THE RISING STARS of OPEN SOURCE 2025

Introduction

Open Source companies operate within a distinctive ecosystem defined by transparent collaboration, active community participation, and unconventional go-to-market strategies. Unlike proprietary software vendors – who monetise through closed licensing models from day one – open source companies often delay commercialisation until their projects have achieved significant technical maturity and organic adoption within a passionate community of contributors and users.

This dynamic poses a unique challenge for investors: how do you evaluate an open source project before it generates meaningful revenue? More specifically, how can you identify signs of early momentum and longterm potential purely from community dynamics?

At Oxx, we believe the answer lies in quantifying early-stage Product Community Fit. Each year, we conduct a deep-dive analysis of the open source ecosystem to uncover high-potential projects that are primed to become scalable, fundable, and ultimately sustainable businesses. Our methodology uses a combination of GitHub stars, forks, and search traffic with deeper engagement metrics such as contributor velocity, issue throughput, and release cadence.

This allows us to detect early traction well before commercial models kick in. By mapping these signals across projects, we spotlight the “Rising Stars of Open Source”: projects that have compelling technical depth as well as the grassroots momentum that predicts future breakout success.

Since launching this analysis in 2021, companies we’ve identified through this methodology have gone on to raise over $3.7bn in venture capital. The signals for the next generation of rising stars may already be visible – if you know where to look…!

Since the publication of the last report, we have seen…

We have also seen several big rounds amongst the 2023 cohort:

Themes

Geography

Global, but not evenly distributed.

From 2021 to 2024, open source innovation has had a clear geographic heartbeat. The United States dominates the landscape – unsurprising given the critical mass of developers, startups, and funding. But look closer, and you’ll spot ecosystems of software development like Germany, the UK, and France serving as fertile ground for the next generation of rising stars.

Notably, Canada, the Netherlands, and New Zealand continue to punch well above their size, often driving deep-tech innovation.

What’s interesting is that this doesn’t just say where code is written, it spotlights from where the next generation of unicorns may emerge.

Understanding Product Community Fit

Entrepreneurs and investors have long relied on signals like Product Market Fit (PMF) and, more recently, Go-to-Market Fit (GTMF) to gauge startup potential. But in the world of open source, those traditional milestones often arrive in reverse. Projects can achieve widespread adoption and deep user affinity before there’s any commercial strategy in place – or even a monetisation model.

That’s why we focus on an earlier, more relevant signal for open source: Product Community Fit (PCF).
PCF measures how tightly a project resonates with its community of users and contributors – who, in open source, often are the market. When a product evolves in lockstep with its community’s real-world needs, it doesn’t just build software – it builds trust, loyalty, and technical credibility.

A strong signal of PCF? A steadily growing, high-quality contributor base. Metrics like commit velocity, pull request activity, issue engagement, and release frequency reveal that users aren’t just consuming, they’re co-creating software. This creates a positive feedback loop: each contribution sharpens the product, amplifies visibility, and lowers the barrier to adoption. The best open source projects become magnets for developer energy. And when that happens, commercial readiness follows suit.

How we identify the Rising Stars

At Oxx, we prioritise depth of engagement over surface-level popularity. Metrics like GitHub stars or download counts may indicate awareness, but these metrics don’t tell the full story of who’s actually contributing to the evolution of the product. For that, we track a far more telling metric: Monthly Active Contributors – developers actively committing code, raising or closing issues, reviewing pull requests, and engaging in discussions.

We’ve built a data pipeline that leverages GitHub’s public events API and enriches with additional data sources, parsing historical project activity across the open source universe. Our methodology has been backtested against 175,000+ projects, spanning 15 years of contribution history. The signal is clear: when a project crosses the threshold of 50–100 active contributors, it often tips into a phase of exponential growth –technically, communally, and eventually, commercially.

Why does this matter? Because high contributor activity isn’t just a vanity metric. It reflects organic governance, distributed ownership, and deep technical buy-in: the very foundations of scalable, defensible open source businesses.
By mapping these patterns of genuine engagement, we’re able to surface high-potential projects long before they show up in VC deal flow. These are the ones we call the Rising Stars of Open Source – projects with not just promise, but clear growth trajectory.

Where the Rising Stars operate: Mapping the Open Source arena

Looking back at our previously-identified cohorts of Rising Stars, we observe some thematic “stalwarts”, in addition to the emergence of new thematic trends in 2025, symbolised by the “modernisers”.

