Open Source Companies

Mistral AI

Paris, France

Mistral AI is a French AI company developing efficient open-weight large language models. Known for Mistral 7B, Mixtral, and enterprise solutions.

TIER IV

Nagoya, Japan

TIER IV is a Japanese deep-tech company and founding member of the Autoware Foundation, maintaining Autoware, the world's first all-in-one open-source software stack for autonomous driving. It develops Level 4 self-driving technology and reference vehicles.

Hugging Face

New York, United States

Hugging Face is the AI community building the future with open source machine learning. Hosts the world's largest collection of ML models and datasets.

Anyscale

San Francisco, United States

Anyscale provides a platform built on Ray, an open-source distributed computing framework, enabling developers to scale Python and machine learning applications from single laptops to large-scale clusters. Their key innovation lies in Ray’s ability to unify compute for the entire AI lifecycle – from data processing and model training to serving and reinforcement learning – with features like automated dependency management and distributed workload observability. Notably, Anyscale has partnered with Microsoft to deliver a first-party, fully managed Ray service on Azure, currently in private preview, and serves customers building and scaling demanding AI applications including large language models.

H2O.ai

Mountain View, United States

H2O.ai develops end-to-end AI platforms, including the open-source H2O and the automated machine learning platform Driverless AI, now extended with generative AI capabilities. Their technology uniquely focuses on enabling private and secure deployments of Large Language Models (LLMs) and Small Language Models (SLMs) – including on-premise and air-gapped environments – giving organizations full ownership of their AI stack. H2O.ai has demonstrated success with major enterprises like Australia’s largest bank, achieving a 70% reduction in fraud, and AT&T, who saw a 2x return on investment in free cash flow through the implementation of h2oGPTe for call center transformation.

Together AI

San Francisco, United States

Together AI delivers a cloud platform specializing in the deployment and scaling of open-source large language models (LLMs) and multimodal AI models, offering both API access and direct GPU cluster access. Their platform distinguishes itself through optimized inference and training on performance-focused GPU clusters, enabling customers to process trillions of tokens and supports models for diverse applications including chat, image generation, and code completion. Notably, Together AI provides OpenAI-compatible APIs, facilitating easy migration from closed-source models and offers a developer-friendly experience for building AI-native applications.

01.AI

Beijing, China

01.AI develops and releases open-weight large language models (LLMs), most notably the Yi series, with a focus on strong multilingual capabilities and efficient deployment. Their flagship model, Yi-Lightning, utilizes a Mixture-of-Experts (MoE) architecture to achieve state-of-the-art performance, and the company emphasizes open-source accessibility for developers and researchers. Founded by Kai-Fu Lee, 01.AI aims to drive innovation in the "AI 2.0" era by providing foundational models and fostering a robust ecosystem around their technology.

Stability AI

London, United Kingdom

Stability AI develops open generative models across image, video, audio, and language, best known for the Stable Diffusion family of text-to-image models.

PostHog

San Francisco, United States

PostHog is an open-source product analytics platform with session replay, feature flags, A/B testing, and surveys. Used by 50,000+ companies.

Replicate

San Francisco, United States

Replicate provides a cloud platform and API for deploying and scaling open-source machine learning models, currently hosting over 8,000 models including Stable Diffusion, Llama 2, and ControlNet. Their key innovation lies in containerizing models with Docker and executing them on scalable cloud infrastructure, simplifying the process of model serving and eliminating the need for users to manage infrastructure. Replicate serves a diverse user base of developers, researchers, and businesses seeking to integrate pre-trained AI capabilities into their applications, and recently achieved significant traction with its support for rapidly deploying and scaling generative AI models.

Union.ai

Bellevue, United States

Union.ai is the enterprise platform behind Flyte, the open-source orchestrator for machine learning and data workflows. Its tooling helps teams build, schedule, and scale reproducible AI pipelines on Kubernetes. Flyte has been downloaded over 80 million times.

Featherless AI

San Francisco, United States

Featherless AI is a serverless AI inference platform that lets developers run thousands of open-source models, especially large language models, via API without managing servers or infrastructure. It integrates with the Hugging Face ecosystem to deploy and scale models for language, vision and audio applications.

Continue

San Francisco, United States

Continue is an open-source AI code assistant platform that integrates into VS Code and JetBrains IDEs. With 20K+ GitHub stars, it allows developers to create custom AI assistants using any model.

Dify

San Francisco, United States

Dify is an open-source LLM app development platform. Provides visual workflow builder, RAG pipeline, prompt IDE, and API backend for building AI applications.

