Novro Platform
Svenska
Sign in

Open source in AI

Language models, speech, embeddings, chat UIs, RAG and agent builders.

Common tools in this category

Show 92 more tools

Open-source products · 191

AI agent & LLM app builders

Dify

Build AI agents and RAG apps visually.

LicenseRef-DifyCustom licence

SaaS: NoClosed product: Unclear – verify

Flowise

Visual AI agent builder. The project is archived – not recommended for new installations.

Apache-2.0PermissiveOpen coreArchived

SaaS: YesClosed product: Yes

Paperclip

Control plane for running AI agents at work: goals, tasks, approvals, budgets and governance.

MITPermissivenot verified

SaaS: YesClosed product: Yes

AutoGPT

Platform for building and running autonomous AI agents that carry out multi-step tasks.

LicenseRef-PolyForm-Shield-1.0.0Source-available (not OSI)

SaaS: NoClosed product: No

Langflow

Visual builder for AI agents and LLM workflows. You assemble flows as nodes and can serve them as an API.

MITPermissive

SaaS: YesClosed product: Yes

LangChain

Framework for developers building apps and agents on top of language models, with ready-made integrations for models, tools and data sources.

MITPermissive

SaaS: YesClosed product: Yes

AutoGen

Programming framework from Microsoft for building AI agents that work together with each other and with people.

MITPermissive

SaaS: YesClosed product: Yes

CrewAI

Python framework for letting several AI agents with different roles work together on a task.

MITPermissive

SaaS: YesClosed product: Yes

LangGraph

Library for building agents as graphs of steps, with persisted state, interruptions and human approval.

MITPermissive

SaaS: YesClosed product: Yes

Agno

Python framework and runtime for building, running and monitoring AI agents and multi-agent systems.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

OpenAI Agents SDK

Lightweight Python framework from OpenAI for multi-agent workflows, hand-offs between agents and tracing.

MITPermissive

SaaS: YesClosed product: Yes

Sim

Web workspace where teams build, deploy and monitor AI agents and workflows visually.

Apache-2.0PermissiveOpen core

SaaS: YesClosed product: Yes

smolagents

Small Python library from Hugging Face for agents that solve tasks by writing and running code.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Mastra

TypeScript framework for building AI agents, workflows and RAG.

Apache-2.0PermissiveOpen core

SaaS: YesClosed product: Yes

Letta

Platform for AI agents with long-term memory that keep their state between conversations.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Coze Studio

Platform with visual tools for creating, debugging and deploying AI agents and chatbots with little code.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

VoltAgent

TypeScript framework for AI agents with tools, memory and workflows, plus a console for following runs.

MITPermissive

SaaS: YesClosed product: Yes

Strands Agents (Harness SDK)

SDK for Python and TypeScript for building production AI agents, independent of model and cloud.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Strands Agents Tools

Collection of ready-made tools (files, web, code, API calls) that agents built with Strands Agents can use.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Strands Agent SOP

Format for describing work procedures in natural language so that AI agents carry out multi-step tasks predictably.

Apache-2.0PermissiveYoung project

SaaS: YesClosed product: Yes

Strands Shell

Gives an AI agent a shell to run commands in, sandboxed so it doesn't get full access to the machine.

Apache-2.0PermissiveYoung project

SaaS: YesClosed product: Yes

Strands Decider

Small, fast model for agent workflows that picks between options or rates on a scale, with a confidence score.

Apache-2.0PermissiveYoung project

SaaS: YesClosed product: Yes

AI chat workspace

LibreChat

Chat interface for multiple AI providers and your own models.

MITPermissive

SaaS: YesClosed product: Yes

Open WebUI

Very popular chat interface for local and remote models.

LicenseRef-Open-WebUICustom licencenot verified

SaaS: Yes, with conditionsClosed product: Unclear – verify

AnythingLLM

RAG workspace: chat with your own documents.

