@luffysolution/omnischolar-pi 0.1.20 → 0.1.22

package/docs/IMAGE_PROVIDERS.en.md

@@ -94,7 +94,9 @@ Vertex AI can use a service-account JSON file:
 
 The service-account file is used locally to obtain an OAuth token and is never written to output artifacts. If `project` is omitted, OmniScholar reads `project_id` from the JSON; `location` defaults to `global`. Model availability still depends on project permissions, location, and model publication status.
 
-Atlas, fal, Vertex, and DashScope/Qwen expose different catalog capabilities. Atlas, fal, Vertex, and DashScope/Qwen have small built-in model sets checked against official documentation; custom services still need an exact `baseUrl`, model ID, capability list, and generation or edit endpoint. Do not infer capabilities from the model name or treat a built-in model as an entitlement guarantee.
+Atlas, fal, Vertex, and DashScope/Qwen expose different catalog capabilities. Atlas, fal, Vertex, and DashScope/Qwen have small built-in model sets checked against official documentation; when `modelCatalogEndpoint` is not configured, custom services make a best-effort request to `baseUrl/models`. If a downstream service has no catalog endpoint, or its models lack explicit capability fields, the tool falls back to the explicit model contracts in `models` and never infers image capabilities from a model name.
+
+The official OpenAI, Google, and Vertex configurations rely on their model catalogs or built-in official catalogs and do not duplicate model pins. The custom example configuration presets `gpt-image-2`, `gpt-image-2.5-sunburst`, `gpt-image-2.5-flare`, `gemini-3.1-flash-image`, `gemini-3.1-flash-lite-image`, and `gemini-3-pro-image` for OpenAI/Gemini-compatible services without a model catalog. These presets are not entitlement guarantees.
 
 ## Generate and edit
 
@@ -109,6 +111,6 @@ Upload only images you may share with the selected service. After files are save
 - DashScope/Qwen supports the Qwen AI Platform native endpoint `https://dashscope.aliyuncs.com/api/v1` and regional Bailian workspace endpoints such as `https://<workspace>.cn-beijing.maas.aliyuncs.com/api/v1`. When `workspace` and `region` are supplied, the tool can derive the regional endpoint.
 - Gemini's `apiKeyEnv` must name an environment variable that exists.
 - Atlas Cloud uses `https://api.atlascloud.ai/api/v1`, submits to `model/generateImage`, and polls `model/prediction/{id}`. The built-in set includes official Nano Banana 2, GPT Image 2, and GPT Image 2.5 Flare/Sunburst IDs. Atlas image tasks are asynchronous and their `outputs` are saved locally.
-- Custom services need a working endpoint and model description.
+- Custom services automatically probe the standard OpenAI-compatible `/models` endpoint, or use `options.modelCatalogEndpoint` for a non-standard catalog URL. If the catalog lacks explicit `capabilities`/`supportedCapabilities`, declare `text-to-image`, `image-to-image`, or `edit` in `models` before calling it.
 
 Review text, structures, mechanisms, scale, and quantitative labels after generation. **AI images are illustrative drafts, not experimental data, real measurements, or scientific conclusions.**

package/docs/IMAGE_PROVIDERS.md

@@ -94,7 +94,9 @@ Vertex AI 可使用 service-account JSON：
 
 服务账号文件只用于本地换取 OAuth token，不会写入输出结果。若省略 `project`，工具会从 JSON 的 `project_id` 自动读取；默认 `location` 为 `global`。模型是否可用仍取决于项目授权、区域和模型发布状态。
 
-Atlas、fal、Vertex、DashScope/Qwen 和自定义服务的模型目录能力不同。Atlas、fal、Vertex、DashScope/Qwen 已提供少量基于官方文档核对的内置模型；自定义服务仍需要准确的 `baseUrl`、模型 ID、功能列表，以及生成或编辑接口。不要只凭模型名填写功能，也不要把内置模型视为账号已开通的保证。
+Atlas、fal、Vertex、DashScope/Qwen 和自定义服务的模型目录能力不同。Atlas、fal、Vertex、DashScope/Qwen 已提供少量基于官方文档核对的内置模型；自定义服务在未配置 `modelCatalogEndpoint` 时会尽力尝试 `baseUrl/models`。如果下游没有模型目录接口，或返回的模型没有明确能力字段，工具会回退到 `models` 中的显式模型合同，不会仅凭模型名推断图片能力。
+
+官方 OpenAI、Google 和 Vertex 配置默认使用模型目录或内置官方目录发现，不需要在 `models` 中重复预设。custom 示例配置预设 `gpt-image-2`、`gpt-image-2.5-sunburst`、`gpt-image-2.5-flare`、`gemini-3.1-flash-image`、`gemini-3.1-flash-lite-image` 和 `gemini-3-pro-image`，供没有模型目录接口的 OpenAI/Gemini 兼容服务使用；这些预设不代表下游账号已经开通对应模型。
 