All Hands AI

Boston, United States

All Hands AI develops OpenHands, the largest open-source platform for AI-driven software development, providing autonomous coding agents and an SDK for building agentic developer workflows. The MIT-licensed project acts as an AI pair programmer that handles tasks like writing tests, managing dependencies and updating documentation.

Meta AI

Menlo Park, United States

Meta AI (formerly FAIR) is Meta's AI research division. Creator of LLaMA, PyTorch, and leading research in computer vision, NLP, and embodied AI.

MLflow (Databricks)

San Francisco, United States

MLflow is an open-source platform for managing the machine-learning lifecycle—experiment tracking, reproducible runs, model packaging and deployment—that works with any ML library or language. Created by Matei Zaharia and launched by Databricks in 2018; later contributed to the Linux Foundation.

StabilityAI Japan

Tokyo, Japan

Stability AI Japan focuses on adapting and deploying open-source generative AI models for the Japanese market, with a core offering centered around Japanese-optimized versions of Stable Diffusion for image generation. They provide enterprise-ready multimodal tools – including image, 3D, and 4D video generation capabilities accessible through platforms like DreamStudio – emphasizing brand safety, customization, and scalable deployment for professional creative workflows. As a regional arm of Stability AI, they aim to empower businesses with AI-driven content creation, offering solutions from initial concept and storyboarding through to final production and asset management.

Monai

Santa Clara, United States

Monai is an open-source, PyTorch-based framework designed to accelerate the development and deployment of AI-powered medical imaging solutions. It provides a comprehensive suite of tools – including pre-processing, model training, and evaluation – specifically tailored for analyzing modalities like MRI, CT, and X-ray images. Developed jointly by NVIDIA and King's College London, Monai is gaining traction within the research community and is being utilized in projects focused on disease detection, segmentation, and image registration, with a growing ecosystem of contributed workflows and models.

OpenMined

Oxford, United States

OpenMined develops open-source tools and protocols for privacy-preserving machine learning, with a focus on enabling federated learning and secure multi-party computation. Their core innovation, Attribution-Based Control, aims to unlock access to previously siloed non-public data by allowing data owners to maintain control and receive attribution for its use in AI models. As a non-profit foundation, OpenMined is actively involved in initiatives like the National AI Research Resource (NAIRR) pilot, seeking to expand data access for research while upholding privacy standards.

Genmo

San Francisco, United States

Genmo is a US-based AI research lab focused on developing open-source video generation models. Their core product is Mochi 1, a state-of-the-art text-to-video model designed for local deployment and customization via open-source repositories like ComfyUI. Genmo targets researchers, developers, and creatives interested in contributing to and utilizing advanced, accessible AI video generation technology, positioning themselves within the emerging field of world model development for AGI research.

Cline

San Francisco, United States

Cline develops an open-source AI coding agent focused on terminal-first workflows and utilizing “Plan/Act” modes for task execution. Their core technology centers around autonomous AI agents capable of writing and executing code, demonstrated through competitive “Thunderdome” environments pitting agents against each other. With reported adoption by over 8 million developers, Cline targets software engineers seeking locally-run, customizable AI coding assistance.

LanceDB

San Francisco, United States

LanceDB develops an AI-native data lakehouse designed to unify data management for the entire machine learning lifecycle. Their core product is a single, searchable table format optimized for raw data, embeddings, and features, eliminating traditional data pipelines and synchronization challenges. Targeting AI infrastructure and MLOps teams, LanceDB aims to streamline data access and processing for large-scale model training and retrieval-augmented generation (RAG) applications, with a focus on GPU utilization efficiency.

Letta

San Francisco, United States

Letta develops infrastructure for building and deploying autonomous AI agents with a focus on persistent memory capabilities. Their core product is a memory-first agent platform enabling continuous learning and stateful operation across various models and infrastructure providers. Founded in 2023, Letta targets developers seeking to create highly personalized and evolving AI agents, with features like memory visualization and portability as key differentiators.

Mem0

San Francisco, United States

Mem0 develops a persistent memory layer designed to enhance the performance of AI agents and applications. Their core product is a drop-in infrastructure enabling continuous learning from user interactions by storing and retrieving contextual data. Founded in 2024, Mem0 targets developers building production-ready AI agents requiring robust, persistent context management.

Prime Intellect

San Francisco, United States

Prime Intellect provides a full-stack MLOps platform focused on agentic AI models. Their core offering is integrated infrastructure for training, evaluating, and deploying these models, encompassing compute, reinforcement learning post-training, simulated environments, and benchmarking tools. Targeting AI developers and researchers, the company emphasizes open-source compatibility and aims to simplify the lifecycle of self-improving agents.