MITPermissive

SaaS: YesClosed product: Yes

Brigade

Ecosystem for self-hosted AI agents ("personal intelligence").

MITPermissiveOpen coreYoung projectnot verified

SaaS: YesClosed product: Yes

NextChat

Chat interface for language models that runs on the web, on desktop or self-hosted, with your own API keys.

MITPermissive

SaaS: YesClosed product: Yes

LobeChat

Chat and agent platform for language models from several providers, with plug-ins and knowledge bases.

LicenseRef-LobeHub-CommunityCustom licence

SaaS: Yes, with conditionsClosed product: Unclear – verify

Cherry Studio

Desktop app for chatting with many different language models, with assistants and agents.

AGPL-3.0Network copyleft (AGPL)

SaaS: Yes, with conditionsClosed product: No

Text Generation WebUI

Web interface for running language models locally, with an OpenAI-compatible API.

AGPL-3.0Network copyleft (AGPL)

SaaS: Yes, with conditionsClosed product: No

Jan

Desktop app for chatting with language models locally, fully offline.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

KoboldCpp

Single-file program that runs GGUF language models locally with its own web interface.

AGPL-3.0Network copyleft (AGPL)

SaaS: Yes, with conditionsClosed product: No

HuggingChat UI

The code behind Hugging Face's HuggingChat, for running your own chat interface for language models.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

LM Studio

Desktop app to download and run local language models (LLMs), with chat and an OpenAI-compatible local server.

LicenseRef-LM-Studio-TermsFree, not open sourcenot verified

SaaS: NoClosed product: No

AI evaluation, red-teaming & auditing

iFixAi

Auditing of AI agents: evaluations, independent judges and scored reports.

Apache-2.0PermissiveYoung projectnot verified

SaaS: YesClosed product: Yes

promptfoo

Command-line tool for testing prompts, agents and RAG setups and for security testing AI apps.

MITPermissive

SaaS: YesClosed product: Yes

DeepEval

Python framework for evaluating language model output with metrics, much like unit tests.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Strands Evals

Framework for evaluating AI agents and LLM apps with test cases and graders.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI meeting notes & screen memory

Meetily

Meeting assistant that transcribes and summarises meetings fully locally on your computer, without a cloud service.

MITPermissive

SaaS: YesClosed product: Yes

Anarlog (formerly Hyprnote)

Local-first meeting notes app that transcribes and helps summarise with AI.

MITPermissiveOpen core

SaaS: YesClosed product: Yes

AI models: agent decisions and tool routing

strands-decider-2B-hobson-v19

Small model from Strands Agents that picks between options or rates on a scale in agent workflows, with a confidence score. Based on Qwen/Qwen3.5-2B-Base.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI models: computer vision

siglip-so400m-patch14-384

Image-text model from Google that matches images against text descriptions, for example for zero-shot classification.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

grounding-dino-base

Model from IDEA Research that finds objects in images from a text description.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

sam2-hiera-large

Segmentation model from Meta that outlines objects in images and video from clicks or boxes.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

dinov2-large

Vision model from Meta that extracts general image features for search, classification and other tasks.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI models: embeddings

all-MiniLM-L6-v2

Small, fast model that turns English sentences into vectors for semantic search and clustering. Based on nreimers/MiniLM-L6-H384-uncased.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

bge-m3

Multilingual embedding model from BAAI for search over long texts, with both dense and sparse vectors.

MITPermissive

SaaS: YesClosed product: Yes

multilingual-e5-large

Multilingual embedding model for semantic search and text similarity across about a hundred languages. The model card lists Swedish support.