Stalwarts

End-user applications

Open-source applications aimed primarily at non-developer end-users remain a stalwart category in open source software. Unsurprisingly, this theme has the widest range of user personas of any category identified. User personas from marketers and technical writers to sales teams and supplier managers benefit from the software in this category.

2021’s cohort featured an open-source Google Calendar alternative (Cal.com) and supply chain management platform (Tauri). 2024’s cohort contains a CRM (TwentyHQ) and a CMS (PayloadCMS). Between the two, we saw another open-source CMS (Builder.io) and a documentation-writing platform (Tiptap, a.k.a. überdosis) enjoy significant growth.

Open-source applications for non-developers have noticeably modernised since 2021. We’ve seen an increasing focus on data capture, which was absent from earlier cohorts. TwentyHQ and PayloadCMS are optimised to extract and label data from customer records and web content, and thus capitalise on the importance of proprietary data assets for improved AI integration. Fundamentally though, the advantages of building an open-source product for non-technical users remain the same.

It enables companies to source feedback and iterate quickly, from both technical and non-technical customers, and to build a selling point on the basis of transparent product roadmaps, accelerated development timelines, and proximity to the customer.

Low-code

Low-code and no-code platforms are represented in both our 2021 cohort (Appsmith, Budibase) and 2024 cohort (ToolJet, FlowFuse). These platforms use simple, intuitive interfaces to enable users to launch and control technical processes – such as building their own applications – with either minimal or no coding required.

Rapid advancement in large language models’ (LLMs) abilities to generate and understand code has made product development much more accessible for non-technical users. Where building an application with a drag-and-drop interface previously entailed a lower degree of control and customisability, it is now possible for LLMs to act as a translation layer between natural language and code. LLMs can output users’ requirements as code and convey any code-level constraints back to users, giving non-technical professionals an unprecedented understanding of ‘under-the-hood’ workflows. This makes ambitious no-code projects much more feasible than they have been in the past.

With (normally) non-technical users as their focus, the function of low-code open-source tools is somewhat similar to that of end-user applications: drawing from a user community with a wide range of technical skill levels, bridging the gap between technical and non-technical teams. 

Across our cohorts, low-code tools have been rolling out new features which put AI application-building front and centre. Appsmith, Budibase and ToolJet all contain modules to facilitate connecting to LLMs  and embedding them in applications. ToolJet brings a distinctly 2024 approach, adding the option to integrate new features into applications via the now-familiar ‘AI chat’ widget. Budibase has always allowed users to set up rules to trigger automated actions – but now supports adding LLMs into these workflows, so that users can choose what actions trigger an LLM response.

Modernisers

Data connectors

Companies in this category deal with data-sharing, streaming, and synchronisation: a problem as old as software itself. Data connectors before the GenAI boom focused on the process of data integration and synchronisation from applications to warehouses. Their differentiators included their ability to provide customisable and automated data pipelines (Airbyte) or optimise for real-time data streaming (Redpanda).

AI-readiness has placed a whole new set of demands on data connectors, which companies including the aforementioned three – all from the 2022 cohort – have adapted to. These companies are increasingly focused on enriching data with relevant context, providing data in formats and structures that are easily consumable by LLMs, and even exposing their APIs to LLMs more easily. For example, both Redpanda and Airbyte now offer a pipeline for directly loading data into vector databases, so that LLMs are able to retrieve it. This functionality brings them in line with some of the earlier stage companies from the 2024 cohort with sophisticated LLM contextualisation offerings, such as Kestra.

Code editing and generation

Code editors from earlier cohorts, such as Warp (2022), Semgrep (2022), and Arch (2021), focused on enhancing core developer workflows. These tools provided features like advanced codebase search and code collaboration tools, all aimed at optimising traditional code writing and project management – for developer teams who were relying only on their brains, Stack Overflow, and documentation if they were lucky. Going open-source allowed them to build a trusting, loyal community of developer users – and a tight feedback loop.

The crop of AI-first code editors from our 2023 cohort – Cursor and Zed.dev – are integrating AI directly into their core functionality. This goes beyond simple code completion, encompassing features like AI-powered code generation, natural language-to-code translation, and intelligent debugging assistance. These editors are moving towards AI-driven contextual understanding, allowing them to comprehend and manipulate code with a level of semantic awareness previously unattainable. Moreover, there’s a growing emphasis on making the editors themselves programmable via LLMs, enabling developers to create custom AI-powered extensions and workflows. Code editors are shifting from tools primarily designed to assist human coding to active collaborators in the development process.

