Why a VC Firm Tracks Open-Source Traction
At Vela we believe the best venture capital decisions start with data, not narratives. We are an AI-native, scientific venture capital firm: every investment thesis we develop is grounded in quantitative signals from the developer ecosystem. Open-source repository traction is one of the strongest leading indicators of where infrastructure demand is heading.
This report is a product of the same tooling we use internally. Our AI-powered research pipeline continuously monitors GitHub, enriches repositories with founder and funding metadata, and clusters emerging activity into investable themes. We publish this analysis to share how we think about product-led growth signals and developer adoption curves with the broader community.
“Stars are noisy. Trends are signal. We cluster repositories into movements, then ask: what infrastructure is missing for this movement to scale?”
Top Trends by Aggregate Stars
Combined GitHub stars of all repositories tagged to each trend. This measures total ecosystem maturity and developer mindshare.
| Rank | Trend | Stars | Growth (90d) | Repos |
|---|---|---|---|---|
| 1 | OpenClaw Ecosystem & Personal AI Assistants | 809.6K | +124.2K | 15 |
| 2 | Terminal-Based AI Coding Agents | 561.6K | +74.6K | 12 |
| 3 | Local & On-Device LLM Inference | 502.8K | +25.1K | 13 |
| 4 | Cross-Platform Proxy & Anti-Censorship | 411.5K | +24.7K | 15 |
| 5 | Multi-Agent Orchestration & Swarms | 359.6K | +29.5K | 14 |
| 6 | AI Voice Cloning & Speech (TTS/STT) | 328.0K | +22.1K | 13 |
| 7 | Model Context Protocol (MCP) Servers | 307.0K | +19.7K | 13 |
| 8 | Agentic Skills Frameworks & Plugins | 305.8K | +120.2K | 24 |
| 9 | Browser Automation & Web Scraping | 278.8K | +21.8K | 9 |
| 10 | Document Parsing & OCR for LLMs | 237.9K | +11.4K | 7 |
| 11 | Persistent Memory & Context Management | 220.6K | +29.8K | 13 |
| 12 | AI-Native Note-taking & Knowledge Bases | 200.9K | +12.8K | 7 |
Fastest Growing Trends
Ranked by stars gained in the last 90 days. This measures momentum: where developer attention is accelerating right now.
| Rank | Trend | 90-Day Growth | Total Stars |
|---|---|---|---|
| 1 | OpenClaw Ecosystem | +124.2K | 809.6K |
| 2 | Agentic Skills Frameworks | +120.2K | 305.8K |
| 3 | Terminal-Based AI Coding Agents | +74.6K | 561.6K |
| 4 | Persistent Memory & Context | +29.8K | 220.6K |
| 5 | Multi-Agent Orchestration | +29.5K | 359.6K |
| 6 | Local & On-Device LLM Inference | +25.1K | 502.8K |
| 7 | Cross-Platform Proxy | +24.7K | 411.5K |
| 8 | AI Voice Cloning & Speech | +22.1K | 328.0K |
| 9 | Browser Automation & Scraping | +21.8K | 278.8K |
| 10 | MCP Servers | +19.7K | 307.0K |
The two fastest-growing categories, personal AI assistants and agentic skill plugins, together gained 244K stars in 90 days. This is a clear signal that developer demand is shifting from “use an AI model” to “build a personalized AI agent with composable capabilities.” Infrastructure that enables this composability is where we see the strongest venture opportunities.
Trend-by-Trend Analysis
Below is a deep dive into each of the 19 trends, with representative repositories and our take on what they mean for infrastructure investors.
OpenClaw Ecosystem & Personal AI Assistants
A massive surge in repositories building around OpenClaw, an open-source framework for creating personal AI assistants. This trend highlights a shift toward highly customizable, cross-platform AI agents that users can run locally or on minimal hardware.