MITPermissive

SaaS: YesClosed product: Yes

Qwen3-Embedding-0.6B

Small embedding model from the Qwen family for multilingual search and text similarity. Based on Qwen/Qwen3-0.6B-Base.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

nomic-embed-text-v1.5

Embedding model from Nomic for long texts, where the vector size can be chosen as needed.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

gte-multilingual-base

Multilingual embedding model from Alibaba for search and text similarity. The model card lists Swedish support.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

sentence-bert-swedish-cased

Swedish sentence model from KBLab at the National Library of Sweden for semantic search and text similarity. The model card lists Swedish support.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI models: image generation

FLUX.1-schnell

Fast model from Black Forest Labs that generates images from text in few steps.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Qwen-Image

Image generation model from Qwen that is good at rendering text in images and can edit images.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

HiDream-I1-Full

Large image generation model from HiDream that creates images from text descriptions.

MITPermissive

SaaS: YesClosed product: Yes

AI models: OCR and document parsing

DeepSeek-OCR

Model from DeepSeek that reads text and layout in document images and compresses them for language models.

MITPermissive

SaaS: YesClosed product: Yes

Florence-2-large

Vision model from Microsoft for captioning, object detection and OCR through text prompts.

MITPermissive

SaaS: YesClosed product: Yes

PaddleOCR-VL

Compact model from PaddleOCR that parses documents with text, tables and formulas in many languages. Based on baidu/ERNIE-4.5-0.3B-Paddle.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

dots.ocr

Model that parses document layout and reads text, tables and formulas in one step, in several languages.

MITPermissive

SaaS: YesClosed product: Yes

granite-docling-258M

Small model from IBM that turns document pages into structured text, for use with Docling.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

olmOCR-7B-0225-preview

Model from the Allen Institute for AI that turns PDF pages into clean text. Based on Qwen/Qwen2-VL-7B-Instruct.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

GOT-OCR-2.0-hf

OCR model from StepFun that reads text, tables, formulas and sheet music from images.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI models: reranking

bge-reranker-v2-m3

Multilingual model from BAAI that reorders search hits by how well they answer the query.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI models: speaker diarization

speaker-diarization-3.1

Model from pyannote that works out who speaks when in an audio recording.

MITPermissive

SaaS: YesClosed product: Yes

AI models: speech to text

KB-Whisper (large)

Swedish speech-to-text model from the National Library of Sweden.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

whisper-large-v3

Speech recognition model from OpenAI that transcribes and translates speech in many languages. The model card lists Swedish support.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

whisper-large-v3-turbo

Faster, smaller version of Whisper large-v3 for transcription with slightly lower accuracy. Based on openai/whisper-large-v3. The model card lists Swedish support.

MITPermissive

SaaS: YesClosed product: Yes

faster-whisper-large-v3

Whisper large-v3 converted to CTranslate2 format for faster transcription with faster-whisper. The model card lists Swedish support.

MITPermissive

SaaS: YesClosed product: Yes

distil-large-v3

Distilled version of Whisper that transcribes English faster with similar accuracy.

MITPermissive

SaaS: YesClosed product: Yes

AI models: text generation (language models)

Gemma 4

Google's open model family in several sizes.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Llama 4 Scout

Meta's multimodal model. Shown as an example of a licence that excludes EU companies.

LicenseRef-Llama-4-CommunityCustom AI licence

SaaS: Yes, with conditionsClosed product: Yes, with conditions

DeepSeek-R1

Large language model from DeepSeek trained for multi-step reasoning such as maths and code.

MITPermissive

SaaS: YesClosed product: Yes

Mistral-7B-Instruct-v0.3

Instruction-tuned language model from Mistral AI with 7 billion parameters and function calling. Based on mistralai/Mistral-7B-v0.3.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Qwen2.5-7B-Instruct

Instruction-tuned language model from Qwen with 7 billion parameters for chat, code and multilingual tasks. Based on Qwen/Qwen2.5-7B.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

phi-4

Smaller language model from Microsoft with 14 billion parameters, focused on reasoning.

MITPermissive

SaaS: YesClosed product: Yes

Qwen3-8B

Language model from Qwen with 8 billion parameters that can switch between quick answers and longer reasoning. Based on Qwen/Qwen3-8B-Base.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

GLM-4.5

Large language model from Z.ai (Zhipu) for reasoning, code and agent tasks.