Databases

Database and backend management systems have always been among the fastest-growing open-source companies – and this is another category where AI-driven workflows have created a new set of feature requirements. As with data connectors, contextualising data for LLMs has become an important part of a database’s role.
Appwrite and Supabase (both 2021), primarily backend-as-a-service providers, have expanded their offerings to include vector storage and search capabilities.

Younger database companies from the 2023 cohort have a focus on AI-native features. DragonflyDB emphasises high-speed, in-memory data processing, which is crucial for low-latency AI applications.

Qdrant specialises in vector similarity search, enabling efficient retrieval of semantically similar data. This is helpful for use cases like chatbots, where an AI model must be able to quickly search a knowledge base and find the information most relevant to a user’s query.

SurrealDB, a multi-model database, offers the flexibility to handle diverse data types and query patterns – including unstructured data like text documents, which is newly accessible thanks to LLMs. These newer databases are designed from the ground up to support vector embeddings, similarity searches, and other AI-specific workloads.

LLMOps (vs NLP)

Natural language processing (NLP) software companies have been among our Rising Stars since the first cohort. These companies applied specialist statistical learning and annotation techniques in combination with linguistic expertise to deal with language data. This was necessary for use cases that required particularly good language comprehension, such as chatbot design or machine translation.

But unsurprisingly, LLMs have revolutionised the methods used to work with language data. Much less rule-based engineering is now needed, as LLMs are excellent at processing language in most situations.

Earlier cohorts show us how the rise of LLMs has caused NLP companies to shift their focus from processing text to facilitating a wide variety of workflows involving LLMs, from fine- tuning to analytics to deployment.
This new emphasis on LLM workflow- building has given rise to a software category of its own: LLMOps.

For example, Botpress (2021) started out as a chatbot-building platform with a conversational flow management system. This was in the days where conversational AI design mostly referred to writing rules to determine how a chatbot would handle particular scenarios – as they lacked the generalisation capabilities of LLMs. Botpress is now a ‘development environment for AI agents’. Instead of a rule-writing interface, it is a data-connection interface, allowing users to integrate their own knowledge bases, CRMs and other data sources with LLMs and launch them in their own environments for custom use cases. Newer entrants to the Rising Stars list, such as Flowise (2023) and LiteLLM (2024), have remarkably similar value propositions.

2023 cohort 
highlights

Two companies from the 2023 cohort have raised over $100M since they were identified, with one of them – Cursor – becoming a unicorn in January 2025.

Cursor

Cursor is a code editor that incorporates AI tools to assist in the programming process. It offers features such as multi-line code completion, AI-driven chat for code-related queries, and instant bug detection. These tools aim to streamline coding tasks and enhance productivity. The editor also includes an agent mode that can generate and execute terminal commands, providing developers with intelligent code suggestions and real-time error fixes.

Cursor has made headlines for becoming “the fastest growing SaaS in history”, exploding to over $100M in ARR in 12 months. The $105M Series B which gave it its unicorn valuation was raised just 4 months after a $60M Series A at a $400M valuation. In June 2025, Cursor raised $900 million at a $9.9 billion valuation, marking their third fundraise in less than a year.

Cursor’s go-to-market strategy has made extensive use of the open source community. Many early adopters contributed to high-profile open-source projects using Cursor. The ‘Edited with Cursor’ footer automatically added to pull requests reportedly drove a third of new signups.

Suno

Suno.ai is a platform that uses generative AI to create music from text prompts. It can produce songs with vocals, instrumentation, and lyrics. Suno.ai’s capabilities include generating tracks with customisable elements such as tempo, instruments, and melodies. The platform has been integrated into various tools, including Microsoft Copilot, to facilitate AI-driven music creation. Suno.ai’s features allow users to explore different musical ideas and bring them to life through its interface.

Since it was identified as a rising star in 2023, Suno’s addition to Copilot skyrocketed it to 25 million users. The most recent funding round valued the company at approximately $500 million.

Thematic trends in Rising Stars

Evolving from direct output to integrated intelligence
Both Cursor and Suno place direct content generation at the centre of their offerings. They are not the only companies from the 2023-24 cohort to have bet big on AI-augmented direct output: alongside them sit Zed.dev, another AI code editor, and LangGenius, a GPT 4-powered AI app builder. After ChatGPT’s public release in late 2022, founders spent 2023 developing products to make the most of AI’s vastly improved content generation capabilities – and this is reflected in the products that came to market.