Arcee AI

San Francisco, United States

Arcee AI develops and deploys open-weight large language models, focusing on rapid iteration and cost-efficient architectures. Their core offering is the “Trinity” model family, releasing three frontier-level models within a six-month period, and utilizes online reinforcement learning for continuous improvement. Arcee AI targets developers, researchers, and the open-source AI community, providing both API access and a grant program to foster adoption of their models.

Tracecat

San Francisco, United States

Tracecat develops an open-source Security Orchestration, Automation, and Response (SOAR) platform leveraging AI agents to automate incident response. Their core product is an AI-native SOAR enabling users to build, manage, and deploy custom AI agents and workflows for security operations. Targeting security teams and developers working with AI-driven security solutions, Tracecat launched in 2024 with a focus on providing an open-source alternative in the emerging AI-powered SOAR space.

Pleias

Paris, France

Pleias develops large language models specifically optimized for energy efficiency and deployment within data-sensitive sectors. Their core offering centers on a data layer technology designed to enhance LLM performance, rather than the models themselves. Founded in 2023, Pleias targets industries where data governance and computational cost are critical concerns, such as finance and healthcare.

Label Studio

San Francisco, United States

Label Studio provides an open-source data labeling and AI evaluation platform supporting a wide range of data types including text, image, audio, and video. Their core offering centers on a customizable labeling interface and tooling for human-in-the-loop AI workflows, with increasing focus on evaluating Large Language Models (LLMs) and Reinforcement Learning from Human Feedback (RLHF). The company targets AI/ML teams building and deploying models across diverse applications, and has gained traction within the open-source community and enterprise adoption for data quality and model improvement.

Ultralytics

Frederick, United States

Ultralytics provides a comprehensive computer vision platform centered around the YOLO (You Only Look Once) family of real-time object detection models. Their core offering is an integrated solution for data annotation, cloud-based model training—leveraging scalable GPU infrastructure—and deployment to edge, mobile, and cloud environments. The company targets developers and businesses requiring custom computer vision solutions, demonstrated by their support for deployment across 43 global regions and various export formats.

ModelBest

Beijing, China

ModelBest develops and distributes open-source large language models (LLMs) primarily for the Chinese market. Their core product is the MiniCPM series of LLMs, designed for efficient deployment and accessibility with a focus on both research and commercial applications. Founded in 2022, ModelBest aims to lower the barrier to entry for utilizing advanced AI through readily available, locally-developed models.

Kyutai

Paris, France

Kyutai is an AI research lab focused on developing open-source components for artificial general intelligence, with a current emphasis on voice AI. Their core technology includes compact, high-quality text-to-speech (Pocket TTS) and speech-to-text models optimized for low-latency, real-time applications—demonstrated by products like Unmute and the speech-native dialogue system, Moshi. Kyutai targets developers and researchers seeking accessible, performant voice AI building blocks, and is actively commercializing research through spin-off ventures like Gradium.

Pruna AI

Munich, Germany

Pruna AI develops and deploys optimized generative image models, focusing on reducing computational costs and model size without sacrificing quality. Their core technology centers around efficient model architectures and deployment pipelines for image generation, leveraging open-source principles. Targeting businesses requiring high-volume, cost-effective image creation, Pruna AI recently launched a suite of performance-focused open-weight image generation models.

Onyx

San Francisco, United States

Onyx develops an open-source enterprise search platform leveraging retrieval-augmented generation (RAG) to enable AI-powered information access. Their core product is an AI assistant connected to a company’s internal data sources, facilitating knowledge discovery and improved team productivity. Founded in 2023, Onyx targets organizations seeking self-hosted, customizable AI search solutions and emphasizes open-source flexibility.

Linagora

Paris, France

Linagora develops and publishes open-source software and Software-as-a-Service (SaaS) solutions for enterprise digital transformation. Their core offering centers around large language models, specifically focusing on sovereign AI and data control for businesses—including the open-source LLM, OpenLLaMA. Targeting the European market with a focus on data privacy and security, Linagora positions itself as a provider of alternatives to dominant, non-European AI platforms.

AkylAI is a Kyrgyzstan startup that built the first conversational AI assistant and language model for the Kyrgyz language, collecting a corpus of over 300 million Kyrgyz words. Developed by The Cramer Project with support from the High-Tech Park of the Kyrgyz Republic and EBRD's Star Venture programme, it released a Kyrgyz-language chatbot and open-source speech models. The team formed in 2022 and released its beta in 2023.