MITPermissive

SaaS: YesClosed product: Yes

Mistral-Small-3.1-24B-Instruct-2503

Instruction-tuned model from Mistral AI with 24 billion parameters that handles both text and images. Based on mistralai/Mistral-Small-3.1-24B-Base-2503. The model card lists Swedish support.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

SmolLM3-3B

Small language model from Hugging Face with 3 billion parameters, with an openly published training recipe. Based on HuggingFaceTB/SmolLM3-3B-Base.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Qwen3-30B-A3B

Mixture-of-experts language model from Qwen: 30 billion parameters, of which about 3 billion are active per step. Based on Qwen/Qwen3-30B-A3B-Base.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

EuroLLM-9B-Instruct

European language model from the EU-funded UTTER project, trained on the official EU languages and a few more. Based on utter-project/EuroLLM-9B. The model card lists Swedish support.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

granite-3.3-8b-instruct

Instruction-tuned language model from IBM with 8 billion parameters for business tasks and reasoning. Based on ibm-granite/granite-3.3-8b-base.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

OLMo-2-1124-13B-Instruct

Fully open language model from the Allen Institute for AI where training data and code are published too. Based on allenai/OLMo-2-1124-13B-Instruct-RLVR2.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI models: text to speech

Kokoro-82M

Very small text-to-speech model that produces natural speech and runs on an ordinary computer. Based on yl4579/StyleTTS2-LJSpeech.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Dia-1.6B

Text-to-speech model from Nari Labs that generates multi-speaker dialogue from a script.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

VibeVoice-1.5B

Text-to-speech model from Microsoft for long multi-speaker conversations, such as podcasts.

MITPermissive

SaaS: YesClosed product: Yes

csm-1b

Speech model from Sesame that generates speech from text and conversational context.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

chatterbox

Text-to-speech model from Resemble AI with voice cloning and adjustable expressiveness. The model card lists Swedish support.

MITPermissive

SaaS: YesClosed product: Yes

piper-voices

Collection of ready-made voices for the Piper text-to-speech engine, in many languages. The model card lists Swedish support.

MITPermissive

SaaS: YesClosed product: Yes

AI models: translation

m2m100_418M

Translation model from Meta that translates directly between 100 languages without going through English. The model card lists Swedish support.

MITPermissive

SaaS: YesClosed product: Yes

madlad400-3b-mt

Translation model from Google trained on over 400 languages. The model card lists Swedish support.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

opus-mt-sv-en

Small translation model from the University of Helsinki that translates from Swedish to English.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

opus-mt-en-sv

Small translation model from the University of Helsinki that translates from English to Swedish.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI models: vision-language

Qwen2.5-VL-7B-Instruct

Vision-language model from Qwen that understands images, documents and video and answers in text.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

SmolVLM-Instruct

Small vision-language model from Hugging Face that describes and answers questions about images. Based on HuggingFaceTB/SmolLM2-1.7B-Instruct.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

AI research notebooks & answer engines

Open Notebook

Private NotebookLM-style AI research notebook you run yourself, with your choice of models.

MITPermissive

SaaS: YesClosed product: Yes

Khoj

Self-hosted AI assistant that answers from your own documents and the web, using local or cloud models.

AGPL-3.0Network copyleft (AGPL)

SaaS: Yes, with conditionsClosed product: No

Vane (formerly Perplexica)

AI-powered search engine that searches the web and summarises the answer with sources. Can run on your own server.

MITPermissive

SaaS: YesClosed product: Yes

GPT Researcher

Agent that does in-depth research on the web or in your own documents and compiles a report with sources.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

SurfSense

Research assistant that searches your own documents and sources with AI, an alternative to NotebookLM.

Apache-2.0PermissiveOpen core

SaaS: YesClosed product: Yes

Local Deep Research

Deep research tool that can run fully locally, with local or cloud models and several search sources.

