Animus is a framework for building a personal AI - one that persists, learns, and serves a single user by design.
Schematics and diagrams that explain artificial intelligence.
ML by itself cannot be intelligent because it lacks reasoning, logic, and doesn’t interact with the environment. ML detects patterns based on statistical analysis of data using math based algorithms. These algorithms are not intelligent per se. Intelligence is much more than that.
Things of interest to me:
- Initializing a raw model in ollama and training it from scratch.
- Persistent memory with a relational database.
- Semantic search.
ONNX Runtime is a cross-platform inference and training machine-learning accelerator.
ONNX Runtime inference can enable faster customer experiences and lower costs, supporting models from deep learning frameworks such as PyTorch and TensorFlow/Keras as well as classical machine learning libraries such as scikit-learn, LightGBM, XGBoost, etc. ONNX Runtime is compatible with different hardware, drivers, and operating systems, and provides optimal performance by leveraging hardware accelerators where applicable alongside graph optimizations and transforms.
ONNX Runtime training can accelerate the model training time on multi-node NVIDIA GPUs for transformer models with a one-line addition for existing PyTorch training scripts.
Whisper is a general-purpose speech recognition model. It is trained on a large dataset of diverse audio and is also a multitasking model that can perform multilingual speech recognition, speech translation, and language identification. A Transformer sequence-to-sequence model is trained on various speech processing tasks, including multilingual speech recognition, speech translation, spoken language identification, and voice activity detection. These tasks are jointly represented as a sequence of tokens to be predicted by the decoder, allowing a single model to replace many stages of a traditional speech-processing pipeline. The multitask training format uses a set of special tokens that serve as task specifiers or classification targets. Multiple models are available. CLI tool, can be used as a Python module.
A complete end-to-end pipeline for training specialized Small Language Models (SLMs) on custom business data. OTTO enables organizations to train domain-specific language models without relying on expensive LLM fine-tuning or external APIs.
OTTO provides a streamlined workflow: Upload → Process → Preprocess → Train → Evaluate
The pipeline automatically handles file processing, data cleaning, tokenization, model training, and evaluation to produce specialized language models tailored to your specific use case. Works best with call transcripts, customer conversations, text-based business documents, natural language content (emails, reports, reviews), conversational data and dialog systems. Limited support for structured data (CSV, JSON, TSV), mixed media that would require video or speech recognition, and highly technical file formats (which will require preprocessing and cleaning). Seems reasonably tunable.
File formats supported: Plain text, Markdown, and ReStyled Text.
Data compression formats supported: .zip, .tar, .tar.gz, .tar.bz2, .gz
The RF Anomaly Detection System is a cutting-edge tool designed to monitor radio frequency (RF) spectrums and identify unusual signal patterns. Using powerful machine learning algorithms (especially unsupervised learning techniques), it can detect potential security threats, interference sources, and unauthorized transmissions—without prior knowledge of the specific threats.
Continuous monitoring of the RF spectrum for rapid anomaly detection. Advanced machine learning algorithms detect unusual signal patterns without the need for pre-existing threat data. Automatically categorizes detected anomalies based on signal characteristics. Compatible with RTL-SDR hardware for capturing real-world RF signals. Rich visualization tools like waterfall displays and heatmaps to simplify signal analysis. Configurable notifications to keep you updated on detected anomalies.
What if a Tamagotchi had a neural network and could learn stuff? Includes tools for visualising and understanding neural networks, Hebbian learning and neurogenesis. Requires PyQt5 and numpy.
The squid moves autonomously, making decisions based on his current state (hunger, sleepiness, etc.) Implements a vision cone for food detection, simulating realistic foraging behavior. Neural network can make decisions and form associations. Weights are analysed, tweaked and trained by Hebbian learning algorithm. Experiences from short-term and long-term memory can influence decision-making. The squid can create new neurons in response to his environment (neurogenesis).
Tracks various needs like hunger, sleepiness, happiness, and cleanliness. Needs change over time and affect the pet's health and behavior. The squid can become sick and die if his needs are neglected. Be aware the squid hates taking medicine and will become depressed and need sleep if made to do so. Seven different personality types (so far).
Basic Memory lets you build persistent knowledge through natural conversations with Large Language Models (LLMs) like Claude, while keeping everything in simple Markdown files on your computer. It uses the Model Context Protocol (MCP) to enable any compatible LLM to read and write to your local knowledge base.
AI assistants can load context from local files in a new conversation. Notes are saved locally as Markdown files in real time. No project knowledge or special prompting required.
Most LLM interactions are ephemeral - you ask a question, get an answer, and everything is forgotten. Each conversation starts fresh, without the context or knowledge from previous ones. Basic Memory addresses these problems with a simple approach: structured Markdown files that both humans and LLMs can read and write to. All knowledge stays in files you control. Both you and the LLM read and write to the same files. LLMs can follow links between topics. Indexed in a local SQLite database.
str2speech is a simple command-line tool for converting text to speech using Transformer-based text-to-speech (TTS) models. It supports multiple models and voice presets, allowing users to generate high-quality speech audio from text.
