Legacy/normal endpoints do not support OCR files.

[jmartinezb3]

What happened?

The problem is that only on normal endpoints (not agents) files that have gone through OCR show an error message when trying to interact. I have done the same process with the agent endpoint and it works correctly.

Interaction with error (normal/legacy endpoints)

Correct interaction (agents endpoints)

In the Gemini endpoint, the error does not appear but instead the context is not taken into account, the model does not know what I am talking about:

Version Information

docker images | grep librechat:

ghcr.io/danny-avila/librechat-rag-api-dev-lite   latest    04c3d3e0b6d4   23 hours ago    1.3GB
ghcr.io/danny-avila/librechat-dev                latest    151c0a323dc8   23 hours ago    913MB

Steps to Reproduce

1. Create agent (for using OCR function).
2. Submit a file (upload as text)
3. Use this file in a normal/legacy endpoint via Attach Files.

What browsers are you seeing the problem on?

Firefox

Relevant log output

Azure OpenAI log:

`
{
  text: "Please provide me with the Reddit post or a description of it. I need the text of the post to unders... [truncated]",
  conversationId: null,
  endpoint: "azureOpenAI",
  resendFiles: true,
    modelOptions.model: "gpt-4o",
  modelsConfig: "exists",
  attachments: [object Promise],
}
2025-03-27T19:03:38.222Z debug: [BaseClient] Loading history:
{
  conversationId: "62e2dbc9-cdba-462b-abeb-1b1f6bb366da",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
}
2025-03-27T19:03:38.747Z debug: [BaseClient] Context Count (1/2)
{
  remainingContextTokens: 127497,
  maxContextTokens: 127500,
}
2025-03-27T19:03:38.752Z debug: [BaseClient] Difference between original payload (1) and context (0): 1
2025-03-27T19:03:38.752Z warn: Prompt token count exceeds max token count (NaN / 127500).
2025-03-27T19:03:38.762Z error: [handleAbortError] AI response error; aborting request: { "type": "INPUT_LENGTH", "info": "NaN / 127500" }
2025-03-27T19:03:39.594Z debug: [AskController] Request closed

`

Gemini log:

`
{
  text: "What is the problem the reddit user is facing?",
  conversationId: null,
  endpoint: "google",
  resendFiles: true,
    modelOptions.model: "gemini-2.0-flash",
  modelsConfig: "exists",
  attachments: [object Promise],
}
2025-03-27T19:20:06.975Z debug: Creating GenAI client
2025-03-27T19:20:06.975Z debug: [BaseClient] Loading history:
{
  conversationId: "eba9b819-3eb9-41bb-a478-587a25d15aef",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
}
2025-03-27T19:20:07.086Z debug: [BaseClient] Context Count (1/2)
{
  remainingContextTokens: 991791,
  maxContextTokens: 991808,
}
2025-03-27T19:20:07.096Z debug: [BaseClient] Context Count (2/2)
{
  remainingContextTokens: 991791,
  maxContextTokens: 991808,
}
2025-03-27T19:20:07.096Z debug: [BaseClient] tokenCountMap:
{
  23374816-01cd-4710-b842-5e6b697794f4: 14,
}
2025-03-27T19:20:07.103Z debug: [BaseClient]
{
  promptTokens: 17,
  remainingContextTokens: 991791,
  payloadSize: 1,
  maxContextTokens: 991808,
}
2025-03-27T19:20:07.108Z debug: [BaseClient] tokenCountMap
{
  23374816-01cd-4710-b842-5e6b697794f4: 14,
}
2025-03-27T19:20:07.108Z debug: [BaseClient] userMessage
{
  messageId: "23374816-01cd-4710-b842-5e6b697794f4",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
  conversationId: "eba9b819-3eb9-41bb-a478-587a25d15aef",
  sender: "User",
  text: "What is the problem the reddit user is facing?",
  isCreatedByUser: true,
  tokenCount: 14,
}
2025-03-27T19:20:07.209Z debug: [saveConvo] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2025-03-27T19:20:08.244Z debug: [spendTokens] conversationId: eba9b819-3eb9-41bb-a478-587a25d15aef | Context: message | Token usage:
{
  promptTokens: 17,
  completionTokens: 34,
}
2025-03-27T19:20:09.261Z debug: [spendTokens] No transactions incurred against balance
2025-03-27T19:20:09.300Z debug: [saveConvo] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2025-03-27T19:20:09.313Z debug: [AskController] Request closed
2025-03-27T19:20:09.322Z debug: Creating GenAI client
2025-03-27T19:20:09.325Z debug: [BaseClient] Context Count (1/2)
{
  remainingContextTokens: 991696,
  maxContextTokens: 991808,
}
2025-03-27T19:20:09.325Z debug: [BaseClient] Context Count (2/2)
{
  remainingContextTokens: 991696,
  maxContextTokens: 991808,
}
2025-03-27T19:20:09.325Z debug: [BaseClient] tokenCountMap:
{
  unknown: [object Object],
}
2025-03-27T19:20:09.325Z debug: [BaseClient]
{
  promptTokens: 112,
  remainingContextTokens: 991696,
  payloadSize: 1,
  maxContextTokens: 991808,
}
2025-03-27T19:20:09.325Z debug: Creating GenAI client
2025-03-27T19:20:09.329Z debug: Identified titling model as GenAI version
2025-03-27T19:20:09.721Z debug: Title response: Reddit User's Problem
2025-03-27T19:20:09.721Z debug: [saveConvo] api/server/services/Endpoints/google/addTitle.js