MITPermissive

SaaS: YesClosed product: Yes

Image generation

ComfyUI

Node-based interface and backend for image generation with diffusion models, running locally.

GPL-3.0Strong copyleft

SaaS: YesClosed product: No

Diffusers

Python library from Hugging Face for running and training diffusion models for images, video and audio.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

InvokeAI

Creative tool for image generation with Stable Diffusion models, with a canvas for editing and inpainting.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Krita AI Diffusion

Plug-in for the Krita painting app that generates and edits images with AI inside the app.

GPL-3.0Strong copyleft

SaaS: YesClosed product: No

SwarmUI

Modular web interface for image generation with Stable Diffusion and similar models.

MITPermissive

SaaS: YesClosed product: Yes

LLM gateway & observability

LiteLLM

AI gateway that calls many LLM providers through one API, with cost tracking.

MITPermissiveOpen core

SaaS: YesClosed product: Yes

Langfuse

Tracing, evaluation and prompt management for LLM applications.

MITPermissiveOpen core

SaaS: YesClosed product: Yes

Opik

Platform for tracing, evaluating and monitoring LLM apps, RAG systems and agents.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Arize Phoenix

Tool for tracing, evaluating and debugging AI and LLM applications.

Elastic-2.0Source-available (not OSI)

SaaS: NoClosed product: No

Helicone

Observability platform for LLM calls: logs, costs and evaluation, often through a proxy.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Coze Loop

Platform for developing, debugging, evaluating and monitoring AI agents across their lifecycle.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Agenta

Workspace for teams to build, test and evaluate LLM apps and AI agents.

MITPermissiveOpen core

SaaS: YesClosed product: Yes

Laminar

Observability platform for AI agents with tracing of runs and evaluations.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

LLM inference & serving

Ollama

Run language models locally on CPU or GPU with one command.

MITPermissive

SaaS: YesClosed product: Yes

vLLM

High-throughput serving of language models in production.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

llama.cpp

C/C++ engine for running language models locally, even without a GPU.

MITPermissive

SaaS: YesClosed product: Yes

LocalAI

Self-hosted drop-in for the OpenAI API that runs language, image and voice models on your own hardware.

MITPermissive

SaaS: YesClosed product: Yes

SGLang

Fast framework for serving large language models and multimodal models in production.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Triton Inference Server

Inference server from NVIDIA for running AI models from several frameworks in the cloud or at the edge.

BSD-3-ClausePermissive

SaaS: YesClosed product: Yes

Xinference

Platform for running open language, speech and multimodal models behind one API, locally or in the cloud.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

LMDeploy

Toolkit for compressing, deploying and serving language models efficiently.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

lms (LM Studio CLI)

Command-line tool for LM Studio: load models and control the local server from the terminal.

MITPermissive

SaaS: YesClosed product: Yes

LM Studio TypeScript SDK

TypeScript library for using local models in LM Studio from your own programs.

MITPermissive

SaaS: YesClosed product: Yes

LM Studio MLX engine

The engine LM Studio uses to run models with Apple MLX on Mac.

MITPermissive

SaaS: YesClosed product: Yes

LM Studio Python SDK

Python library for using local models in LM Studio from your own programs.

MITPermissiveLow activity

SaaS: YesClosed product: Yes

Machine translation

LibreTranslate

Machine translation API that runs fully locally. Quality is often lower than DeepL – test it on your own texts.

AGPL-3.0Network copyleft (AGPL)

SaaS: Yes, with conditionsClosed product: No

Argos Translate

Python library and app for machine translation that works offline.

MITPermissive

SaaS: YesClosed product: Yes

CTranslate2

Fast inference engine for Transformer models, often used for translation and speech recognition.

MITPermissive

SaaS: YesClosed product: Yes

Firefox Translations (training)

The code and training pipeline behind the translation models in Firefox, which translate locally in the browser.

