencePenalty = inRequest.PresencePenalty
+
+	// render prompt
+	systemMessage := ""
+	userMessage := ""
+	assistantMessage := ""
+	for _, message := range inRequest.Messages {
+		if message.Role == "system" {
+			if systemMessage != "" {
+				systemMessage += "\n"
+			}
+			systemMessage += message.Content
+			continue
+		}
+		if message.Role == "user" {
+			if userMessage != "" {
+				userMessage += "\n"
+			}
+			userMessage += message.Content
+			if systemMessage != "" {
+				userMessage = systemMessage + "\n" + userMessage
+				systemMessage = ""
+			}
+			continue
+		}
+		if message.Role == "assistant" {
+			if assistantMessage != "" {
+				assistantMessage += "\n"
+			}
+			assistantMessage += message.Content
+
+			if outRequest.Input.Prompt != "" {
+				outRequest.Input.Prompt += "\n"
+			}
+			if userMessage != "" {
+				outRequest.Input.Prompt += fmt.Sprintf("<s> [INST] %s [/INST] %s </s>", userMessage, assistantMessage)
+			} else {
+				outRequest.Input.Prompt += fmt.Sprintf("<s> %s </s>", assistantMessage)
+			}
+			userMessage = ""
+			assistantMessage = ""
+		}
+		// unknown role
+		log.Println("[processReplicateRequest]: Warning: unknown role", message.Role)
+	}
+	// final user message
+	if userMessage != "" {
+		outRequest.Input.Prompt += fmt.Sprintf("<s> [INST] %s [/INST] ", userMessage)
+		userMessage = ""
+	}
+	// final assistant message
+	if assistantMessage != "" {
+		outRequest.Input.Prompt += fmt.Sprintf("<s> %s </s>", assistantMessage)
+	}
+	log.Println("[processReplicateRequest]: outRequest.Input.Prompt:", outRequest.Input.Prompt)
+
+	// send request
+	outBody, err := json.Marshal(outRequest)
+	if err != nil {
+		c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+		return errors.New("[processReplicateRequest]: failed to marshal request body " + err.Error())
+	}
+
+	// http add headers
+	req, err := http.NewRequest("POST", model_url, bytes.NewBuffer(outBody))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Token "+upstream.SK)
+	// send
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+		return errors.New("[processReplicateRequest]: failed to post request " + err.Error())
+	}
+
+	// read response body
+	outBody, err = io.ReadAll(resp.Body)
+	if err != nil {
+		c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+		return errors.New("[processReplicateRequest]: failed to read response body " + err.Error())
+	}
+
+	// parse reponse body
+	outResponse := &ReplicateModelResponse{}
+	err = json.Unmarshal(outBody, outResponse)
+	if err != nil {
+		c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+		return errors.New("[processReplicateRequest]: failed to parse response body " + err.Error())
+	}
+
+	if outResponse.Stream {
+		// get result
+		log.Println("[processReplicateRequest]: outResponse.URLS.Get:", outResponse.URLS.Stream)
+		req, err := http.NewRequest("GET", outResponse.URLS.Stream, nil)
+		if err != nil {
+			c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+			return errors.New("[processReplicateRequest]: failed to create get request " + err.Error())
+		}
+		req.Header.Set("Authorization", "Token "+upstream.SK)
+		req.Header.Set("Accept", "text/event-stream")
+		// send
+		resp, err := http.DefaultClient.Do(req)
+		if err != nil {
+			c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+			return errors.New("[processReplicateRequest]: failed to get request " + err.Error())
+		}
+
+		// get result by chunk
+		var buffer string = ""
+		var indexCount int64 = 0
+		for {
+			if !strings.Contains(buffer, "\n\n") {
+				// receive chunk
+				chunk := make([]byte, 1024)
+				length, err := resp.Body.Read(chunk)
+				if err == io.EOF {
+					break
+				}
+				if length == 0 {
+					break
+				}
+				if err != nil {
+					c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+					return errors.New("[processReplicateRequest]: failed to read response body " + err.Error())
+				}
+				// add chunk to buffer
+				chunk = bytes.Trim(chunk, "\x00")
+				buffer += string(chunk)
