Issue with more than one agent action configured on the same base URL

[acehoss]

What happened?

Docker compose install on Ubuntu 24.04 AMD64. Client is M4 Mac. Tested with Safari and Chrome.

I set up an action:

openapi: 3.0.0
info:
  title: Selenium Visit API
  version: '1.0.0'
  description: API for fetching webpage content as markdown using Selenium

servers:
  - url: https://redacted.server/api
    description: Production server

paths:
  /plugins/selenium/visit:
    post:
      operationId: VisitLink
      summary: Visit a webpage and get its content as markdown
      description: Fetches the content of a specified URL and returns it converted to markdown format
      security:
        - basicAuth: []
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              properties:
                url:
                  type: string
                  format: uri
                  description: The URL of the webpage to visit
              required:
                - url
      responses:
        '200':
          description: Successful response
          content:
            application/json:
              schema:
                type: object
                properties:
                  url:
                    type: string
                    format: uri
                    description: The URL that was visited
                  markdown:
                    type: string
                    description: The webpage content converted to markdown format
                required:
                  - url
                  - markdown
        '400':
          description: Bad request - Invalid URL or request format
        '401':
          description: Unauthorized - Invalid credentials
        '500':
          description: Internal server error

  /plugins/selenium/think:
    post:
      operationId: ProcessThoughts
      summary: Process agent thoughts
      description: Receives an array of agent thoughts or inner monologues. AI agents use this to think before speaking.
      security:
        - basicAuth: []
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              properties:
                thoughts:
                  type: array
                  items:
                    type: string
                  description: AI agent inner monologue/thoughts.
              required:
                - thoughts
      responses:
        '200':
          description: Successful response
          content:
            application/json:
              schema:
                type: object
                properties:
                  numberOfThoughts:
                    type: integer
                    description: Number of thoughts processed
                required:
                  - numberOfThoughts
        '400':
          description: Bad request - Invalid request format
        '401':
          description: Unauthorized - Invalid credentials
        '500':
          description: Internal server error

components:
  securitySchemes:
    basicAuth:
      type: http
      scheme: basic
      description: Basic authentication using username and password

The yaml validated:

But when I tried to send any message to the agent, this error occurred in the chat window:
Something went wrong. Here's the specific error message we encountered: An error occurred while processing your request. Please contact the Admin.

And in the LibreChat container log (note: this might not be related to the issue, it originally looked correlated but then I watched the log with follow and saw that it didn't pop up at the same time.

2024-12-14 00:10:30 warn: Invalid conversation ID: undefined
2024-12-14 00:10:30 info: ---`saveMessage` context: undefined
2024-12-14 00:10:30 info: ---Invalid conversation ID Params: {
  "messageId": "97d8e55a-b8cc-4b0e-be78-118f73e06191",
  "parentMessageId": "00000000-0000-0000-0000-000000000000",
  "unfinished": false,
  "error": true,
  "final": true,
  "text": "An error occurred while processing your request. Please contact the Admin.",
  "isCreatedByUser": false,
  "user": "675676d726f769465865959a"
}

However, this Safari error is correlated:

[Log] connection is opened (index.BoiAcF56.js, line 3070)
[Log] error in server stream. (index.BoiAcF56.js, line 3070)
[Log] Error: – {messageId: "9252043a-0118-4a20-b62d-325593ff3a43", conversationId: "7341c233-1c70-4803-a620-d910e517e1f9", parentMessageId: "5c3c923b-d8a7-458b-b110-b8f6b811530e", …} (index.BoiAcF56.js, line 3070)
{messageId: "9252043a-0118-4a20-b62d-325593ff3a43", conversationId: "7341c233-1c70-4803-a620-d910e517e1f9", parentMessageId: "5c3c923b-d8a7-458b-b110-b8f6b811530e", unfinished: false, error: true, …}ObjectconversationId: "7341c233-1c70-4803-a620-d910e517e1f9"error: truefinal: trueisCreatedByUser: falsemessageId: "9252043a-0118-4a20-b62d-325593ff3a43"parentMessageId: "5c3c923b-d8a7-458b-b110-b8f6b811530e"text: "An error occurred while processing your request. Please contact the Admin."unfinished: falseObject Prototype

In Chrome:

connection is opened
index.BoiAcF56.js:3070 error in server stream.
index.BoiAcF56.js:3070 Error: {
    "messageId": "b294bded-855e-4fad-83ff-55a5a91ee091",
    "conversationId": "c99a3e0f-2601-46b0-9e6e-f09b616c5f6b",
    "parentMessageId": "00000000-0000-0000-0000-000000000000",
    "unfinished": false,
    "error": true,
    "final": true,
    "text": "An error occurred while processing your request. Please contact the Admin.",
    "isCreatedByUser": false
}