Terminal-Based AI Coding Agents
Developers are increasingly adopting CLI-first AI coding assistants like Claude Code and OpenCode instead of traditional IDE extensions. These tools operate directly in the terminal, allowing them to seamlessly integrate with existing developer workflows, execute shell commands, and autonomously edit codebases.
Local & On-Device LLM Inference
Driven by privacy concerns and hardware advancements, there is a massive ecosystem growing around running frontier AI models locally. Projects are focusing on extreme optimization, 1-bit quantization, and efficient inference engines that allow powerful models to run on consumer-grade GPUs or even CPUs.
Cross-Platform Proxy & Anti-Censorship Clients
A highly active community is building and maintaining modern GUI clients for network proxies and anti-censorship protocols. These tools are essential for users in restricted regions to access the global internet and AI APIs securely.
Multi-Agent Orchestration & Swarm Frameworks
Moving beyond single-prompt interactions, developers are building frameworks to orchestrate swarms of specialized AI agents. These platforms allow multiple agents to collaborate, delegate tasks, and execute complex, multi-step workflows autonomously.
AI Voice Cloning & Speech-to-Text (TTS/STT)
Open-source audio AI is exploding, with projects offering highly accurate, offline speech recognition and few-shot voice cloning. These tools are being used to generate audiobooks, transcribe meetings locally for privacy, and give AI agents natural-sounding voices.
Model Context Protocol (MCP) Servers
The adoption of the Model Context Protocol is standardizing how LLMs connect to external data sources and tools. Developers are rapidly building MCP servers to bridge AI agents with everything from web browsers and databases to game engines and social media platforms.
Agentic Skills Frameworks & Plugins
A rapidly growing ecosystem of Skills that can be injected into AI agents to give them new capabilities. These modular plugins allow agents to perform specific tasks like marketing analysis, UI design, or interacting with external APIs.
Browser Automation & Web Scraping for AI Agents
Traditional web scraping is evolving into AI-driven browser automation. These tools convert complex, dynamic web interfaces into structured markdown or JSON, allowing AI agents to autonomously navigate websites and extract data.
Document Parsing & OCR for LLMs
To feed complex documents into RAG pipelines and LLMs, developers are building advanced parsing tools. These projects specialize in extracting structured text, tables, and metadata from messy PDFs and images, making them LLM-ready.
Persistent Memory & Context Management for AI Agents
Solving the amnesia problem in LLMs, these projects provide infrastructure for long-term agent memory. By using vector databases, knowledge graphs, and memory-centric OS layers, they allow AI systems to retain context across sessions and continuously learn.
AI-Native Note-taking & Knowledge Bases
A new generation of personal knowledge management tools is emerging, built from the ground up with AI integration. These self-hosted workspaces use local LLMs to automatically tag, summarize, and connect notes, acting as a second brain.
API Proxies & Gateways for LLM Access
As developers juggle multiple AI providers, there is a surge in unified API gateways. These tools proxy requests, manage quotas, translate API formats, and allow users to pool subscriptions or bypass regional restrictions.
AI Video Generation & Editing
Open-source models and automated pipelines for video generation are gaining massive traction. These tools allow users to generate short dramas, automate video editing, and create high-quality video content entirely from text prompts.
AI-Powered Penetration Testing & Cybersecurity Agents
Cybersecurity is being automated via specialized AI agents capable of autonomous vulnerability discovery. These frameworks use LLMs to orchestrate standard security tools, analyze code for exploits, and conduct end-to-end penetration tests.
Desktop GUI Clients for CLI AI Tools
While many AI coding tools are terminal-first, a parallel trend is building cross-platform desktop GUIs to manage them. These wrappers provide visual dashboards, quota tracking, and easier project management for CLI tools.
AI Financial Trading & Hedge Fund Agents
Developers are combining multi-agent LLM frameworks with financial data to create autonomous trading systems. These AI hedge funds conduct deep market research, analyze sentiment, and execute trades automatically.
