Replies: 14 comments 15 replies
-
|
To let users choose an LLM model provider instance when configuring a Dify plugin, you should add a parameter to your plugin's configuration file (usually YAML) that represents the provider selection. You can define this parameter as a string or a select type, and optionally provide a list of available providers as options. Here's a concrete example from an official plugin config: parameters:
- name: vlm_model_provider
type: string
required: false
default: openai
label:
en_US: VLM Model Provider
zh_Hans: VLM模型提供商
ja_JP: VLMモデルプロバイダー
human_description:
en_US: The provider of the VLM model
llm_description: The provider of the VLM model
form: form
# If you want a dropdown, add options:
# options:
# - value: openai
# label:
# en_US: OpenAI
# - value: azure
# label:
# en_US: AzureSee a real example here. If you want the UI to show a dropdown, use In your plugin's Python code, access the selected provider value from the parameters dictionary and use it to route requests or validate credentials. For best practices, localize labels/descriptions and validate the selected provider before use. For more examples, check how plugins like Qwen/Tongyi and OpenRouter handle user-selectable parameters in their YAML and Python files (reference). To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
I want to add an existed node's output variable type is a json array[object]. My plugin want to use this json array[object] as an input variable. I want restict the ui can only show the array type variables while user link nodes and configure it. |
Beta Was this translation helpful? Give feedback.
-
|
To restrict your plugin input parameter so the UI only shows variables of type JSON array[object] when linking nodes, define the parameter in your YAML like this: parameters:
- name: context
type: any
scope: array[object]
required: true
label:
en_US: Context
zh_Hans: 上下文This setup tells Dify to only display variables of type array[object] for this input in the UI. You can see this pattern in official plugins and discussions (example, discussion). Important: If the array[object] variable comes from a plugin node, make sure the output schema for that variable includes detailed item definitions (i.e., the structure of each object in the array). Without this, the UI may not show the variable as selectable in downstream nodes due to frontend filtering logic (details). Let me know if you need an example of a full output schema or have other questions! To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
I used this to define input parameter
but ui the generaete result don't show the input, just show the lable |
Beta Was this translation helpful? Give feedback.
-
|
The UI only shows the label (and not the input field) because the upstream node’s output schema for your array[object] variable is missing a detailed items definition. Dify’s frontend will only let you select array[object] variables as inputs if the output schema of the previous node includes a full description of the structure of each object in the array. To fix this, update the output_schema of the node producing the array so that it includes an explicit items schema. For example, in official plugins, the output schema looks like this: output_schema:
type: object
properties:
result:
type: array
items:
type: object
properties:
parent_content:
type: string
description: The parent content
child_contents:
type: array
items:
type: string
description: List of child contents
required: [parent_content, child_contents]
description: List of parent-child chunk pairsOr, you can reference an external schema: output_schema:
type: object
properties:
result:
$ref: "https://dify.ai/schemas/v1/parent_child_structure.json"See this example and the corresponding JSON schema. Once the upstream node’s output schema includes a detailed items definition, the variable picker UI will show the input field as expected. If you want, share your upstream node’s output_schema and I can help review it. To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
我的plugin以debug方式部署后,在plugin的配置窗口里我定义的这个
|
Beta Was this translation helpful? Give feedback.
-
|
你的参数定义 如果你确实需要传递多个字符串，可以用字符串参数，用户输入逗号分隔的内容，然后在插件逻辑里用代码分割成数组。例如： parameters:
- name: json_array
type: string
required: true
label:
en_US: json array
zh_Hans: json数组
human_description:
en_US: "Comma separated values"
zh_Hans: "逗号分隔的字符串"插件代码里处理： json_array = params['json_array'].split(',')如果你的场景必须用数组类型，只能通过 OpenAPI schema 或官方支持的特殊类型实现，普通插件参数暂时无法直接支持数组输入框 参考。 有具体需求可以补充说明，我可以帮你找合适的实现方式。 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
但是我去掉scope后,我将type定义为string或者files,在UI上可以显示出来变量的选择框可以选择文件或者字符串的变量.
我想知道如果我需要给这个插件定义一个是json数组的input变量,允许用户在ui界面上选择minerU节点生成的json格式的array数组作为输入,如何去定义? |
Beta Was this translation helpful? Give feedback.