 ## 生成与编辑
 
@@ -109,6 +111,6 @@ Atlas、fal、Vertex、DashScope/Qwen 和自定义服务的模型目录能力不
 - DashScope/Qwen 支持千问 AI 平台的通用原生地址 `https://dashscope.aliyuncs.com/api/v1`，也支持百炼按地域的 workspace 地址，例如 `https://<workspace>.cn-beijing.maas.aliyuncs.com/api/v1`。当填写 `workspace` 和 `region` 时，工具可以自动拼接地域地址。
 - Gemini 配置中的 `apiKeyEnv` 必须指向已存在的环境变量。
 - Atlas Cloud 使用 `https://api.atlascloud.ai/api/v1`、`model/generateImage` 提交和 `model/prediction/{id}` 轮询；默认内置 Nano Banana 2、GPT Image 2、GPT Image 2.5 Flare/Sunburst 的官方模型 ID。Atlas 的图片任务是异步的，结果中的 `outputs` 会保存到本地。
-- 自定义服务必须给出可用的服务地址与模型说明。
+- 自定义服务会自动探测标准 OpenAI 兼容的 `/models`；也可以用 `options.modelCatalogEndpoint` 指定非标准目录地址。目录结果缺少明确 `capabilities`/`supportedCapabilities` 时只展示为未绑定能力，必须在 `models` 中声明 `text-to-image`、`image-to-image` 或 `edit` 后才能调用。
 
 生成后仍需人工检查文字、结构、机制、比例和定量描述。**AI 图片是示意草稿，不是实验数据、真实测量或科研结论。**

package/omnischolar.config.example.json

@@ -131,12 +131,7 @@
         "apiKey": "",
         "apiKeyEnv": null,
         "baseUrl": "https://api.openai.com/v1",
-        "models": {
-          "gpt-image-1": {
-            "capabilities": ["text-to-image", "image-to-image", "edit", "multi-reference"],
-            "paid": true
-          }
-        },
+        "models": {},
         "options": {}
       },
       "google": {
@@ -214,12 +209,33 @@
         "apiKeyEnv": null,
         "baseUrl": "https://images.example.edu/v1",
         "models": {
-          "explicit-model-id": {
+          "gpt-image-2": {
+            "capabilities": ["text-to-image"],
+            "paid": true
+          },
+          "gpt-image-2.5-sunburst": {
+            "capabilities": ["text-to-image"],
+            "paid": true
+          },
+          "gpt-image-2.5-flare": {
+            "capabilities": ["text-to-image"],
+            "paid": true
+          },
+          "gemini-3.1-flash-image": {
+            "capabilities": ["text-to-image"],
+            "paid": true
+          },
+          "gemini-3.1-flash-lite-image": {
+            "capabilities": ["text-to-image"],
+            "paid": true
+          },
+          "gemini-3-pro-image": {
             "capabilities": ["text-to-image"],
             "paid": true
           }
         },
         "options": {
+          "modelCatalogEndpoint": null,
           "generationEndpoint": "images/generations",
           "editEndpoint": "images/edits"
         }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@luffysolution/omnischolar-pi",
-  "version": "0.1.20",
+  "version": "0.1.22",
   "description": "Pi MCP bridge for the OmniScholar research toolkit",
   "type": "module",
   "license": "MIT",

package/pi-extension/dist/index.js

@@ -1,7 +1,7 @@
 import { Client } from "@modelcontextprotocol/sdk/client/index.js";
 import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
 import { Type } from "typebox";
-const CLIENT_INFO = { name: "omnischolar-pi", version: "0.1.20" };
+const CLIENT_INFO = { name: "omnischolar-pi", version: "0.1.22" };
 export function resultText(result) {
     return result.content
         .map((item) => {