The Race to Build Specialized, High-Context AI DevTools

Across the board, we’re seeing AI devtools shift from broad, general assistants to tightly scoped solutions that understand the context, the stack, and the real work developers are trying to do. It’s less about showing off and more about showing up, right where the problem is.

Oh, and if you thought acquisitions were slowing down for the summer… think again. It’s been a week of fun, games, and billion-dollar talent grabs.

🏆 Funding Wins

ParadeDB Raises $12M to Supercharge Postgres Search

ParadeDB has closed a $12 million Series A led by Craft Ventures, with participation from Y Combinator. The company, founded by Philippe Noël and Ming Ying, is building an open-source Postgres extension that replaces Elasticsearch-style search and analytics without requiring external databases. Their approach keeps everything inside Postgres, cutting latency and complexity for high-frequency workloads. Enterprise traction has followed, with customers including Alibaba, Modern Treasury, and Bilt Rewards.

🔍 Explore ParadeDB’s Postgres-native search tools at paradedb.com

$200M Unicorn Leap for Lovable’s AI-Powered Site Builder

Lovable has raised a $200 million Series A round led by Accel, just eight months after launch, at a $1.8 billion valuation. Based in Stockholm, the startup helps users build apps and websites using natural language, attracting more than 2.3 million active users and 180,000 paying subscribers. The company says it reached $75 million ARR within seven months on a lean team of 45. Existing investors Creandum, 20VC, and Visionaries Club joined the round, along with angels from Klarna, Slack, Remote, and HubSpot. As CEO Anton Osika puts it;

“Every day, brilliant founders and operators with game-changing ideas hit the same wall: they don’t have a developer to realize their vision quickly and easily.”

💗 Try Lovable for yourself and explore what’s possible at lovable.dev

$7.3M for Blaxel’s AI Agent Infrastructure Play

Blaxel (YC X25) has raised $7.3 million in seed funding led by First Round Capital, fresh out of Y Combinator’s X25 batch.

“We believe that there is a huge economy which is starting around the agents… the infrastructure we have today has not been designed for this new wave.”

– Paul Sinaï (co-founder and CEO of Blaxel)

The six-founder team, is building cloud infrastructure designed specifically for autonomous AI agents, with features like sub-25ms boot times, agent-level policies, and HIPAA/SOC2 compliance. Their agentic platform already processes billions of runtime seconds across 16 regions, with clients using it to analyse videos, replays, and behavioural data.

🕸️ Find out more about Blaxel here.

🚀 3, 2, 1… Launches

Embedder Wants to Be the IDE Embedded Engineers Actually Like

Tired of parsing 100-page datasheets or hand-writing boilerplate I2C drivers? Embedder (YC S25) has launched a web-based AI IDE built specifically for embedded developers. It generates production-ready drivers from a single PDF, provides a datasheet-aware chat agent for debugging, and replaces PDF scanning with a searchable component viewer. Founded by Ethan Gibbs and Bob Wei, the platform already supports STM32, ESP32, Arduino, and Raspberry Pi Pico, with a CLI-based agent mode coming soon that will flash, test, and debug in real time.

🔌 Sign up for free here and reach out to the team to be part of their first design partnerships at founders@embedder.dev

Cactus Brings Local AI Model Deployment to Your Phone

Cactus (YC S25) is now live with a powerful open-source framework that lets developers run LLMs, VLMs, and embedding models directly on smartphones and other low-power devices. Built by Roman Shemet, Cactus supports Flutter and React Native bindings, accommodates GGUF models from Hugging Face, and runs anything from FP32 to 2-bit quantisation. Unlike Google and Apple’s platform-specific AI runtimes, Cactus gives mobile teams full control across devices with local inference and optional fallback to the cloud when needed. Early users are already building RAG pipelines, phone-based tool agents, and privacy-first medical applications.

🌵 Explore the framework on GitHub and join the community on Discord.

Async Ships Dev Assistant That Writes PRs While You Focus Elsewhere

Async has launched a multi-agent coding platform designed to take low-priority engineering tasks off your plate. Founded by Paul Bae and Mikkel Kim, Async runs in the cloud, learns your codebase, and spins up specialised agents for bugs, migrations, and repetitive tasks. Once complete, it creates a pull request with linted, tested, and deployable code. It’s already outperforming other assistants on SWE-Bench and comes with a clean mobile interface for managing tasks on the go. If your team spends too much time on engineering gruntwork, Async might be worth a look.

🔗 Try the agent and see how it works at async.build

StakPak Launches DevOps Agent Built for Real Infrastructure Work

Most AI agents fall apart when it comes to real-world DevOps. Stakpak is taking a different approach. Built in Rust and designed to run in your terminal, it understands your stack, handles long-running deployments, and manages infrastructure without exposing secrets. CEO George Fahmy announced the launch on Product Hunt today after months of development, highlighting features like mutual TLS, dynamic secret redaction, and local indexing of Terraform and Kubernetes configs. StakPak is open source, CLI-native, and now available with a lifetime launch offer.