MPL-2.0Weak copyleft

SaaS: YesClosed product: Yes, with conditions

Model training, fine-tuning & data labelling

Unsloth

Tool for fine-tuning and running language models faster and with less memory, with a local interface.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

LLaMA-Factory

Framework for fine-tuning over 100 language and vision-language models, with a web interface.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Ray

Distributed computing framework that scales Python and ML workloads from one machine to a cluster.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Label Studio

Tool for labelling and annotating data (text, images, audio, video) for machine learning.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

MLflow

Platform for tracking experiments, managing models and evaluating ML and LLM applications.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

CVAT

Tool for annotating images and video and building computer vision datasets.

MITPermissive

SaaS: YesClosed product: Yes

Axolotl

Tool for fine-tuning language models with configuration files, supporting several training methods.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

OCR & document AI

MarkItDown

Python tool from Microsoft that converts Office documents, PDFs and other files to Markdown.

MITPermissive

SaaS: YesClosed product: Yes

PaddleOCR

OCR toolkit that reads text and structure from images and PDF files, with support for many languages.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Tesseract

Established OCR engine that recognises text in images, with support for over 100 languages.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Docling

Converts PDFs, Office files and images into structured text and tables that AI models can use.

MITPermissive

SaaS: YesClosed product: Yes

Marker

Converts PDF files to Markdown and JSON, including tables and equations.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

OCRmyPDF

Adds a searchable text layer to scanned PDF files using OCR.

MPL-2.0Weak copyleft

SaaS: YesClosed product: Yes, with conditions

olmOCR

Toolkit that turns PDF pages into clean, readable text using a language model.

Apache-2.0PermissiveLow activity

SaaS: YesClosed product: Yes

Unstructured

Library that turns documents in many formats into structured data for language models and search.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

docTR

Python library for deep-learning OCR: detects and reads text in document images.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

RAG & enterprise AI search

PageIndex

Vectorless RAG: builds a tree index over documents and lets the language model reason its way to the answer.

MITPermissive

SaaS: YesClosed product: Yes

Hindsight

Memory for AI agents that learns over time.

MITPermissiveOpen corenot verified

SaaS: YesClosed product: Yes

RAGFlow

RAG engine that ingests documents, searches them and lets language models answer with citations.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

LlamaIndex

Framework for connecting your own documents and data to language models, for RAG and agents among other things.

MITPermissive

SaaS: YesClosed product: Yes

GraphRAG

Method and library from Microsoft for RAG that builds a knowledge graph from the texts before searching.

MITPermissive

SaaS: YesClosed product: Yes

Onyx

AI chat and search across company documents and tools, for any language model.

MITPermissiveOpen core

SaaS: YesClosed product: Yes

WeKnora

Knowledge platform that makes documents searchable with RAG and an AI agent.

MITPermissive

SaaS: YesClosed product: Yes

Haystack

Python framework for building LLM apps as pipelines, with search, RAG and agents.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

kotaemon

Web interface for chatting with your own documents via RAG, with support for several models.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Speech-to-text & dictation

OpenWhispr

Dictation (voice to text) with local models or your own API keys.

MITPermissive

SaaS: YesClosed product: Yes

Whisper

Speech recognition model from OpenAI that transcribes and translates speech in many languages.

MITPermissive

SaaS: YesClosed product: Yes

whisper.cpp

Fast C/C++ port of Whisper for transcription on ordinary computers, mobiles or embedded devices.

MITPermissive

SaaS: YesClosed product: Yes

Handy

Desktop dictation app that turns speech into text fully offline.

MITPermissive

SaaS: YesClosed product: Yes

faster-whisper

Faster variant of Whisper built on CTranslate2, with lower memory use.

MITPermissive

SaaS: YesClosed product: Yes

WhisperX

Whisper-based transcription with word-level timestamps and speaker separation.

BSD-2-ClausePermissive

SaaS: YesClosed product: Yes

Buzz

Desktop app that transcribes and translates audio offline with Whisper.

