Skip to content

3.0.25

Choose a tag to compare

View all tags
@metantonio metantonio released this 23 Feb 02:23
3.0.25
94b47fd
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

QLX GPT 3.0.25

Important

Before install QLX-GPT be sure of:

  • Have installed CMake
  • Have installed 7zip
  • Have Ollama Installed (recommended)
  • (Optional if you are using Ollama, but if you want to use llamaCPP you will need to do this step) Have installed Visual Studio Community and during installation choose Universal Windows Platform development and C++ CMake tools for Windows.
  • By default QLX-GPT will use CPU for LLM inference, but if you want to use the GPU inference you will need to have installed CUDA (NVIDIA GPU needed for this, recommended with minimum of 8GB VRAM):
    • For Installer Type, choose exe (network).
    • During the installation, choose Custom (Advanced).
    • The only required components are:
    • Cuda
      • Runtime
      • Development
      • Integration with Visual Studio
    • Driver components
      • Display driver
  • SO: Windows 10/11 with 64-bits architecture
  • Have at least 15GB of free space on disk just to install QLX-GPT with default LLM, database, embedding models and vector store.
  • Have at least 16GB RAM.
  • CPU with AVX2 instruction capabilities if you are planning to use the CPU for LLM inference, or NVIDIA RTX GPU if you are planning to use the GPU for LLM inference.

Note: You don't need to have python installed, a custom portable version of python will be downloaded, and it won't interfere with other python versions or environment that you could have.

Tip

After installation, when you execute QLX-GPT app for the very first time:

  • It will download a portable version of python for this specific app... be patient
  • Backend windows will rebuild some python libraries for your specific CPU/GPU.
  • LLM model and Embedding models will be downloaded, be patient.
  • First Time the relational database will need to be created; it will create it automatically with SQLite. If you want to use another, add the string connection to the .env file. Note that this database will register users, prompts and path to documents and embeddings, by default MySQL is supported, if need another you will need to install a python library to be able to connect to it.
  • On the front-end, create a new account (click on Have an account? ). Then upload any document for the first time, this will enable the LLM to answer

Each time you open the application:

  • Wait for backend to be ready (it could take 30-60 seconds).
  • When LLM is asked for the first time (each time you open the app) it could take some seconds, or minutes, to be loaded on memory depending on your hardware. After being loaded in memory, it will be running as intended depending on your hardware.

Warning

When you make updates through the automatic updater, please be sure to have closed the backend windows before install any updates. Failing on doing this could cause potential corruption of the application, making necessary uninstall it and re-install again. By default, updates have an expiration date. Starting from version 3.0.12 the next expiration date for updates will be in January 2026.

Warning

If need to reinstall the python libraries on the portable python version:

Navigate to the `C:/python_embedded` folder (or the corresponding path if you have changed it).
Open the installation_log.txt file
Change the variable to true: BuildLlama=True
Open the app normally. The libraries will be reinstalled automatically.

If need to reinstall or reset the local database (SQLite):

Navigate to the C:/python_embedded folder (or the corresponding path if you have changed it).
Open the installation_log.txt file
Change the variable to true: CreateDB=True
Navigate to the C:/ folder (or the corresponding path if you have changed it).
Delete the migration folder
Delete the qlxgpt.db file
Open the app normally. The database will be reinstalled automatically.

If the database was reset and you want to reset the vector store to forget the current knowledge, follow these steps:

Navigate to the C:/qlx_models/local_data/vectorDB folder (or the corresponding path if you have changed it).
Delete all the contents of the folder.
Vector store will be created the next time you ingest documents

If need to change the Ollama LLM model:

Navigate to the C:\python_embedded folder (or the corresponding path if you have changed it).
Open the installation_log.txt file
Change the ollama model to the desired model, eg: OllamaModel=qwen2.5-coder:14b
Navigate to the `C:\Users\my-name\AppData\Local\Programs\QLX-GPT\resources\gpt` folder (or the corresponding path if you have changed it).
Open settings.yaml
Update de LLM model on Ollama sections, to make it coincide with the model in the installation_log.txt file. Note that by default `qwen2.5-coder:7b` is used.
Open the app normally and the new model will be used (and downloaded if doesn't exist yet)

Caution

If need to make a fresh installation again. Please uninstall first, then:

Delete the `C:/python_embedded` folder (or the corresponding path if you have changed it).
Delete the `C:/qlx_models` folder (or the corresponding path if you have changed it).
Delete the `C:/qlxgpt.db` file
Delete the `C:/migrations` folder
  • Note: Normally, the updater will work for most cases. Still, if you find an error and need to re-install from scratch, these are the steps to follow.

