Open Source Startups funded by Y Combinator (YC) 2026

GitLab is the first single application for the entire DevOps lifecycle. Only GitLab enables Concurrent DevOps, unlocking organizations from the constraints of today’s toolchain. GitLab provides unmatched visibility, radical new levels of efficiency and comprehensive governance to significantly compress the time between planning a change and monitoring its effect. This makes the software lifecycle 200% faster, radically improving the speed of business. GitLab and Concurrent DevOps collapses cycle times by driving higher efficiency across all stages of the software development lifecycle. For the first time, Product, Development, QA, Security, and Operations teams can work concurrently in a single application. There’s no need to integrate and synchronize tools, or waste time waiting for handoffs. Everyone contributes to a single conversation, instead of managing multiple threads across disparate tools. And only GitLab gives teams complete visibility across the lifecycle with a single, trusted source of data to simplify troubleshooting and drive accountability. All activity is governed by consistent controls, making security and compliance first-class citizens instead of an afterthought. Built on Open Source, GitLab leverages the community contributions of thousands of developers and millions of users to continuously deliver new DevOps innovations. More than 100,000 organizations from startups to global enterprise organizations, including Ticketmaster, Jaguar Land Rover, NASDAQ, Dish Network and Comcast trust GitLab to deliver great software at new speeds.

Caveman is the optimization stack for AI agents. It observes real agent workloads, finds cost and performance inefficiencies, tests improvements, and safely deploys the ones that work.

Lucebox makes workstations for local AI inference. They're private, always-on machines that arrive with Lucebox's inference engine installed and tuned, so developers point Claude Code, Codex or OpenCode on their own hardware. Lucebox is building toward a new category of self-improving personal AI computers that learn from the work you do every day, training small models on your own context until they're faster and better than the cloud at your tasks. The engine is open source, with 2.9K stars on GitHub and 95K+ downloads on Hugging Face.

Acyclic builds the infrastructure for agent swarms: open-source primitives for running agents in parallel, and a managed agents platform that runs swarms for you. Complex, long-running tasks are split across many agents, each working in its own instant fork of the workspace, and finish in a fraction of the time. We built Acyclic because we believe there are huge gains to be had in how agents are run, beyond the models themselves. Agents can already do real work, but they still do it one step at a time, so complex tasks take hours or days. Running them as swarms is an infrastructure problem: isolating each agent, dividing the work, and merging the results without conflicts. Today, teams use Acyclic to run heavy document-processing pipelines in minutes instead of hours. At Cambridge, we researched ML and auto-parallelising compilers, and then we built multi-agent systems in production: Ram as a Palantir FDE for pharma clients, shipping systems for clinical trial design, and Varun leading agent architecture at Tracelight for Big 4 accounting firms. We kept asking the same question: the agents were good enough, so why couldn't we run thousands of them at once to make everything faster and cheaper? Acyclic is our answer.

Connect to your AI Agent in 2 minutes. Get expert level advice tailored to your business backed by real data like keyword search traffic, backlinks, and technical audits of your site. Here are some good skills to start with: /seo-audit What 1-2 changes to your website will drive more traffic? /keyword-research Find topics worth writing about? /competitor-analysis Understand your competitors’ SEO strategy.

Agent Relay lets software teams' AI coding agents message each other directly, so people stop copy/pasting between Claude Code, Codex, Cursor, and their teammates' agents. A frontend agent can ask the backend agent what an API returns, two agents can sort out a merge conflict between themselves, and your agent can answer a colleague's agent while you sleep. Agents connect wherever they run: terminals, desktop apps, phones, laptops, cloud boxes, or a Mac mini across the world. Every session goes into shared, searchable history, so decisions don't get stuck in one person's chat. You can even let a customer's agent debug with yours. Setup takes about 5 minutes: give your agent the link and it installs itself. Use our hosted version free, or self-host the open-source version

Tracer is an AI research lab building more capable and efficient AI through model coordination. We research adaptive inference, ensembling, and systems in which multiple open-source models work together to produce capabilities beyond any single model. Our work spans reasoning, code, research, and agentic systems, with the goal of making advanced intelligence more accessible and economical.