+				continue
+			}
+
+			// cut the first chunk by find index
+			index := strings.Index(buffer, "\n\n")
+			chunk := buffer[:index]
+			buffer = buffer[index+2:]
+
+			// trim line
+			chunk = strings.Trim(chunk, "\n")
+
+			// ignore hi
+			if !strings.Contains(chunk, "\n") {
+				continue
+			}
+
+			// parse chunk to ReplicateModelResultChunk object
+			chunkObj := &ReplicateModelResultChunk{}
+			lines := strings.Split(chunk, "\n")
+			// first line is event
+			chunkObj.Event = strings.TrimSpace(lines[0])
+			chunkObj.Event = strings.TrimPrefix(chunkObj.Event, "event: ")
+			// second line is id
+			chunkObj.ID = strings.TrimSpace(lines[1])
+			chunkObj.ID = strings.TrimPrefix(chunkObj.ID, "id: ")
+			chunkObj.ID = strings.SplitN(chunkObj.ID, ":", 2)[0]
+			// third line is data
+			chunkObj.Data = lines[2]
+			chunkObj.Data = strings.TrimPrefix(chunkObj.Data, "data: ")
+
+			record.Response += chunkObj.Data
+
+			// done
+			if chunkObj.Event == "done" {
+				break
+			}
+
+			sendCORSHeaders(c)
+
+			// create OpenAI response chunk
+			c.SSEvent("", &OpenAIChatResponseChunk{
+				ID:    "",
+				Model: outResponse.Model,
+				Choices: []OpenAIChatResponseChunkChoice{
+					{
+						Index: indexCount,
+						Delta: OpenAIChatMessage{
+							Role:    "assistant",
+							Content: chunkObj.Data,
+						},
+					},
+				},
+			})
+			c.Writer.Flush()
+			indexCount += 1
+		}
+		sendCORSHeaders(c)
+		c.SSEvent("", &OpenAIChatResponseChunk{
+			ID:    "",
+			Model: outResponse.Model,
+			Choices: []OpenAIChatResponseChunkChoice{
+				{
+					Index: indexCount,
+					Delta: OpenAIChatMessage{
+						Role:    "assistant",
+						Content: "",
+					},
+					FinishReason: "stop",
+				},
+			},
+		})
+		c.Writer.Flush()
+		indexCount += 1
+		record.Status = 200
+		return nil
+
+	} else {
+		var result *ReplicateModelResultGet
+
+		for {
+			// get result
+			log.Println("[processReplicateRequest]: outResponse.URLS.Get:", outResponse.URLS.Get)
+			req, err := http.NewRequest("GET", outResponse.URLS.Get, nil)
+			if err != nil {
+				c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+				return errors.New("[processReplicateRequest]: failed to create get request " + err.Error())
+			}
+			req.Header.Set("Authorization", "Token "+upstream.SK)
+			// send
+			resp, err := http.DefaultClient.Do(req)
+			if err != nil {
+				c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+				return errors.New("[processReplicateRequest]: failed to get request " + err.Error())
+			}
+			// get result
+			resultBody, err := io.ReadAll(resp.Body)
+			if err != nil {
+				c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+				return errors.New("[processReplicateRequest]: failed to read response body " + err.Error())
+			}
+
+			// parse reponse body
+			result = &ReplicateModelResultGet{}
+			err = json.Unmarshal(resultBody, result)
+			if err != nil {
+				c.Request.Body = io.NopCloser(bytes.NewBuffer(inBody))
+				return errors.New("[processReplicateRequest]: failed to parse response body " + err.Error())
+			}
+
+			if result.Status == "processing" || result.Status == "starting" {
+				time.Sleep(3 * time.Second)
+				continue
+			}
+
+			break
+		}
+
+		// build openai resposne
+		openAIResult := &OpenAIChatResponse{
+			ID:      result.ID,
+			Model:   result.Model,
+			Choices: []OpenAIChatResponseChoice{},
+			Usage: OpenAIChatResponseUsage{
+				TotalTokens:  result.Metrics.InputTokenCount + result.Metrics.OutputTokenCount,
+				PromptTokens: result.Metrics.InputTokenCount,
+			},
+		}
+		openAIResult.Choices = append(openAIResult.Choices, OpenAIChatResponseChoice{
+			Index: 0,
+			Message: OpenAIChatMessage{
+				Role:    "assistant",
+				Content: strings.Join(result.Output, ""),
+			},
+			FinishReason: "stop",
+		})
+
+		record.Response = strings.Join(result.Output, "")
+		record.Status = 200
+
+		// gin return
+		sendCORSHeaders(c)
+		c.JSON(200, openAIResult)
+
+	}
+
+	return nil
+}