-
|
要让你的插件参数在 Dify UI 上支持选择 minerU 节点生成的 JSON 数组（array[object]），需要满足两个条件：
output_schema:
type: object
properties:
images:
type: array
items:
type: object
properties:
url:
type: string
description: 图片链接
label:
type: string
description: 标签
required: [url, label]
description: 图片数组或者引用外部 schema（官方 chunker 示例）： output_schema:
type: object
properties:
result:
$ref: "https://dify.ai/schemas/v1/parent_child_structure.json"
parameters:
- name: json_array
type: any
scope: array[object]
required: true
label:
en_US: json array
zh_Hans: json数组
human_description:
en_US: "Pdf Chunk Plugin"
zh_Hans: "分块"
llm_description: "Pdf Chunk Plugin"
form: llm只要上游节点的 output_schema 定义详细，UI 就会显示变量选择器，允许你选择 minerU 的 JSON 数组变量作为输入。 为什么 type: string/files 能选变量，但 array 类型不行？ 如果无法修改上游 schema，可以用字符串参数，传递 JSON 字符串，然后在插件里解析。 如需进一步帮助，可以贴出你的 minerU 节点的 output_schema，我可以帮你 review。 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
但是为什么迭代器节点可以选择mineru输出的节点中的json数组作为输入呢 |
Beta Was this translation helpful? Give feedback.
-
|
这是 Dify 设计上的一个区别：迭代器节点（Iteration/Loop Node）在前端和后端有专门的逻辑，允许它直接选择上游节点输出的任何数组类型（包括 minerU 的 JSON 数组）作为输入，即使上游 output_schema 没有详细定义数组元素结构。前端会对迭代器节点的输入做 array 类型专属过滤，允许 array、arrayObject 等类型直接作为输入变量选择（详见前端实现）。 而自定义插件节点的参数，前端会更严格地校验上游 output_schema，只有当 output_schema 里数组元素（items）有详细属性定义时，才会在变量选择器中显示 array[object] 变量，否则只显示 label，不显示输入框（详见讨论）。 所以，迭代器节点能选，是因为它有特殊的 array 输入处理逻辑；自定义插件要实现类似效果，必须让上游节点的 output_schema 详细定义数组元素结构（如 items 里写明每个 object 的属性），这样变量选择器才会显示出来。 如果你需要让自定义插件支持选择 minerU 的数组输出，建议让 minerU 的 output_schema 里 array 字段的 items 加上详细属性定义。这样插件参数用 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
首先我看到了mineru官方的插件中已经定义了images如下,那么为什么这个讨论(https://github.com/langgenius/dify-official-plugins/issues/2074)中问题还会出现? 是必须指明是image,而不能是object? 其次我如果要将json部分的数组输出,我也需要参照images,这样定义吗? |
Beta Was this translation helpful? Give feedback.
-
|
这个问题的根本原因是：Dify 的前端变量选择器（variable picker）对于 array[object] 类型非常严格，只有当 output_schema 里的数组元素（items）有详细的属性定义时，插件参数才能在 UI 里显示为可选变量。如果只是 正确做法：
output_schema:
type: object
properties:
images:
type: array
items:
type: object
properties:
url:
type: string
description: 图片链接
label:
type: string
description: 标签
required: [url, label]
description: 图片数组
output_schema:
type: object
properties:
jsons:
type: array
items:
type: string
description: jsons结论：
如果你要让 jsons 像 images 一样作为结构化数组被下游插件选择，必须根据实际需求定义 items 的属性结构。如果只是字符串数组，直接用 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
但是我在本地为这个mineru添加了 然后我需要做一个patch版本的时候为什么会提示下面的错误 dify plugin package mineru |
Beta Was this translation helpful? Give feedback.
-
|
这个错误不是因为你修改了 output_schema 或添加了 请检查以下几点：
cd your-workspace
dify plugin package ./mineru
只要目录和文件都在，命令路径正确，就不会出现 “The system cannot find the file specified” 的错误。这个问题和你 YAML 里 参考：官方文档说明 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
那这样build的patch版本的mineru的插件,在dify中安装时候出现错误 req_id: 7ca83c0560 PluginDaemonBadRequestError: plugin verification has been enabled, and the plugin you want to install has a bad signature |
Beta Was this translation helpful? Give feedback.
-
|
这个错误是因为 Dify 启用了插件签名验证（plugin verification），而你打包的 patch 版本 mineru 插件没有有效签名或签名不被当前系统信任。只要插件被修改或重新打包，原有签名就会失效，Dify 就会拒绝安装并报 “bad signature” 错误。 解决方法：
注意事项：
如需详细签名流程或 repackaging 工具用法，可以补充说明你的环境和需求。 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
我已经可以自己build出来一个修改了mineru的output_schema的patch版本的 那么我如何修改下面的定义的soul_pdf_chunk_plugin输入参数json_array,可以使用这个patch版本定义的jsons数组? identity:
|
Beta Was this translation helpful? Give feedback.