Graphify is a Knowledge Graph control plane for enterprise software. The open-source tool has 116K+ GitHub stars and 6.5M+ downloads, and is used in production by engineers at Shopify, Datadog, JP Morgan, American Express, Harvey, Automation Anywhere, and Vanguard. On top of that graph, our enterprise layer reviews every pull request and formally verifies each code change: it proves the new behavior matches the old, or hands back the exact input that breaks it. That makes it built for the changes teams fear most, like refactors, framework upgrades, and large-scale migrations, where you need proof that nothing broke, not a guess. The graph updates itself as the code changes and keeps context instead of forgetting it, so your whole team and their coding agents work from one shared, always-current view of the codebase instead of a thousand private threads. It runs entirely inside your perimeter, fully air-gapped on-premise or self-hosted in your own cloud, so no code or data ever leaves your environment.

OneCLI (YC S26) is an open-source platform for delegating business-critical work to AI agents, from Salesforce updates to data analysis and engineering investigations. Agents work across your company’s apps, prepare changes, and pause for approval when your rules require it. Your team controls what they can access and change, while credentials stay out of the agents’ reach and every call is logged. https://www.onecli.sh

Orchestra is an inference cloud. We capture your AI work traces, train smaller models, and automatically deploy them (only when they outperform the expensive model you are currently paying for). Instead of renting the same frontier model forever, your AI infrastructure gets faster, smarter, and cheaper every time it runs. Suppose your product uses Claude for a repetitive legal workflow. Orchestra observes successful production runs, learns the specific tools and judgment required for that workflow, and trains a specialized open-weight model. Once that model matches or beats Claude on a held-out evaluation, Orchestra starts routing production traffic to it. As more work is completed, the model keeps improving. We started Orchestra because we believe every company using AI should be accumulating intelligence, not accumulating API bills.

Inference has surpassed training as AI’s largest compute cost, accelerated by increasing demand for AI and agents. By using proprietary KV cache transfer tech, OneTriangle has cut inference costs by 20% and time by 40% compared to present standards.

Robocurve builds open-source tools and independent benchmarks to measure how well robots can do real-world jobs. Instead of relying on unverified demo videos from frontier labs, we score their models on reproducible benchmarks that anyone can trust. Today there are no well-run, standardized robotics benchmarks. Labs evaluate in-house, and no independent group has stepped in to run continuous benchmarking as a service. The result is that no one actually knows how good anyone else is, or where the real frontier sits. Good benchmarks require operating and maintaining physical hardware and real-world setups. Simulation only goes so far, since models that look strong in sim can show large performance gaps once deployed in the real world. Building real-world benchmarks means coordinating job-domain experts, evals engineering, and hands-on robotics all at once. Frontier labs are targeting general-purpose robotics by 2028, yet the field of robotics evals barely exists. Whoever builds the trusted measure of robot capability becomes the reference everyone relies on. We combine backgrounds in AI evals and robotics to build and grow this field as fast as possible. We already shipped v1 of our open-source framework, Inspect Robots, and ran our first pilot scoring a frontier model on a real robot.

Sentry and Datadog dump a stream of alerts. Most are duplicates, most lack context, and you still have to solve the issue yourself. Superlog is AI-native observability that's meant not to be opened. A wizard scans your repo and installs proper OpenTelemetry instrumentation, then runs daily to keep up as you ship new code. When something breaks, Superlog groups the errors into a single incident, investigates with full context (logs, traces, recent deploys, past Slack threads), and posts one mergeable PR in Slack. You merge it, ignore it, or open it as a Claude Code session and modify it. Telemetry is vendor-neutral, so you keep every log, trace, and metric we install, even if you leave.

RunAnywhere runs open-weight models at full speed wherever a company needs them. Wally is our inference stack for open models: serverless inference, BYOC, or on-prem, so code and data never leave their network. One command puts a Wally-served model inside Claude Code, DeepSeek Harness, OpenCode, Hermes, OpenClaw, etc. When a model fits on the device, the same tooling runs it locally through our open-source SDKs for iOS, Android, macOS, Windows, Linux, and web, built on kernels written by our agentic harness for Apple and Qualcomm silicon (10K+ GitHub stars).