Vibe Coding & Natural Language Software Development
A cultural and technical movement termed Vibe Coding is emerging, focusing on building software entirely through natural language prompts without writing traditional code. Repositories are popping up to provide tutorials, guidelines, and environments optimized for this workflow.
Nano Banana Pro (AI Image Generation)
A highly specific, viral trend centered around the Nano Banana Pro AI image generation model. The community is rapidly building prompt libraries, slide generators, and social media content creation tools based on this specific architecture.
What This Means for Investors
Three macro themes emerge from this data that directly inform our investment strategy at Vela:
- The agent stack is unbundling. Developers are not building monolithic AI applications. They are assembling agents from composable pieces: skills, memory layers, browser connectors, and orchestration frameworks. The investable layer is the infrastructure that makes this assembly reliable.
- Local-first is not a niche. On-device inference, self-hosted knowledge bases, and privacy-preserving AI tools collectively represent over 700K stars. This is mainstream developer demand, not an ideological fringe. Companies that solve the UX gap between cloud and local AI will capture significant value.
- Developer tools are the new distribution moat. Terminal-based coding agents, MCP servers, and skill plugins are creating ecosystems with strong network effects. The winning platforms will be those where the community builds the long tail of integrations.
“We don't predict trends. We measure them. When 854 repositories independently converge on the same architecture, that is not a trend forecast. That is demand already in motion.”
Our Methodology
We scanned GitHub's public event stream via BigQuery for repositories that gained 20+ stars in the last 90 days. From the top 854 shortlisted repos, we enriched each with company, founder, and funding data using Gemini-powered web search grounding. Trend identification was performed by Gemini 3.1 Pro, which analyzed repo descriptions and company context to cluster repositories into specific thematic ideas. A single repository can appear in multiple trends. Stars are aggregated per trend.
Frequently Asked Questions
How does Vela Partners use open-source data for venture capital?
Vela is an AI-native, scientific venture capital firm. We continuously monitor developer ecosystem signals, including GitHub repository traction, contributor growth, and dependency adoption, to identify emerging infrastructure trends before they become consensus. This data-driven approach complements traditional deal sourcing and due diligence.
What is a scientific venture capital firm?
A scientific VC applies quantitative methods, AI-powered analysis, and reproducible research to investment decisions. Instead of relying solely on pattern matching and gut instinct, scientific VCs like Vela build systematic pipelines that measure developer adoption, product-led growth signals, and market timing with data.
What is product-led venture capital?
Product-led VC evaluates companies primarily by their product adoption metrics (user growth, developer engagement, open-source traction, and organic distribution) rather than just team pedigree or market size estimates. Vela tracks these signals at scale using AI agents that monitor thousands of repositories and developer communities.
How were these 19 trends identified?
We scanned GitHub's public event stream via BigQuery for repositories that gained 20+ stars in the last 90 days. From 854 shortlisted repos, we enriched each with company, founder, and funding metadata using Gemini-powered web search grounding. Trend clustering was performed by Gemini 3.1 Pro analyzing repo descriptions and company context.
What is the most popular open-source AI trend in 2026?
By aggregate GitHub stars, the OpenClaw ecosystem for personal AI assistants leads with 809.6K stars. By recent momentum, Agentic Skills Frameworks gained 120.2K stars in just 90 days, making it the fastest-accelerating category alongside OpenClaw.
What is MCP (Model Context Protocol) and why does it matter?
MCP is an open standard for connecting LLMs to external data sources and tools. It matters because it creates a universal interface for AI agents, similar to how HTTP standardized web communication. With 307K aggregate stars and growing, MCP adoption signals a maturing agent infrastructure layer.
What are multi-agent orchestration frameworks?
Multi-agent orchestration frameworks like CrewAI, Dify, and LangGraph allow developers to coordinate multiple specialized AI agents working together. Instead of one model doing everything, these frameworks let you build swarms of agents that collaborate, delegate, and execute complex multi-step workflows autonomously.