RAG breaks results

OCR works fine without RAG, but when I add the same document to RAG it now doesn’t work properly.

Felix Croizon About 1 month ago

1

Couldn't deserialize server error

From today we are receiving this in response for our model and file we tested yesterday succesfully ["HTTP -1 - Unknown error :: 000-000 - Couldn't deserialize server error. Found: {\n \"data\": false,\n \"code\": 0\n}"] thrown by MindeeV2HttpUnknownError

Foreo IT 2 months ago

8

error

my model is not accepting JPEG image and also it is not working in the automation

Archit Gupta 2 months ago

1

Can't update my instructions - changes aren't saved

Hi there, anyone else having issues updating instructions? When I click save nothing happens, can’t update a models instruction for a given field in the data schema. Here’s a loom: https://www.loom.com/share/8b6dea47be8148f695cf29f5e0c32dcc Tried logging in and out, clearing cookies - happening on both Chrome and Firefox. Would appreciate if someone could let me know how to solve this. Thx, Roman

Roman Leeb 2 months ago

5

I can't add rows in RAG documents

Hello again! When I try to add a row in the RAG document review, the page doesn’t load. The browser console shows the following: I am using Firefox, but I also tested in Chrome and it happen again I can add guidelines, but when I click on the row button, the page turns black and doesn’t display anything. This happened yesterday, after I delete a document from the RAG and created a new one. However, I haven’t made any change to this for a while, so I don’t know it’s related to this. Best regards

Sergio Rigo 3 months ago

3

API Call error

Hello! When we tray t execute API call API https://api-v2.mindee.net/v2/inferences/enqueue with the Model_id 'b0f84468-d7bd-49c8-8818-eec5f01d3e9c' we get the following error, that 3 days ago when we migrated from V1 to V2 didn’t occured: { "statusCode": 500, "headers": { "Strict-Transport-Security": "max-age=31536000; includeSubDomains", "x-ms-environment-id": "default-172d5c03-04d9-45ba-adb5-b6d7b8f02e6a", "x-ms-tenant-id": "172d5c03-04d9-45ba-adb5-b6d7b8f02e6a", "x-ms-dlp-re": "-|-", "x-ms-dlp-gu": "-|-", "x-ms-dlp-ef": "-|-/-|-|-", "x-ms-apihub-cached-response": "false", "x-ms-apihub-obo": "false", "Date": "Wed, 29 Oct 2025 15:42:46 GMT", "Content-Length": "123", "Content-Type": "application/json" }, "body": { "status": 500, "detail": "An unexpected server error occurred.", "title": "Internal Server Error", "code": "500-003", "errors": [] } } Thank you

Fundació Kalida 3 months ago

4

### ❓ Issue: "HTTP 422 - Empty file sent" when using `Base64Input` with Mindee V2 SDK

Hi Mindee team 👋 I’m integrating the Mindee OCR API inside a NestJS backend (running in Docker), and I’m getting a consistent HTTP 422 - Empty file sent error when using Base64Input with the V2 client. 🧩 Code example this.client = new mindee.ClientV2({ apiKey: config.apiKey, }); async parseReceipt(imageBase64: string): Promise { const inputSource = new mindee.Base64Input({ inputString: imageBase64, filename: 'receipt.jpg', }); // I also tried manually initializing: // await inputSource.init(); // The Buffer is correctly filled here. const apiResponse = await this.client.enqueueAndGetInference( inputSource, { modelId: this.modelId, rawText: true, } ); return {}; } ⚙️ Context SDK: mindee (latest, ClientV2) Environment: NestJS (Node 20, Docker) Model: Input: Base64-encoded JPEG (I confirmed it’s valid and decodes properly) 🧠 What I noticed Even when I manually run await inputSource.init(), the inputSource.buffer is correctly filled with the decoded image data. However, as soon as I call: await this.client.enqueueAndGetInference(inputSource, { modelId: this.modelId }) the SDK still throws: MindeeHttpErrorV2: HTTP 422 - Empty file sent. 🙏 Question Is there something I’m missing when using Base64Input with ClientV2? Does it need a specific encoding format or initialization step before sending? Thanks for your help!

Folly Epiphane AVLAH 4 months ago

4

Bug in assistant ia edit

I'm focused on helping with document extraction. Could you please clarify what fields you'd like to extract from your documents? Everytime this reply at everything.

Clément DEBOVES 4 months ago

3

Bug in creation model with assistant

Hello, cant create model with assistant however i said. Since yesterday.

Nathan Sebban 4 months ago

2

500 Error when submitting documents

Good morning, last night we got a 500 error on all the documents sent. This morning I tried again and it gives the same error, I’m attaching it below. { "statusCode": 500, "headers": { "Date": "Fri, 03 Oct 2025 00:33:48 GMT", "Connection": "keep-alive", "Strict-Transport-Security": "max-age=31536000; includeSubDomains", "Content-Type": "application/json", "Content-Length": "62" }, "body": { "status": 500, "detail": "An unexpected server error occurred." } } { "statusCode": 500, "headers": { "Date": "Fri, 03 Oct 2025 07:28:59 GMT", "Connection": "keep-alive", "Strict-Transport-Security": "max-age=31536000; includeSubDomains", "Content-Type": "application/json", "Content-Length": "62" }, "body": { "status": 500, "detail": "An unexpected server error occurred." } }

Sergio Rigo 4 months ago

5

Performance Issue On some Invoices

On some invoices it takes hours to process it which normally needs to be at most 1 min. Is there any thing you can suggest us to do?