Expected Parrot helps companies simulate their customers with AI agents to explore pricing, product, marketing, communications and other scenarios at scale. Our open-source library and no-code web app let you design custom agent personas and conduct interviews and surveys with them using LLMs of your choice to generate results and reports. Everything is cached and reproducible, and you can send the same surveys to human respondents to validate results in the same interface.

Sourcebot is an open source code understanding platform for massive codebases. We’re used by thousands of engineers in some of the largest companies in the world, including NVIDIA, Red Hat, Wikimedia, and Arista Networks. Understanding code, not writing it, is the primary bottleneck for large engineering teams. For developers, this means onboarding onto complex codebases faster. For AI agents, it means getting the necessary code context to minimize hallucinations and maximize cohesion within the wider codebase. Sourcebot solves this by giving developers and AI agents the ability to regex search across millions of lines of code instantly, as well as ask questions across thousands of repos using any flagship reasoning model. Being open source and on-prem, we can get deployed in enterprises within minutes before other tools even get approval. We met a decade ago at McGill University and have been building together full time for over 2 years. We’ve personally felt the pain of understanding large codebases while working at Microsoft, EA, Ubisoft, Google, and Meta.

S2 is an API for durable streams. We make working with streaming data dead simple and reliable. The more we worked on data systems, the more we felt like there was a missing building block. The seamless experience of object storage simply did not exist for durable streams – so we set out to fix that! S2 reimagines streams as an unlimited and access-controlled web resource. Like if Kafka and S3 had a baby. - Building agents? Use granular streams to persist context in real time, make workflows auditable, and coordinate between agents. - Publishing fast-moving data? Streams support high fanout and can be accessed directly over the internet, so they are a great fit for broadcasting real-time feeds. - Sandboxing? Emit to per-instance streams, and read history or live events just as easily. - Local-first and multiplayer experiences? Propagate updates reliably and efficiently. Stop dealing with clusters, and start streaming.

Char is a daily notepad that finishes your todos for you. Type a checkbox, and an agent picks it up — researching, drafting, or scheduling on your behalf. Stay in command. Char does the work.

Epicenter is an ecosystem of open-source, local-first apps that share a memory. We store all your work—notes, transcripts, chat histories—in a single folder of plain text and SQLite. Every tool we build shares this memory: your text editor, personal assistant, etc.. It's open, tweakable, and yours. Grep it, open it in Obsidian, host it wherever you like. The choice is yours. Our vision is to build a personal workspace where you own your data, choose your own AI models, and replace siloed apps with open, interoperable alternatives. All while preserving authenticity and being free and open source. ❤️

TraceRoot is the open-source AI agent improvement platform. Turn production traces into actionable feedback and evals, closing the improvement loop with your coding agent.

Manufact is the MCP cloud platform to get your product ready for Claude Connectors marketplace, ChatGPT / Codex plugins store and any AI agent that supports MCP. We're the team behind mcp-use SDK, the open-source framework for building MCP Apps with 10k github stars and 10M+ downloads across python and typescript. It's the fastest and most complete solution for production-ready MCP servers.

Pangolin delivers seamless access and connectivity across your apps, infrastructure, and AI workloads while continuously enforcing enterprise-grade security. Built for agile IT teams, Pangolin eliminates the friction of bloated, legacy systems with an open platform that’s easier to deploy for your workforce. It's open-source.

You built your product in a weekend and deployed to Vercel, Supabase, Railway or AWS. Now you need to figure out how to pick the right region, scale to multiple instances, add monitoring and alerts, and who to ping when something breaks… and you don't have a cloud engineer. We spent a decade at our previous startup (CodeSignal), building a $500M platform while scaling to millions of users and customers like Meta & Netflix. Now we're standardizing how web applications are operated in production, so developers can build reliable production apps without becoming cloud infrastructure experts. Modelence is a production cloud that's designed for enterprise-grade apps, abstracting away the infrastructure layer: - Deployments & hosting - Background jobs and persistence - File storage, real-time data and networking - Monitoring and alerts Platforms like Vercel and Supabase are easy to start with, but most companies that start getting real customers eventually need a persistent backend where processes can run beyond just a few minutes at a time, unlike pure serverless. And low-level cloud platforms like AWS are meant to be operated by cloud engineers and are too complicated for engineers who want to focus on product rather than infra. Modelence is your managed backend infrastructure, built on decades of our experience operating web apps in the cloud.