2024 saw a shift: excitement about the raw power of AI-generated content gave way to a more sober focus on its potential shortcomings. Much of the most exciting commercialised AI research in 2024 has applied a layer of factual information or reinforcement learning on top of LLM-generated content. This serves the purpose of preventing hallucination: LLMs’ tendency to fabricate plausible-sounding but untrue information.

Hallucinations are arguably the most fundamental problem with LLMs, and result from a key feature of their architecture: they are designed to predict the next output in a sequence based on statistical trends in their training data, rather than to accurately reference factual information. This causes them to generate output that superficially looks like a plausible next word, but has no grounding in reality. To date, there has been little progress preventing the LLM architecture itself from hallucinating – so AI companies have had to focus on giving models access to facts in other ways.  Perplexity’s integration of facts and sources from search results, and the rise of ‘reasoning models’ like DeepSeek R1, are good examples. 



Some of the 2023-24 cohort were ahead of the curve on this trend. SurrealDB focuses less on direct AI content generation and more on creating a robust data infrastructure that can support complex AI applications. MindsDB works on bringing machine learning models directly into the database layer, allowing for more structured and controlled interaction with AI. But it is in 2024-25 that we really saw a rise in open-source startups addressing the inherent limitations of direct output by focusing on precision and modularity.

The 2025 cohort revealed

It’s really important to be aware that these, mostly ‘traditional’, metrics that we have described can give biased results depending on your GTM strategy, and that averages themselves are blunt and can obfuscate from what’s really going on under the surface.
The metrics that will be of most value to you will depend on your GTM strategy. You will have realised that we in several instances have stated that certain metrics are of less importance, or need to be tweaked, for PLG strategies. You might also find these metrics less helpful if your GTM strategy is focused on enterprise sales with high levels of upsell, or community driven adoption, for example.

There is a story behind each number, so even though the data may look good at first glance, sharing the story behind them is equally important. Conversely, numbers that look less than ideal can also hide positives. In other words, adding context is of high importance. For instance, we’re increasingly finding that more thorough analysis and customer segmentation is needed for accurate and reliable performance insights.

Concluding thoughts on the 2025 cohort

What characterises a company in the 2025 cohort?

As we look to 2025, the Rising Stars cohort offers a glimpse into the next frontier of open source innovation. This year’s most exciting open source projects aren’t just riding the momentum of established best practices like high contributor velocity and strong product-community fit, they’re developer-first by design, while at the same time deeply attuned to the tectonic shifts being driven by AI, automation, and intelligent tooling. They’re building at the intersection of code and cognition, where generative models, natural language interfaces, and autonomous agents are redefining how software is created, used, and scaled.

What sets them apart is a shared mindset: solving real, emerging problems with the openness, modularity, and speed that only open source can deliver. These projects are not just technical innovations. They’re infrastructure for the next era of software.

Below, we unpack four defining themes that capture the shape of this new wave.

1. Minimising hallucinations with rules and facts

Retrieval-augmented generation (RAG) is a technique for grounding LLM outputs in information from an external knowledge base, by converting that information into vector embeddings. In simple terms, RAG is like giving an LLM a smart search engine to fact-check its answers. While LLMs have significantly improved at integrating external context into their responses – mainly thanks to longer context windows – RAG techniques have evolved too. They remain a
cost-effective way to counteract LLMs’ inherent tendencies towards generalisation and hallucination.
The 2025 cohort contains a number of companies that are reimagining RAG:

• 1. PayloadCMS looks to differentiate on the basis of being a RAG-native CMS. It allows developers to set rules for converting documents into embeddings, then automatically stores the embeddings in a database and keeps them synced with document updates. This means that the content in the CMS remains “RAG-ready”: essential if customers want to continuously feed their content into AI models to keep them grounded in reality.
• 2. FalkorDB’s approach to RAG integrates graph database capabilities with language models. The graph relationships provide more comprehensive context for LLM queries, as well as transparent logic that traces how information will be retrieved by LLMs. The ability to follow the trail of information through the graph is another effective safeguard against hallucinations.

Beyond RAG, other companies in our cohort are designing with hallucination-prevention in mind. n8n.io has integrated AI chatbot creation as one of the custom workflows that can be built via the platform – and a key feature of this workflow is the ability to specify rules and link external logic to their chatbots. Apify, a web scraping tool, has placed sourcing high-quality data to improve LLM ground truth at the core of its offering. LiteLLM facilitates workflows involving calls to multiple LLM APIs, making hallucination detection through multi-agent interactions much easier.
At the application layer, applications optimised for data model flexibility help customers pull and access data in any format they need. Twenty is an open-source CRM with data model flexibility at the core of its offering. Like PayloadCMS, it is well-suited for workflows involving continuously enriching AI models with high-quality customer data.