Aydin Karaman 4 months ago

1

RAG innacuracy

Hi! Recently I added a document to the RAG adding all the lines not retrieved by the invoice model. Then I tested it with the live test with the same pdf and it seems like it doesn’t work. The document has 30 lines of articles and the model only reads 23. Why? The live test is the following: https://app.mindee.com/model/573e3cec-2b4f-4f43-97c3-3f9841167919/live-test An the document on the RAG is: https://app.mindee.com/model/573e3cec-2b4f-4f43-97c3-3f9841167919/review?docs=b561b29d-b40b-4766-9955-537c312b041a

Juan Manuel Plaza Ruiz 4 months ago

2

Document review form resets after changing window

Hi! The document review form resets always I change the window. It’s very annoying because I want to copy data from the document and I can’t. Here you have a link to a loom video reproducing the issue: https://www.loom.com/share/5599fc6ea1494b8285bb3686c87a7ab3?sid=e6bcf3e2-3d8d-4834-8564-d8dab2aaf749

Juan Manuel Plaza Ruiz 4 months ago

4

lovable supabase mindee v2 from mindee v1

Hi Mindee team, I’m integrating the expense_receipts model via the V2 API from a Supabase Edge Function and I’m consistently receiving 401 Unauthorized. Endpoint being called POST https://api.mindee.net/v1/products/mindee/expense_receipts/v5.3/predict Authorization: Token md_******************** Content-Type: image/jpeg Accept: application/json (body: raw JPEG bytes) What I’ve verified Using a V2 key (prefix md_…), trimmed (no whitespace/newlines). Same key tested via curl returns 401 as well. Model path uses path segment versioning (/v5.3/), not @v5.3. Also tried predict_async; falls back to predict when not supported. Content types tried: image/jpeg and image/png. No proxy; requests originate from Supabase Edge (Deno) and from local curl. Recent logs (redacted) Mindee base: https://api.mindee.net/v1/products/mindee/expense_receipts/v5.3 Mindee response (sync) 401 Mindee error 401 … Questions / help requested Is the expense_receipts v5.3 public model accessible with standard V2 keys, or is a specific entitlement/workspace toggle required? Do V2 keys need to be created inside the same workspace/project where the receipts model is enabled? (Could a key from another workspace cause 401?) Can you confirm the correct product path for the receipts model in V2 (/v1/products/mindee/expense_receipts/v5.3/predict) and that raw binary (not multipart) with Content-Type: image/jpeg is acceptable? Are there any IP restrictions or organization settings that would return 401 instead of 403? If possible, could you check this key’s scope/status on your side (last 24h) to see why auth fails? Repro with curl export MINDEE_API_KEY='md_xxxREDACTEDxxx' curl -i -X POST \ -H "Authorization: Token $MINDEE_API_KEY" \ -H "Content-Type: image/jpeg" \ -H "Accept: application/json" \ --data-binary @sample.jpg \ "https://api.mindee.net/v1/products/mindee/expense_receipts/v5.3/predict" Thank you! Happy to provide full timestamps, workspace name, and any additional headers privately if that helps you trace the request.

chanua4 johnson 4 months ago

7

Not all responses are being sent to the webhook

Good morning, We have a flow in Power Automate from which we send the documents to be processed. Then we have another flow with a webhook where the document is sent. The issue is that since last night, we are not receiving all the responses to the webhook. Out of 23 documents sent last night, we only received the response for 2. I thought it might be due to too many API calls per second (although I haven’t received any errors, and in subsequent submissions I do receive a 202 response for all documents), so I added a 15-second delay between each call. In the first batch sent a little while ago, we received responses for 4 documents, in the next batch 4 more, then 6, and in the last one 7. I also want to point out that in the web panel the number of processed pages is increasing, so the documents seem to be processed, but the webhook responses are not always being received. This has never happened to us before. Could you please let me know if there is any kind of issue or limitation with the webhooks on your side? Thank you, Best regards

Sergio Rigo 5 months ago

8

Ruby pre release gem

Hi there, my team is working on the upgrade to Mindee V2 for a Ruby on Rails application. We’re currently using version 4.6.0 but are having issues with upgrading to version 4.7.0.pre.rc2, getting the following error in the console: bin/rails aborted! /opt/homebrew/lib/ruby/gems/3.3.0/gems/mindee-4.7.0.pre.rc1/lib/mindee/http/api_settings_v2.rb:23:in ` ': uninitialized constant Mindee::VERSION (NameError) USER_AGENT = "mindee-api-ruby@v#{Mindee::VERSION} ruby-v#{RUBY_VERSION} #{Mindee::PLATFORM}".freeze ^^^^^^^^^ Did you mean? VersionGem Is there any additional config we need to do/is anyone else having similar issues. With v1 becoming end of life very soon we’re keen to get this migration as seamless as possible.

Freya Dixon 5 months ago

6

Document parsing failed even for the Live Test test samples provided

I chose the Driver License model, and tested out the provided downloadable test samples, and none of them work. Then I tried my own real driver license and it also failed.

Mike Schmit 6 months ago

4

URL Error on Landing Page

You guys have a typo in your CTA on this page. Didn’t find a email address to send this to. You can delete the post here, if you like.

Thomas Grömer 6 months ago

1

Mindee Model

Hi, I was trialling Mindee and created a custom UK bank statement model, although when I try connecting a HTTP Make module, I am getting a 404 error code in Make. Is this because I am not a paid user or on a trial anymore with yourselves ?

Scott Ward 6 months ago

8

java SDK problem

Hello Mindee Team, I’m currently working with Java SDK to extract data from a URL source file, It was working perfectly fine on Friday, but since this morning I keep getting the same error [com.mindee.http.MindeeHttpExceptionV2: Invalid fields in form - HTTP 422 - Invalid fields in form], the same documents work just fine on live test but not using the Java SDK. Could you please help me resolve this issue ? Best regards

Said El haddati 6 months ago

3