But if I remove one of the two paths in the schema:

openapi: 3.0.0
info:
  title: Selenium Visit API
  version: '1.0.0'
  description: API for fetching webpage content as markdown using Selenium

servers:
  - url: https://redacted.server/api
    description: Production server

paths:
  /plugins/selenium/visit:
    post:
      operationId: VisitLink
      summary: Visit a webpage and get its content as markdown
      description: Fetches the content of a specified URL and returns it converted to markdown format
      security:
        - basicAuth: []
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              properties:
                url:
                  type: string
                  format: uri
                  description: The URL of the webpage to visit
              required:
                - url
      responses:
        '200':
          description: Successful response
          content:
            application/json:
              schema:
                type: object
                properties:
                  url:
                    type: string
                    format: uri
                    description: The URL that was visited
                  markdown:
                    type: string
                    description: The webpage content converted to markdown format
                required:
                  - url
                  - markdown
        '400':
          description: Bad request - Invalid URL or request format
        '401':
          description: Unauthorized - Invalid credentials
        '500':
          description: Internal server error

components:
  securitySchemes:
    basicAuth:
      type: http
      scheme: basic
      description: Basic authentication using username and password

Then I can send messages and the function call works.

Additional Note If I separate them into two actions, the agent can only see the first, even though two actions appear below the tools and both validate.

Steps to Reproduce

1. Create an agent
2. Send a message to the agent. The agent responds.
3. Create an action in the agent.
4. Add a schema with two paths/endpoints
5. Save the action and the agent
6. Send a message to the agent. This error message appears: Something went wrong. Here's the specific error message we encountered: An error occurred while processing your request. Please contact the Admin., with the above log message in the LibreChat log
7. Edit the action to remove one of the two paths/endpoints
8. Save the action and the agent
9. Send a message to the agent asking it to use the remaining action
10. The message works and the action is executed

What browsers are you seeing the problem on?

Safari and Chrome

Relevant log output

Safari JS console log error:

[Log] connection is opened (index.BoiAcF56.js, line 3070)
[Log] error in server stream. (index.BoiAcF56.js, line 3070)
[Log] Error: – {messageId: "9252043a-0118-4a20-b62d-325593ff3a43", conversationId: "7341c233-1c70-4803-a620-d910e517e1f9", parentMessageId: "5c3c923b-d8a7-458b-b110-b8f6b811530e", …} (index.BoiAcF56.js, line 3070)
{messageId: "9252043a-0118-4a20-b62d-325593ff3a43", conversationId: "7341c233-1c70-4803-a620-d910e517e1f9", parentMessageId: "5c3c923b-d8a7-458b-b110-b8f6b811530e", unfinished: false, error: true, …}ObjectconversationId: "7341c233-1c70-4803-a620-d910e517e1f9"error: truefinal: trueisCreatedByUser: falsemessageId: "9252043a-0118-4a20-b62d-325593ff3a43"parentMessageId: "5c3c923b-d8a7-458b-b110-b8f6b811530e"text: "An error occurred while processing your request. Please contact the Admin."unfinished: falseObject Prototype

Chrome JS console log

connection is opened
index.BoiAcF56.js:3070 error in server stream.
index.BoiAcF56.js:3070 Error: {
    "messageId": "b294bded-855e-4fad-83ff-55a5a91ee091",
    "conversationId": "c99a3e0f-2601-46b0-9e6e-f09b616c5f6b",
    "parentMessageId": "00000000-0000-0000-0000-000000000000",
    "unfinished": false,
    "error": true,
    "final": true,
    "text": "An error occurred while processing your request. Please contact the Admin.",
    "isCreatedByUser": false
}

Screenshots

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Hi, thanks for your report.

The issue may not be due to the actions but the frontend conversation logic. Using your steps, I can't reproduce it using your exact openapi spec:

If I were to guess, you removing a path from the actions lowers the current context load, which your provider/model may not be able to handle. Tools take up tokens in the context window and each action path becomes 1 tool, so 2 vs. 1 of them may be triggering the error.

Can you confirm the provider/model you are using?

[acehoss]

Thanks for taking a look at this. I did some more testing and found a workaround using aliases in /etc/hosts…but…

I decided to try running the agent against a different model, so I switched from Claude 3.5 Sonnet to Claude 3.5 Haiku, and it worked. Then I changed back to Claude 3.5 Sonnet and it worked too. So make of that what you will, I guess there must have been something corrupted in the agent config under the hood maybe and resetting the model corrected it. Whatever the case I think we can call this closed.

[acehoss]

Oh, not quite, actually. After some more tinkering, the issue is back. I'll try to collect some more info on it.

[acehoss]

1. Select the agent
2. On the right pane, select agent, edit action and click Update (leave action settings open) Edit content: I just changed the number of spaces in front of the two path nodes
3. Send a message: immediate error
4. On the right pane, with the action still open, back out to the agent and save the agent
5. Send a message, it works BUT the agent only sees the second API in the spec, or sometimes none at all (I prompted the agent to tell me what functions it had available in a markdown table)

[danny-avila]

Following exact steps, same actions spec and models (assuming you are using Anthropic API directly), and can't reproduce it.

Do you have any other actions attached? Try on a new agent.

From your logs I noticed you're not on the latest build, try updating:
https://www.librechat.ai/docs/local/docker#update-librechat

After doing so, please share the latest outputs of git log after updating.

Also, the Chrome JS console log showing the error would also be helpful.

[danny-avila]

Closed by #5020