structure.go

@@ -2,6 +2,7 @@ package main
 
 import (
 	"log"
+	"net/url"
 	"os"
 
 	"gopkg.in/yaml.v3"
@@ -9,15 +10,26 @@ import (
 
 type Config struct {
 	Address       string            `yaml:"address"`
+	Hostname      string            `yaml:"hostname"`
 	DBType        string            `yaml:"dbtype"`
 	DBAddr        string            `yaml:"dbaddr"`
 	Authorization string            `yaml:"authorization"`
+	Timeout       int64             `yaml:"timeout"`
+	StreamTimeout int64             `yaml:"stream_timeout"`
 	Upstreams     []OPENAI_UPSTREAM `yaml:"upstreams"`
 }
 type OPENAI_UPSTREAM struct {
-	SK       string `yaml:"sk"`
-	Endpoint string `yaml:"endpoint"`
-	Timeout  int64  `yaml:"timeout"`
+	SK            string   `yaml:"sk"`
+	Endpoint      string   `yaml:"endpoint"`
+	Timeout       int64    `yaml:"timeout"`
+	StreamTimeout int64    `yaml:"stream_timeout"`
+	Allow         []string `yaml:"allow"`
+	Deny          []string `yaml:"deny"`
+	Type          string   `yaml:"type"`
+	KeepHeader    bool     `yaml:"keep_header"`
+	Authorization string   `yaml:"authorization"`
+	Noauth        bool     `yaml:"noauth"`
+	URL           *url.URL
 }
 
 func readConfig(filepath string) Config {
@@ -48,6 +60,145 @@ func readConfig(filepath string) Config {
 		log.Println("DBAddr not set, use default value: ./db.sqlite")
 		config.DBAddr = "./db.sqlite"
 	}
+	if config.Timeout == 0 {
+		log.Println("Timeout not set, use default value: 120")
+		config.Timeout = 120
+	}
+	if config.StreamTimeout == 0 {
+		log.Println("StreamTimeout not set, use default value: 10")
+		config.StreamTimeout = 10
+	}
+
+	for i, upstream := range config.Upstreams {
+		// parse upstream endpoint URL
+		endpoint, err := url.Parse(upstream.Endpoint)
+		if err != nil {
+			log.Fatalf("Can't parse upstream endpoint URL '%s': %s", upstream.Endpoint, err)
+		}
+		config.Upstreams[i].URL = endpoint
+		if config.Upstreams[i].Type == "" {
+			config.Upstreams[i].Type = "openai"
+		}
+		if (config.Upstreams[i].Type != "openai") && (config.Upstreams[i].Type != "replicate") {
+			log.Fatalf("Unsupported upstream type '%s'", config.Upstreams[i].Type)
+		}
+		// apply authorization from global config if not set
+		if config.Upstreams[i].Authorization == "" && !config.Upstreams[i].Noauth {
+			config.Upstreams[i].Authorization = config.Authorization
+		}
+		if config.Upstreams[i].Timeout == 0 {
+			config.Upstreams[i].Timeout = config.Timeout
+		}
+		if config.Upstreams[i].StreamTimeout == 0 {
+			config.Upstreams[i].StreamTimeout = config.StreamTimeout
+		}
+	}
 