Note

Changelogs 3.0.24-3.0.25:

  • Fixed plotly graphs
  • Added programmatic tool calls
  • Added agents to search, read files
  • Added agents to search on web and search on specific website
  • Added agents to create workflows

Changelogs 3.0.24-3.0.23:

  • Updated the UI on database chat
  • Changed how communication with database works
  • Added plotly and react-plotly for graphs

Changelogs 3.0.23-3.0.17:

  • Updated the UI
  • Added new OCR features
  • Can select downloaded Ollama LLM models from the UI
  • Can upload documents directly from the conversation.
  • Added help section.
  • Added compatibility with QWEN3-VL models
  • Updated frontend libraries.
  • Updated transformers library to 5.0.0.
  • Added more notifications to the users.

Changelogs 3.0.17-3.0.16:

  • Added multi-watchdog events in order to scan changes in folders to ingest.

Changelogs 3.0.16-3.0.15:

  • UI Improved.
  • Standardized some buttons and icons
  • Added back Gmail node API help message.
  • Added back prompt cluster function.

Changelogs 3.0.15-3.0.14:

  • Added Credential Manager to the Gmail Node.
  • Added dropdown instead of the text input for the API Key in the Gmail Node.
  • Improved selection in the user menu, and icons.

Changelogs 3.0.14-3.0.12:

  • Added Credential Manager.
  • API Keys can be shared in the Credential Manager.
  • Owners of an API Key cannot be unshared/unlinked to it.
  • Shared API Keys can be shared by other users.

Changelogs 3.0.12-3.0.9:

  • Fixed bug for the default workflow that was not working correctly.
  • Added reload of folders list (group context) when the web scraping Node is executed.
  • Added new states to update old components based on the workflow actions.

Changelogs 3.0.9-3.0.8:

  • Fixed Gmail node to work with the workflow when it is the target of a prompting node.

Changelogs 3.0.8-3.0.5:

  • Added Gmail node, right now can send emails derivate from the concatenation node and the prompting node result.
  • Added a script on the installation to download the edge driver and updated it each time that the backend loads
  • Fixed login error
  • Improve workflow logic

Changelogs 3.0.5-3.0.4:

  • Rebuild of the Prompting Node to start accepting MCPs
  • Enabled backend when communication with server updater is broken
  • Fixed minor bugs/visuals

Changelogs 3.0.4-3.0.3-3.0.2:

  • Visuals Enhanced
  • Search bar works to find embeddings directly without AI response (search for documents)
  • Fixed minor bugs

Changelogs 3.0.2-3.0.0:

  • Added topological structure and order to be able to execute all the nodes in one click
  • Added execute workflow button
  • Fixed minor bugs

Changelogs 3.0.0-2.0.183:

  • Delegate the download of ollama models during the first installation and check on start.
  • Updated the README.md file and release instructions.

Changelogs 2.0.183-2.0.182:

  • Update to be compatible with Ollama and use it by default.

Changelogs 2.0.182-2.0.181:

  • Added support for Deepseek-R1-Destilll, recommended version 14B.
  • Changed default LLM model to DeepSeek working locally. Depending on the question it will use "Time Inference" technique, and use more or less time to get an answer.

Changelogs 2.0.181-2.0.180:

  • Added simulation of navigation with actual browser, instead of downloading html document.
  • Added tricks to avoid scraping agent detection.
  • Added capability to press buttons, links or any clickable element (just need a property attribute and the value, could be "id", "class", "data", "innerHTML", etc) Very useful for Single Page Applications, and/or +21 legal websites.
  • Changed how web scraping is made, in order to "convert" it into markdown format, this will make it easier for LLM to ingest the data.

Changelogs 2.0.179-2.0.178:

  • Changed how text files are being created when using the Web Scraping Node, changed the context_windows to 512 for it.
  • Added max_width for the Web Scraping Node to 500px
  • Added use_contex logic for the Prompting Node, when group_context_filter is detected with content

Changelogs 2.0.178-2.0.177:

  • Hot fix when passing group_context_filter variable from Concatenation Text Node

Changelogs 2.0.177-2.0.168:

  • Added Web Scraping Node
  • Added Node Resize for Web Scraping Node
  • Scraped websites will be ingested to reduce the context window for the LLM
  • The embedding for scraped data will be linked to a group for the node of the user
  • Group_Context_Filter field is passed through nodes to the Prompting Node
  • General Fixes

Changelogs 2.0.168-2.0.163:

  • Added Node Resize for Text Node
  • Improved logic when saving workflow json data for SSH Nodes
  • Improved visuals for SSH node terminal output
  • Improved visuals for Concatenation Text node.
  • Improved visuals for Text node.
  • General fix of known bugs

Changelogs 2.0.163-2.0.120:

  • Added the SSH Node (with backend and frontend functionalities)
  • Added Node Resize for Concatenation Text Node and SSH Node
  • Improved logic when loading workflow json data for SSH Nodes
  • Improved visuals
  • General fix of known bugs

Changelogs 2.0.120-2.0.109:

  • Enabled headers for the API Request Node
  • Enabled toggle side bar button for workflow
  • Improved workflow data that is being saved into a json file
  • Improved styles for easier reading
  • Added version text to welcome view.
  • Added License when installing for the first time.

Changelogs 2.0.108-2.0.98:

  • Added API Request Node, basic mode is able to do: GET, POST, PUT and PATCH Methods, using JSON format or equivalent Key-Value format. API's response can be use as output for the Concatenation Text Node and Prompting Node

Changelogs 2.0.97-2.0.62:

  • Added Text node
  • Added Concatenation node
  • Added dark, light and system theme
  • Refactorized nodeTypes and nodeVariables in one file
  • Added save and load buttons to save workspace into a json file locally.
  • Improved frontend logic when loading json files for the workspace

Changelogs 2.0.72-2.0.61:

  • Added prompt node
  • Prompt node works with socket connection, and now each node has separated socket thread
  • Added new functions and visualizations for workflow with nodes
  • Improved backend logic to manage sockets of each node

Changelogs 2.0.60:

  • Changed backend windows when open the app
  • Added ReactFlow library as main framework for nodes workflow
  • Added custom nodes and basic workflow window in electron to integrate ReactFlow to QLX-GPT

Changelogs 2.0.46-2.0.59:

  • Added new chunk text function for web scraping feature.
  • Changed Parsed body for more readability for the LLM

Changelogs 2.0.45:

  • Added feature to add directly to selected folder, the web scraping/youtube video to import.
  • Changed the user input, that was missing on recent updates.
  • Tested new option to write the folder's name.

Changelogs 2.0.30-2.0.42:

  • Added a parser for the web-scraping option, this should improve the readability of the websites for LLMs.
  • Added ftfy library to process non-unicode text
  • Added clean_utf8_text function
  • Added redundant text cleaning when injecting to vector database
  • Added cleaning function for characters not found due to language
  • Added more text cleaning features in based on printable characters
  • Added function to test utf-8 compatibility and extract website apparent encoding
  • Force utf-8 encoding when a document is ingested and then converted to binary

Changelogs 2.0.29:

  • Added web-scraping option, saving scraped websites into embeddings.

Changelogs 2.0.28 - 2.0.0:

  • Changed how updates are made making the process easier.
  • Added torch and torchvision for CPU/GPU, enabling experimental multi-modal vision, aiming to do OCR with it.
  • Eliminated notification when you have the latest version.
  • Added experimental tab, enabled only on development mode due to compatibility issues.
  • Added byaldi library.

Changelogs 1.12.0:

  • Changed where upload folder could be, can be changed at .env file

Changelogs 1.11.1:

  • Added bootstrap to the preloader, this should fix dropdowns and accordions behaviors.

Changelogs 1.11.0:

  • Vector store folder will be at C:\qlx_models\local_data path, this will help to keep the same vector store when installing updates.
  • The settings.yml file is changed to have C:\qlx_models\local_data as path for the vectorDB with qdrant

Changelogs 1.10.0:

  • Changed how initial installation is started, this was necessary when app is updated through updater.

Changelogs 1.9.0:

  • Database now will be at C:\ root, this will help to keep the same database when installing updates.
  • Python embedded changed to C:\, this will help to keep the same database when installing updates.
  • Added auto-updater
  • Refactorized some code in order to made easier future changes

Changelogs 1.8.0:

  • When you executed it for a second time, you had to press N on the backend window to complete the installation, now is automatic
  • The Installation and download of necessary files are now fully automatic when you open the app for the very first time.
  • The SQLite database installed on C:/Windows/System32 folder was changed to the project folder
  • The sensitive_words.json created at C:/Windows/System32 folder was changed to the project folder

Changelogs 1.6.0 and 1.7.0:

  • Fixed the download and open of documents with authentication for this desktop version

Changelogs 1.5.0:

  • Added execution policy
  • Added and implemented a semi-automatic setup of LLM when app is opened for the very first time.
Assets 7
Loading