2. Leveraging the fluidity between natural language and code

Building and orchestrating automated workflows – for tasks like reporting, sharing and analysing data – has historically been code-intensive. However, two of this year’s cohort are taking advantage of the growing ease of translating between natural language and code with the help of AI. Having built up loyal developer customer bases using community-led growth, both of these companies have been shifting their focus to less technical users with new releases.

n8n offers built-in AI nodes that enable non-technical users to implement sophisticated capabilities like document summarisation, question answering, and content generation. Their integration with LangChain allows users to describe what they want to accomplish in plain language, with the system suggesting appropriate nodes and connections. These nodes abstract away the complexities of working with large language models.

Kestra launched a redesigned no-code workflow creation tool in February 2025. The tool offers a simplified interface for visualising and modifying workflow dependencies and plugin properties.

AI makes translation between workflows built in the no-code tool and objects stored as code seamless. As a result, achieving deep precision and customisability is now a possibility even for non-engineers.

3. Integrating agentic workflows

No tech spotlight would be complete without some discussion of agentic AI, an application of generative AI which began in 2023 and shows no sign of slowing down.Unsurprisingly, multiple companies in our cohort have focused on rolling out their own AI copilots or assistants – or enabling customers to launch their own AI copilots or assistants – who can autonomously interact with internal codebases, knowledge bases, and other resources.

Improved translation between workflows built in the no-code tool 
and objects stored as code makes it increasingly possible for domain experts to achieve the same level of precision and customisability as engineers.

FlowFuse is a fast-growing, low-code visual development environment for building workflows which connect hardware with APIs and online services. This year, it launched an AI agent which generates code on the fly. This augments its traditional drag-and-drop node interface with the increasingly familiar chat-based interface, which is fast becoming the standard way to interact with AI agents.

Daytona is an open-source tool aimed specifically at agent builders. It offers a fast, scalable, and secure infrastructure for executing AI code and managing agents. It enables AI agents to run code concurrently, with real-time output streaming, and maintain context across parallel runs. This means that the agent’s state is preserved, enabling more advanced workflows where context needs to be carried over multiple executions.

Outside of the Rising Stars cohort, there are many more agent-centred companies which saw impressive growth this year. Multi-agent systems involving interactions between multiple LLMs – such as those which LiteLLM helps to facilitate – have become more common amongst open-source projects in 2024. 

LangChain’s LangGraph framework allowed developers to orchestrate stateful interactions between multiple LLMs, enabling complex task automation and conversational agents capable of handling multi-step queries.
As evolving AI capabilities enable more and more workflows to be fully automated, we expect ‘agent infrastructure’ to emerge as a distinct category.

4. ‘Traditional’ developer tooling

It is also notable that several companies in this year’s cohort are ‘traditional’ developer tools, including API management (Unkey), self-hosting (Coolify) and Python tooling (Astral, Pydantic). These are products aimed squarely at developers which continue to leverage the open-source ecosystem to build loyalty and gather feedback. This is not to say that AI has not shaped the offerings of these tools; Pydantic has launched a framework specifically for building with LLMs, for example. But it does demonstrate that for all the changes in software engineering workflows in the last 12 months, building an engaged developer community and working towards product community fit is still a tried and true way to scale a product for highly technical users.

Summary

We have high conviction in the 2025 Rising Stars cohort as they bring a new wave of foundational technologies to market, built for a landscape increasingly shaped by data-centric architectures, intelligent automation, and AI-native developer tooling. These projects aren’t just thematically aligned with today’s trends – they’re laying the groundwork for the next generation of infrastructure and applications in B2B software.

While the macro environment remains complex and capital efficiency in scaling software companies is critical, the technical calibre of these projects, combined with their strong contributor networks and alignment with high-growth end-markets, positions them well for sustained traction. 

Their open source DNA also gives them a powerful talent advantage, drawing in top-tier engineers who want to co-build meaningful, scalable tools in domains like agent orchestration, real-time data processing, and modular AI integration.

We’re not just bullish on their potential – we’re tracking their progress closely, and already excited to share what’s next. The future of enterprise software is being built in the open, and these are the teams leading the charge.