MITPermissive

SaaS: YesClosed product: Yes

FunASR

Speech recognition toolkit: training, streaming recognition, punctuation and speaker diarization.

MITPermissive

SaaS: YesClosed product: Yes

NVIDIA NeMo Speech

NVIDIA's framework for training and running speech AI such as speech recognition and text to speech.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

sherpa-onnx

Speech to text, text to speech and speaker diarization that run offline, including on mobile and embedded devices.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Vibe

Desktop app that transcribes audio and video locally on your computer.

MITPermissive

SaaS: YesClosed product: Yes

Speaches

Server for speech to text and text to speech with an OpenAI-compatible API.

MITPermissive

SaaS: YesClosed product: Yes

Text-to-speech & voice cloning

GPT-SoVITS

Voice cloning and text-to-speech tool that can be trained on a short voice recording.

MITPermissive

SaaS: YesClosed product: Yes

VoiceStudio

Local app for voice cloning, text to speech, dubbing and transcription.

AGPL-3.0Network copyleft (AGPL)Young project

SaaS: Yes, with conditionsClosed product: No

ChatTTS

Text-to-speech model tuned for everyday dialogue, mainly Chinese and English.

AGPL-3.0Network copyleft (AGPL)

SaaS: Yes, with conditionsClosed product: No

Chatterbox

Text-to-speech model from Resemble AI with voice cloning and adjustable expressiveness.

MITPermissive

SaaS: YesClosed product: Yes

IndexTTS

Text-to-speech system that can mimic a voice from a short sample and control pacing and expression.

LicenseRef-bilibili-Model-UseCustom AI licence

SaaS: Yes, with conditionsClosed product: Yes, with conditions

CosyVoice

Multilingual text-to-speech model with code for inference, training and deployment.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

F5-TTS

Research code for the F5-TTS text-to-speech model, which can mimic a voice from a short sample.

MITPermissive

SaaS: YesClosed product: Yes

Piper

Fast local text to speech that runs well on small computers such as a Raspberry Pi.

GPL-3.0Strong copyleft

SaaS: YesClosed product: No

Coqui TTS (Idiap fork)

Maintained continuation of Coqui TTS, a deep-learning toolkit for text to speech.

MPL-2.0Weak copyleft

SaaS: YesClosed product: Yes, with conditions

Vector database

Milvus

Vector database for fast similarity search over large numbers of embeddings, built to scale in cloud environments.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Qdrant

Vector database and search engine for similarity search with filtering, written in Rust.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Chroma

Embedding database and search layer used to give AI apps and agents access to your own data.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

pgvector

PostgreSQL extension for storing vectors and running similarity search.

PostgreSQLPermissive

SaaS: YesClosed product: Yes

Weaviate

Vector database that stores objects and vectors and combines vector search with filters.

BSD-3-ClausePermissiveOpen core

SaaS: YesClosed product: Yes

LanceDB

Embedded vector database for multimodal data that runs inside your app without a separate server.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Vespa

Platform for search, recommendations and vector search at scale, with ranking at query time.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Voice agents

Pipecat

Python framework for voice agents and real-time apps that combine speech recognition, a language model and text to speech.

BSD-2-ClausePermissive

SaaS: YesClosed product: Yes

LiveKit Agents

Framework for building AI agents that talk and listen in real time over LiveKit.

Apache-2.0Permissive

SaaS: YesClosed product: Yes

Other

Hexabot

Platform for chatbots and AI workflows with agents, actions and conversational channels.

LicenseRef-FCL-1.0-ALv2Source-available (not OSI)Young projectnot verified

SaaS: NoClosed product: No

Dyad

Local AI app builder – build web apps with AI on your own machine.

Apache-2.0PermissiveOpen core

SaaS: YesClosed product: Yes

Orca

Agent development environment for running several coding agents (Claude Code, Codex and more) in parallel git worktrees, with terminals, diffs and SSH.

MITPermissivenot verified

SaaS: YesClosed product: Yes