turbovec is a fast rising and impressive vector index library built on the TurboQuant algorithm. Support for it was just added to TxtAI and will be available next release!
https://t.co/PuI5uhzHDz
Paper: https://t.co/WRIKWhsh7g
TxtAI Issue: https://t.co/fnZYkxBXCK
TxtAI 9.10 is out! TxtAI continues to invest heavily in local and edge device AI!
This release adds support for generating vectors via LiteRT for edge device use cases. It also adds support for training small models via Knowledge Distillation.
https://t.co/FOZ7F3498w
π Did you know that TxtAI Embeddings instances support SQL and openCypher queries? An embeddings graph automatically uses the vector similarity model to build an entire graph network of nodes.
β Training tiny models requires a different playbook. Check out this example article that covers how to progressively distill knowledge into a tiny 250K parameter model.
https://t.co/XhmGXfmn7U
Tiny AI isn't just about tiny models. It's also about the install footprint in tiny spaces.
With TxtAI's minimal install you can say run a full RAG+LLM+Embeddings solution with only 10 packages and GPU (or NPU) support.
π Want a vector model that's less than 1M parameters can be as small as less than 1 MB? Want to run it on Mobile? Check this model out then. It's our export of the popular BERT Hash series!
https://t.co/DdNPTiYkUp
π₯ The next version of TxtAI will support running LiteRT vector models (formerly known as TensorFlow Lite). Check out this version of the popular all-MiniLM model!
https://t.co/7CI1fDUuxU
π₯ TxtAI is an all-in-one AI framework. With the new minimal install it can also be the none-in-one or some-in-one framework.
Check out this example that has zero dependencies where TxtAI can be a simple JSON object store.
Why care about TxtAI's zero dependency install? Well Transformers and Torch bring in a lot of dependencies. That's great if you need them but if you just want to run say a llama.cpp focused solution or only use the Textractor pipeline, it's a lot of unnecessary transitive dependencies and increases the overall image size. Problem solved!
π TxtAI 9.9 is out! This release brings a big and important change: the zero dependency build. Previously, the base install required Transformers and Torch which brought the install up to at least 4GB. Now with providers like llama.cpp and LiteRT, a base install can be under 100MB with full GPU support!
Release Notes: https://t.co/LMRd2U3CNu
GitHub: https://t.co/4v0zmzyeWG
Ever since the original v1.0 release back in 2020, TxtAI has relied on a Transformers and Torch install. But now with more lightweight options such as llama.cpp, it's time to allow TxtAI to run without those libraries!
Issue: https://t.co/wAWOhdNmOy
https://t.co/t6KZHx3y8G
We all use AI to summarize, process and digest information in 2026. What about if you'd rather have AI read a document and automatically highlight important concepts? Then still read the source.
If this sounds interesting, check out AnnotateAI! Works great with small local models such as Gemma 4 and Qwen. Fully open source (Apache 2.0) and it comes with a Docker quickstart image. Enjoy.
https://t.co/5GVc28cglt
Important change coming with the next TxtAI release - the ability to run without torch and with llama-cpp for edge device use cases.
https://t.co/IgcQXiSnnz
TxtAI 9.8 is out! This release adds a number of performance, security and compatibility improvements!
GitHub: https://t.co/t6KZHx3y8G
Release Notes: https://t.co/woMhEhuFrV
We're excited to release the new BiomedBERT Small series of models. These 22.7M parameter models, trained for medical literature, are similarity sized to the popular all-MiniLM-v2 models and pack quite a punch.
Read more here: https://t.co/esuMtOLWep