On-device AI for phones, wearables, smart home and small robots.

stagewise is building Klex Bots, autonomous AI coworkers that can take on real work inside companies. Unlike traditional AI assistants, Klex Bots don't need to be prompted for every task. They operate continuously, maintain their own memory and workspaces, and use the same tools and systems as human employees. They can execute complex, multi-step tasks independently, collaborate with other bots or humans, and can use multiple sandboxes/computers depending on configuration. Our Klex Cloud platform makes it easy for businesses to deploy and manage these digital coworkers securely, using their preferred AI models and infrastructure, hosting agents either in cloud or on custom infrastructure.

Magnitude is an open source inference engine that tunes its kernels on your hardware to find the fastest way to run each model. It's over 2x faster than Ollama and works on Apple Silicon, NVIDIA, AMD, or just a CPU. One click connects it to the agent you already use (Pi, OpenCode, Hermes, Codex, and more)

As AI agents replace traditional software, they need a way to store, recall, and reason over contextual data. HelixDB gives them all of that in one system. We’re building the knowledge infrastructure every AI agent will depend on.

Kaelio is the company behind ktx, the open-source context layer that makes data agents reliable. Data agents today fail because they lack structured context and determinism. Metrics are defined in five places, governed in none. Teams spend months on semantic layer projects that never finish. ktx fixes that. Connect your warehouse, BI tools, and docs. ktx builds the context layer automatically. Your team governs it. Any agent can query it. One context layer across your entire data stack.

nao is an open source framework to build and deploy analytics agent. Create the context of your analytics agent with nao-core cli: data, metadata, modeling, rules, etc. Deploy a UI for anyone to chat with your agent and run analytics on your data.

Give every agent a cloud desktop. Built for Claude Code, Codex, OpenClaw, and any computer-use agents.

Codevisor is an open-source native macOS app for agentic engineering. Run Claude Code, Codex, Pi, and any ACP-compatible coding agent on your own machines in one place. Codevisor brings agent chats, projects, and an integrated terminal into a single workspace. Connect remote Macs or Linux servers to run agents where your code lives and keep working across machines.

Corsair connects your app to any other API. Use Corsair to build custom agents, dashboards, and workflows with hundreds of different integrations. Get started here: https://docs.corsair.dev/quick-start

Mentra is building the open source operating system for smart glasses. The smart glasses hardware is finally ready, so we're building the OS and app store to power the next computing platform.

superglue learns how your systems work from your company's knowledge and performs the implementation that normally requires human coordination and engineering work. Implement NetSuite, Sage Intacct, SAP, Business Central or Acumatica in days. Agents map and migrate legacy data, configure the system, and keep data in sync post go-live. Connect ERP, CRM, databases, and internal systems to AI platforms like Claude. Create governed data access for AI agents and track data usage across your org. Connect customer systems, import historical data and get new integrations live in hours. Let agents manage the entire implementation process end-to-end.

Mastra is the easiest way to build a production AI application, built by the team that brought you Gatsby. We provide a complete set of AI primitives: workflows for complex operations, agents for autonomous decisions, RAG for knowledge integration, and evals for quality & accuracy. Mastra is open-source, modular, and built in Typescript for product developers. You can deploy Mastra on any cloud, and it gives you a straightforward syntax to solve the hardest problems building with AI: human-in-the-loop workflows, agent memory and tool selection, and output quality.

Confident AI provides enterprises an all-in-one platform to evaluate, observe, red team, and govern AI applications. Built by the creators of DeepEval (16k stars, 8m monthly downloads), Confident AI is able to offer battle-tested, open-source evaluation algorithms while providing the infrastructure needed for teams to stay confident their AI systems.

