记录一下快速模板,整体很简单,如果不接auth,只需要以下:
- 提供一个
/.well-known/ai-plugin.json接口,返回openAI所需要的格式 - 提供openAPI规范的文档
- CORS设置
其他的和普通的web开发类似.
本地开发就直接使用localhost即可,前几天官方localhost无法联通,最近应该修复了.
要让GPT更好理解接口内容,接口需要写详细的文档,在文档内写清楚各个参数作用和可选值以及示例.
Spring Boot
增加对文档的依赖
<dependency> <groupId>org.springdoc</groupId> <artifactId>springdoc-openapi-starter-webmvc-ui</artifactId> <version>2.1.0</version> </dependency>
增加一个bean配置:
@Bean public OpenAPI openAPI() { return new OpenAPI() .info(new Info().title("html fetcher") .description("get content from url") .version("1.0")); }
文档的地址为/v3/api-docs
增加ai-plugin.json接口
@GetMapping(value = "/.well-known/ai-plugin.json", produces = "application/json") public String aiPlugin() { return """ { "schema_version": "v1", "name_for_human": "html fetcher Plugin", "name_for_model": "html_fetcher", "description_for_human": "Plugin for getting content from url", "description_for_model": "Plugin for getting content from url", "auth": { "type": "none" }, "api": { "type": "openapi", "url": "http://localhost:8080/v3/api-docs", "is_user_authenticated": false }, "logo_url": "http://localhost:8080/logo.png", "contact_email": "support@example.com", "legal_info_url": "http://www.example.com/legal" } """; }
logo直接放到resourcesstatic中
内容根据自己插件修改,本地开发直接写localhost,部署写对应的网站地址.
CORS设置
测试的时候允许可以写*, 后续上线更改为openai.com:
@Bean public CorsFilter corsFilter() { UrlBasedCorsConfigurationSource source = new UrlBasedCorsConfigurationSource(); CorsConfiguration config = new CorsConfiguration(); config.addAllowedOrigin("*"); config.addAllowedHeader("*"); config.addAllowedMethod("*"); source.registerCorsConfiguration("/**", config); return new CorsFilter(source); }
要写自定义文档可以用spring-doc的相关注解.
比如写在接口上用@Operation, 字段上使用@Schema注解.
@Schema内可以用allowableValues和example等来约束openai的查询.
fastapi
fastapi自带openAPI集成,只需要把json dump成yaml即可, setup比较简单, 这里直接全部放一起了:
app = FastAPI() app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"], ) # static 文件夹用来做静态资源host 放logo app.mount("/static", StaticFiles(directory="static"), name="static") # json -> yaml @app.get("/openapi.yaml") async def get_openapi(): return Response(content=yaml.dump(app.openapi()), media_type="application/yaml") @app.get("/.well-known/ai-plugin.json") async def openai_api_plugin(): return { "schema_version": "v1", "name_for_human": "", "name_for_model": "", "description_for_human": "", "description_for_model": "", "auth": { "type": "none" }, "api": { "type": "openapi", "url": "http://localhost:8000/openapi.yaml", "is_user_authenticated": False }, "logo_url": "http://localhost:8000/static/logo.png", "contact_email": "support@example.com", "legal_info_url": "http://www.example.com/legal" }
自定义文档内容对于接口的可以用summary,response_description参数, query参数可以用Annotationd, 一个例子:
@app.get("/api/query_profit_data", summary='query profit data by company code, year and quarter', response_description=""" return profit data in format {"key":{"0":"value"}}, panda's dataframe""") async def query_profit_data(code: Annotated[str, Query(description="the company code", example="sh.600000")], year: Annotated[int, Query(description="year to get profit", example=2023)], quarter: Annotated[ int, Query(description="quarter to get profit. allow values:1,2,3,4", example=1)]):
参考资料
chatgpt plugin: https://openai.com/blog/chatgpt-plugins
spring doc: https://springdoc.org/v2/
fastapi: https://fastapi.tiangolo.com/