 	return config
 }
+
+type OpenAIChatRequest struct {
+	FrequencyPenalty float64 `json:"frequency_penalty"`
+	PresencePenalty  float64 `json:"presence_penalty"`
+	MaxTokens        int64   `json:"max_tokens"`
+	Model            string  `json:"model"`
+	Stream           bool    `json:"stream"`
+	Temperature      float64 `json:"temperature"`
+	Messages         []OpenAIChatRequestMessage
+}
+
+type OpenAIChatRequestMessage struct {
+	Content string `json:"content"`
+	Role    string `json:"role"`
+}
+
+type ReplicateModelRequest struct {
+	Stream bool                       `json:"stream"`
+	Input  ReplicateModelRequestInput `json:"input"`
+}
+
+type ReplicateModelRequestInput struct {
+	Prompt           string  `json:"prompt"`
+	MaxNewTokens     int64   `json:"max_new_tokens"`
+	Temperature      float64 `json:"temperature"`
+	Top_p            float64 `json:"top_p"`
+	Top_k            int64   `json:"top_k"`
+	PresencePenalty  float64 `json:"presence_penalty"`
+	FrequencyPenalty float64 `json:"frequency_penalty"`
+	PromptTemplate   string  `json:"prompt_template"`
+}
+
+type ReplicateModelResponse struct {
+	Model   string                     `json:"model"`
+	Version string                     `json:"version"`
+	Stream  bool                       `json:"stream"`
+	Error   string                     `json:"error"`
+	URLS    ReplicateModelResponseURLS `json:"urls"`
+}
+
+type ReplicateModelResponseURLS struct {
+	Cancel string `json:"cancel"`
+	Get    string `json:"get"`
+	Stream string `json:"stream"`
+}
+
+type ReplicateModelResultGet struct {
+	ID      string                      `json:"id"`
+	Model   string                      `json:"model"`
+	Version string                      `json:"version"`
+	Output  []string                    `json:"output"`
+	Error   string                      `json:"error"`
+	Metrics ReplicateModelResultMetrics `json:"metrics"`
+	Status  string                      `json:"status"`
+}
+
+type ReplicateModelResultMetrics struct {
+	InputTokenCount  int64 `json:"input_token_count"`
+	OutputTokenCount int64 `json:"output_token_count"`
+}
+
+type OpenAIChatResponse struct {
+	ID      string                     `json:"id"`
+	Object  string                     `json:"object"`
+	Created int64                      `json:"created"`
+	Model   string                     `json:"model"`
+	Choices []OpenAIChatResponseChoice `json:"choices"`
+	Usage   OpenAIChatResponseUsage    `json:"usage"`
+}
+
+type OpenAIChatResponseUsage struct {
+	PromptTokens     int64 `json:"prompt_tokens"`
+	CompletionTokens int64 `json:"completion_tokens"`
+	TotalTokens      int64 `json:"total_tokens"`
+}
+
+type OpenAIChatResponseChoice struct {
+	Index        int64             `json:"index"`
+	Message      OpenAIChatMessage `json:"message"`
+	FinishReason string            `json:"finish_reason"`
+}
+
+type OpenAIChatMessage struct {
+	Role    string `json:"role"`
+	Content string `json:"content"`
+}
+
+type ReplicateModelResultChunk struct {
+	Event string `json:"event"`
+	ID    string `json:"id"`
+	Data  string `json:"data"`
+}
+
+type OpenAIChatResponseChunk struct {
+	ID      string                          `json:"id"`
+	Object  string                          `json:"object"`
+	Created int64                           `json:"created"`
+	Model   string                          `json:"model"`
+	Choices []OpenAIChatResponseChunkChoice `json:"choices"`
+}
+
+type OpenAIChatResponseChunkChoice struct {
+	Index        int64             `json:"index"`
+	Delta        OpenAIChatMessage `json:"delta"`
+	FinishReason string            `json:"finish_reason"`
+}