We enable AI to control the browser. Browser-use is the biggest open-source web agent project with 110k+ Github stars.

Random Labs is building Slate. Slate is a coding agent designed to work with you for long hours on hard problems.

Unsloth helps builders create custom models better & faster using RL and custom kernels. We're developing the all in one solution to help you create highly-accurate custom models 30x faster with 90% less memory use. With over 10 million monthly model downloads and 40K GitHub stars, our mission is to make training and open-source the best it can be!

Rowboat is an AI coworker for everyday work - email, meetings, projects, and people. It builds a local knowledge graph from your work and uses it to draft emails, prep meetings, track projects, and get work done on your own computer. It comes with batteries included - meeting note capture, in-app browsing, search, and app integrations - plus MCP support for custom tools.

Opslane catches the bugs code review misses. On every PR, it runs your app against test versions of the services it calls, from Stripe to the LLM. Then it tries what breaks in production: a service failing, two requests at once, a restart mid-job.

Dench: AI CRM + Full-Service Deployment Most CRMs sell you software and hand you a 200-page setup guide. Dench is different. We're the only CRM that ships with an embedded team: we migrate your data, build your object schema, configure your integrations, train your people, and deploy AI agents that keep your pipeline clean without a Salesforce developer, a RevOps contractor, or a six-month implementation project. From day one, our team works inside your stack. We pull your existing data out of HubSpot, Salesforce, or whatever spreadsheet graveyard you're currently using, and rebuild it properly in Dench. Custom objects, custom fields, automation rules, integration config across Gmail, Slack, Notion, GitHub, LinkedIn, and 50+ other tools. Everything set up the way your business actually works, not the way a generic CRM template assumes it does. Then we train your team. Not a one-hour Zoom and a PDF. Real sessions that build CRM discipline: how to log activity consistently, how to keep data clean, how to use AI agents to do the repetitive work so your reps spend time selling instead of updating fields. Quarterly business reviews keep the system evolving as your pipeline grows. The AI layer is what makes it permanent. Dench agents enrich leads automatically, move deals through stages based on real signals, draft outreach, log interactions, and flag hygiene issues before they compound. Your CRM stays accurate without someone babysitting it. No implementation consultant invoice. No dedicated Salesforce admin headcount. No duct-taped automation stack. One engagement, everything included, starting at $1,500/month.

Mem0 enhances AI assistants and agents with an intelligent memory layer, enabling personalized AI interactions. It remembers user preferences, adapts to individual needs, and continuously learns over time - ideal for customer support chatbots, AI assistants, and autonomous systems.

An open source native macOS terminal built on Ghostty. Vertical tabs, notification rings when agents need attention, split panes, and a CLI for programmability.

We work on methods to autonomously scaling evals/environments for post-training agents.

Hatchet is an orchestration platform for managing and observing AI agents, background tasks, and mission-critical workflows. Our mission is to provide developers with a simple, powerful, and reliable orchestration layer, enabling them to focus on building great applications.

Identifying performance bottlenecks and strategizing ways to solve them takes 4-8x longer than actually writing the code to fix them. We're building an agent that is an expert at analyzing the performance GPU systems like inference engines at all levels of the stack - from CPU-GPU interactions down to GPU kernels. Pairing our agent with Cursor / Claude Code allows you to automate both the reasoning and code implementation steps of performance optimization. What used to take weeks can now be done in days. Along with our AI agent, we have features such as running diffs on system traces as well as sharing and collaboration features that make our VSCode extension the most powerful way to work on performance optimization.

Tracecat helps AI-native security teams build agents and automate work. All-in-one platform with agents, skills, prompt-to-workflows, and case management. Scale cyber defense with agents you own. Open source and community-driven at it's core: https://github.com/TracecatHQ/tracecat

Superagent helps AI-native developers trust what they merge, control what their agents do, and show users it's safe. We work in your coding agent and on GitHub: - Trust the change: checks on every pull request, before merge - Control it where it runs: boundaries around what your agents can see, say, and do - Break it before it ships: red teaming against what you actually run Free for open source.