-
|
你只需要把 - name: json_array
type: any
scope: array[string]
required: true
label:
en_US: json array
zh_Hans: json数组
human_description:
en_US: "Pdf Chunk Plugin"
zh_Hans: "分块"
llm_description: "Pdf Chunk Plugin"
form: llm这样，Dify UI 就会自动识别并允许你选择 mineru 节点输出的 如果 UI 还是不显示变量选择器，请确认 mineru 的 output_schema 里 jsons:
type: array
items:
type: string
description: jsons如遇缓存或前端未刷新，建议重启 Dify 前端或刷新页面。如果还有问题可以贴出你的 mineru output_schema 和插件参数定义，我可以帮你进一步排查。 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
但是现在问题是在我新的Pdf Chunk Plugin中,我定义了json_array,但是我在UI上没有显示出来json_array的输入框, 需要怎么修改才能让webUI显示显示出来json_array,的输入框. |
Beta Was this translation helpful? Give feedback.
-
|
你遇到的 UI 不显示
如需进一步排查，请贴出你当前的插件参数 YAML 和 mineru 的 output_schema。 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
请dify官方尽快取消掉限制吧!!! 2025-12-03 19:00:38.086 INFO [GraphWorker-3] [_client.py:1038] - HTTP Request: POST http://sandbox:8194/v1/sandbox/run "HTTP/1.1 200 OK" |
Beta Was this translation helpful? Give feedback.
-
|
你遇到的 “Output length exceeds 400000 characters” 错误，是 Dify 后端对单个节点输出长度的限制。这个限制由环境变量 CODE_MAX_STRING_LENGTH 和 TEMPLATE_TRANSFORM_MAX_LENGTH 控制，默认都是 400000 字符。你可以通过如下方式提升上限：
注意：提升这些限制会增加内存和 CPU 占用，可能导致系统变慢或不稳定，建议只在确有需要时调整，并观察资源消耗。配置参考 如果你只是偶尔需要处理超大数据，建议把数据拆分成多批处理，或者用外部存储/分页等方式规避单节点超大输出。更多讨论和最佳实践 至于 “This parameter only accepts one file but got multiple files while invoking.”，说明你的插件参数定义为单文件（如 type: file），但实际传入了文件数组。请检查参数定义和工作流变量映射，确保类型一致。如果需要支持多文件，参数应定义为 array[file] 并在插件逻辑中处理数组。 如需进一步排查，可以贴出你的参数定义和实际传参示例。 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
那对与系统预置好的比如模板转换节点是直接作为一个node节点存在,还是也是作为一个plugin存在的?我自定义的plugin能直接继承自模板转换节点吗? |
Beta Was this translation helpful? Give feedback.
-
|
系统预置的模板转换节点（Template Transform）在 Dify 里是作为内置节点（built-in node）实现的，不是 plugin。它在后端有专门的 Python 类（TemplateTransformNode），前端也有专属的 UI 组件，属于系统核心节点类型之一（NodeType.TEMPLATE_TRANSFORM），和插件（tool/plugin）是两套机制，互不继承、扩展。 自定义 plugin（工具插件）不能直接继承或扩展模板转换节点的功能。如果你想用模板转换的能力，可以在工作流里先加一个模板转换节点，再接你的 plugin，或者在 plugin 代码里自己实现类似的模板处理逻辑（比如用 Jinja2 渲染）。 如果你有特殊的节点扩展需求，只能通过后端开发自定义 Node 类（需要修改 Dify 源码），不能通过 plugin manifest 或 handler 实现直接继承。 To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
parameters:
Traceback (most recent call last): |
Beta Was this translation helpful? Give feedback.
-
|
Dify does parameter type checking and conversion before your plugin code runs, so errors like “The selector must be a dictionary.” can happen even if your However, the If you need a reference for modern plugin parameter config: parameters:
- name: mode
type: select
label:
en_US: Mode
options:
- value: fast
label:
en_US: Fast
- value: accurate
label:
en_US: Accurate
required: true
form: formFor plugin compatibility and best practices, check the latest plugin development docs. To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
-
|
但是在plugin的可选类型中还是有这个model-selector吗,也没有提示这个已经过时的参数,那我如何实现让用户在plugin的配置介面上让用户去配置一个系统级的llm模型呢? |
Beta Was this translation helpful? Give feedback.
-
|
目前 plugin 的参数类型中确实还包含 - name: model
type: model-selector
scope: llm
required: true
label:
en_US: Model
zh_Hans: 模型
human_description:
en_US: "Select LLM model"
zh_Hans: "选择大模型"
llm_description: "Select LLM model"
form: form这样用户在插件配置界面就能选择系统已配置的 LLM 模型实例， 注意事项：
目前没有官方废弃公告， To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
|
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Self Checks
1. Is this request related to a challenge you're experiencing? Tell me about your story.
I want to develop a new dify plugin that the plugin will permit customer to select LLM model provider instance. How can I add this parameter to plugin configuration file to define this required parameter?
Where Can I find some examples that has similar functions?
2. Additional context or comments
No response
Beta Was this translation helpful? Give feedback.
All reactions