`

Screenshots

No response

Code of Conduct

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

[danny-avila]

This is by design. Everything will move to agents eventually

[jmartinezb3]

Really? in my opinion I think it is not a good approach for the common user, since it requires configuration, it is much easier to open a new chat or directly change the endpoint in the interface, configuring an agent for each need is very cumbersome. I don't think it is good under that approach. Just an opinion.

[danny-avila]

To the user, it will be seamless on regular endpoints. They won't even know they will be using the same framework as agents, other than having additional options for file uploads and other features that will be more easily accessible soon.

[jmartinezb3]

To the user, it will be seamless on regular endpoints. They won't even know they will be using the same framework as agents, other than having additional options for file uploads and other features that will be more easily accessible soon.

Oh, I see. So what you're planning to do is essentially migrate all the endpoints to the Agents framework. I assume that once this is done, the legacy endpoints will be removed, and in the end, it will be practically the same for users who are already accustomed to the legacy endpoints. That's what I understand.

The issue I see lies in the setup of an agent every time the user wants a new interaction. This is quite confusing for the average user, having to configure agent names, instructions, etc. every single time they simply want to have a quick interaction, especially with different LLMs quickly, as is currently possible with the legacy endpoints.

In my opinion, agents are great for exactly that: having a specialized agent with precise instructions, pre-loaded knowledge, etc. But they’re not ideal for everyday interactions, quick queries, casual conversations, etc.

What you're talking about sounds more like an integration of everything into the Agents endpoint functionality? If that's the case, it seems great to me. I hope you can make the integration user-friendly for the average user.

Best of luck :)

[onestardao]

Hey, just jumping in because I’ve seen this OCR-related frustration pop up in multiple LibreChat + RAG threads recently.

This issue — where OCR-based content doesn’t work properly with certain endpoints — is actually a known failure mode we’ve catalogued in our open-source diagnostic map.

In this specific case, it's Failure Mode No.14–16 — related to undeployed or misaligned upstream components like unready OCR parsing, endpoint routing mismatch, or schema desync between front/backend.

I’ve run into and fixed these in production pipelines, so we created a complete list + corresponding fixes:

🔗 [WFGY ProblemMap – Full diagnostic map with real solutions]
(https://github.com/onestardao/WFGY/blob/main/ProblemMap/README.md)

And since OCR is central to many of these cases —

this project is also backed by the creator of Tesseract.js, who's endorsed our OCR handling modules.

If you're stuck trying to wire OCR + RAG + agent routing together, this might help avoid months of debugging.

Let me know if you want a pointer to the specific workaround we used for endpoint handoff + OCR ingestion.