🧰 Show your support on Product Hunt and see what StakPak can do here

Herdora Turns Your PyTorch Into GPU-Optimised Gold

The best GPU engineers work at companies like Nvidia and OpenAI, commanding huge salaries to wring every drop of performance from hardware. Most teams don’t have that luxury. They’re stuck with PyTorch code that burns through compute while barely touching the GPU’s full potential. Herdora fixes this. Founded by Emilio Andere and Steven Arellano, Herdora automatically writes the kind of custom kernels those engineers would hand-tune over weeks. It also monitors models in production, catching and fixing inefficiencies in real time. No CUDA knowledge required. Just faster models, lower latency, and more hardware flexibility.

🐄 Book a call or join the pilot at herdora.com, find out more in this YC write-up and give this video a watch.

Nia Gives Your Coding Agent the Context It’s Been Missing

Nozomio has launched @Nia, a powerful context indexing engine that helps coding agents like Cursor, Continue, and Cline answer deeper technical questions with less handholding. Founded by Arlan Rakhmetzhanov, Nia acts as an MCP server that indexes entire codebases and documentation sites, so agents “just know” your setup. In internal evaluations, Nia improved Cursor’s performance by 27% by surfacing information that isn’t covered in training data or web search. It also supports deep research prompts, like generating comparison tables across frameworks with preloaded repo context.

🧠 Get your API key and start indexing with Nia here

🔦 DevTool of the Week: Railtown Technologies AI

Founded in Vancouver by Cory Brandolini, Marwan Haddad, Dr. Elliot Holtham, and Kevin O’Neil. Each founder brought decades of experience in software development and tech startups. The company was officially incorporated in 2011 (originally under the name “Railtown Capital Corp.” before focusing on AI developer tools in 2019).

Railtown AI Technologies is now focused on one over-arching goal: giving developers back the time they lose to debugging and maintenance. After years working in software and AI, the team set out to build tools that fit inside your stack and solve problems before you need to ask.

2025 has been a huge year for the company. Their core product, Conductr, was launched. Conductr connects to your codebase, CI/CD pipeline, and production logs. It traces the root cause of new errors, drafts the fix, and creates the pull request. It can even generate documentation and release notes. Recent additions include Auto Triage, which flags and prioritises issues as they appear, and a new Visual Studio extension that brings Conductr into the editor.

“This is about getting the right eyes on the right problem, instantly… Auto Triage gives engineering teams the context and clarity they need to act quickly and effectively, without slowing down development.”

Beyond Conductr, the team has released Langtracks, a framework for building RAG pipelines with minimal setup. A partnership with Quantum.IQ is now underway to support hybrid quantum-classical development, alongside plans to acquire AI Partnerships Corporation and expand into enterprise AI tooling.

Railtown’s trajectory points to a team steadily expanding its toolkit, staying close to real developer pain points while building towards a more efficient future. Their work aligns closely with the principles of Developer Productivity Engineering, using automation where it counts, to reduce friction without adding complexity.

If you’re exploring AI tools that fit naturally into modern dev workflows, Railtown is well worth a closer look.

🔎 Find out more 🧪 See it in action – GutHib

🤝 Acquisitions – a summary!

Acquisitions, acquihires, and talent grabs have dominated the AI coding space over the past week. It started with OpenAI’s $3 billion offer to acquire Windsurf falling through. Within hours, Google stepped in with a $2.4 billion reverse-acquihire, hiring the co-founders and R&D leaders and licensing key parts of Windsurf’s tech. Just days later, Cognition , the company behind AI coding agent Devin, acquired the rest of Windsurf’s product, IP, and remaining team in a fast-moving weekend deal. The agreement ensured all employees received equity participation and waived vesting cliffs.

Over at Cursor, the team quietly acquired CRM startup Koala. The product will shut down in September, but several top engineers will join Cursor to help strengthen its enterprise offering. With Google, Cognition, and Cursor all stepping up their M&A activity, the battle for dominance in AI coding is only accelerating.

🧠 Read more on Cognition’s acquisition here 🐨 Learn more about Cursor’s move here and what this means for Koala

 

As the tools get sharper and more focused, so do we. Thanks for reading and being part of this growing community of builders who care about doing things thoughtfully.

Got a launch, raise, or tool we should feature in next week’s 100th edition?

Hit reply and send it our way.

Louise Ogilvy (Recruitment Director) – UK/USA

Becca Combe MIRP CertRP (Lead Recruiter) – USA

Ben Chipchase (Lead Recruiter) – UK

Natalie Harper (Administrator)

Looking for your next DevTools role?

Visit our jobs page