Supports multiple TTS models, including suno/bark-small, suno/bark, and various facebook/mms-tts models. Allows selection of voice presets. Supports text input via command-line arguments or files. Outputs speech in .wav format. Works with both CPU and GPU.
Looks like the speech models have to be installed locally to work.
Yet Another Soft Robot Evolver. Evolutionary computation experiments using Evolution Gym as a base to play with evolutionary computation algorithms and other weirder things. It can also be useful as a minimalist codebase to learn how to use evogym without having to worry about PPO and stuff.
Open WebUI is an extensible, feature-rich, and user-friendly self-hosted WebUI designed to operate entirely offline. It supports various LLM runners, including Ollama and OpenAI-compatible APIs. Effortlessly integrate OpenAI-compatible APIs for versatile conversations alongside Ollama models. Customize the OpenAI API URL to link with LMStudio, GroqCloud, Mistral, OpenRouter, and more. Seamlessly integrate custom logic and Python libraries into Open WebUI using Pipelines Plugin Framework. Launch your Pipelines instance, set the OpenAI URL to the Pipelines URL, and explore endless possibilities. Examples include Function Calling, User Rate Limiting to control access, Usage Monitoring with tools like Langfuse, Live Translation with LibreTranslate for multilingual support, Toxic Message Filtering and much more. Enjoy a seamless experience across Desktop PC, Laptop, and Mobile devices.
Txtify is a free and open-source web app for converting audio and video to text using advanced AI models. It supports YouTube videos and personal media files, offering fast and accurate transcriptions. Txtify can be self-hosted, giving you full control over your transcription process.
Singulatron is an app that lets you run AI anywhere! It is private, works offline, and can run on your laptop, PC, or even on your company computers or servers. It's not just an app but also a platform that enables building other AI applications on top of it.
Singulatron aims to be both a desktop app for local usage and also to work as a distributed daemon to drive servers, with a web app frontend client that is the same as the local app. Private: your chats never leave your computer.
Works without an internet connection. The prompt queue system lets you input many prompts at once - even across threads - they will be processed sensibly. You can leave threads and return - streaming won't be interrupted. A download manager makes sure your models are well kept.
Unfortunately, it's an Electron app.
Perplexica is an open-source AI-powered searching tool or an AI-powered search engine that goes deep into the internet to find answers. Inspired by Perplexity AI, it's an open-source option that not just searches the web but understands your questions. It uses advanced machine learning algorithms like similarity searching and embeddings to refine results and provides clear answers with sources cited. Using SearxNG to stay current and fully open source, Perplexica ensures you always get the most up-to-date information without compromising your privacy.
You can make use local LLMs such as Llama3 and Mixtral using Ollama. Normal or Copilot modes. Special modes to better answer specific types of questions. Some search tools might give you outdated info because they use data from crawling bots and convert them into embeddings and store them in a index. Unlike them, Perplexica uses SearxNG, a metasearch engine to get the results and rerank and get the most relevant source out of it, ensuring you always get the latest information without the overhead of daily data updates.
Has a documented installation process that doesn't require Docker.
Chunking documents is a challenging task that underpins any RAG system. High quality results are critical to a sucessful AI application, yet most open-source libraries are limited in their ability to handle complex documents. Open Parse is designed to fill this gap by providing a flexible, easy-to-use library capable of visually discerning document layouts and chunking them effectively.
Visually driven. Parses Markdown. Can analyze data tables by extracting them into Markdown tables.
Surya is a document OCR toolkit.
- Accurate OCR in 90+ languages
- Line-level text detection in any language
- Layout analysis (table, image, header, etc detection) in any language
Built on top of PyTorch. Multiple models.
Cross-platform, open-source voice assistant and framework to build fully-featured, offline machines you can talk to. Self-hosted. Desktop and mobile clients. Repos of note:
- sepia-stt-server - Public SEPIA server to support open-source speech recognition via WebSocket connection.
- sepia-assist-server - Core server of the SEPIA Framework responsible for NLU, conversation, smart-service integration, user-accounts and more.
- sepia-html-client-app - Application to communicate with SEPIA via browser, iOS and Android. Works as chat messenger with personal-assistant, ASR and TTS integration.
- sepia-admin-tools - Tools to connect to your SEPIA servers, change settings, organize users, test the APIs and more.
- sepia-installation-and-setup - Instructions, scripts, tools and files to install, setup and run the SEPIA-Framework on Raspberry Pi, Windows, Mac and other Linux systems.
AlgorithmWatch is a human rights organization based in Berlin and Zurich. We fight for a world where algorithms and Artificial Intelligence (AI) do not weaken justice, democracy, and sustainability, but strengthen them.
Bark is a transformer-based text-to-audio model created by Suno. Bark can generate highly realistic, multilingual speech as well as other audio - including music, background noise and simple sound effects. The model can also produce nonverbal communications like laughing, sighing and crying. To support the research community, we are providing access to pretrained model checkpoints, which are ready for inference and available for commercial use.