Inside EagerHQ: The Projects We Build, Open Source and Otherwise

EagerHQ is a UK software studio. We build cloud products, SaaS platforms, web apps, and agentic AI systems, and we try to be honest about the difference between the two sides of the workshop. Some of what we make is open source and lives on GitHub for anyone to fork. Some of what we make is product, polished and pitched to a specific audience. This post is a tour through both.

The goal here is not a glossy pitch. It is a field report. Expect real technical detail on the open source work, and a straight-up marketing look at the product we actually want you to use.

Voxlit is our voice dictation and AI agent for macOS. You say "Hey Voxlit" and the app listens, transcribes, and hands the result to an agent that can write, explain, translate, or edit text in whatever app you are focused on. It lives at voxlit.co and the source is public.

Architecture

• Native macOS client written in Swift and SwiftUI, shipped as a universal binary for Apple Silicon and Intel. No Electron, no web view.
• On-device hotword detection. A small CoreML model listens for the wake phrase locally, so audio never leaves the machine until the user explicitly triggers dictation.
• Streaming speech-to-text over a persistent WebSocket to Voxlit Cloud. The client batches 20ms audio frames, runs a voice-activity pass, and streams them up. Partial transcripts stream back so text appears as you speak.
• The agent layer is a thin orchestration process. It hands the transcript, the active app, the focused window title, and any selected text to a tool-enabled model. Tools include write_to_cursor, translate, explain_error, summarise_selection, and a macOS-specific paste_and_format that respects the target app's clipboard conventions.
• Voxlit Cloud is a Go backend on serverless infrastructure. It routes STT through a primary provider with automatic fallback, caches short completions, and issues short-lived tokens so no long-lived API keys ever land on the client.

Why it is open source

A tool that sits between your voice and your keyboard needs to be trustable. The fastest way to earn that trust is to publish the source. You can read exactly what audio is captured, where it goes, and how long it lives. Beta is free, and the open source tier will stay free once pricing is announced for the cloud.

For the full engineering breakdown, see Voxlit Under the Hood.

Patchbay is a browser-to-browser audio pipe. Two people open a room, share a short code, and audio flows peer to peer. No install, no login. It started as a weekend experiment and it is still online at patchbay.eagerhq.com for anyone who needs it.

• Pure WebRTC peer-to-peer. The signaling server only exchanges SDP offers and ICE candidates. Once the RTCPeerConnection is established, audio never touches our infrastructure.
• Signaling is a small Cloudflare Worker backed by Durable Objects. A room is a single Durable Object instance, so presence and ordering are trivial.
• Two profiles. Lossless PCM at 48kHz for musicians, HD Opus at 256kbps for general voice. A single toggle switches between them.
• The frontend is a single Svelte component plus a handful of Web Audio nodes. No service worker, no analytics, no tracking scripts.

Full protocol and code walkthrough in Patchbay: Browser-to-Browser Audio in 800 Lines of Code.

Now the other side of the studio. Webnite is a product. We market it, we polish it, and we are pitching it to a specific audience. It is a story-driven learning game for the CompTIA Security+ exam at webnite.app.

Security+ prep is dry. Flashcards do the job, but they do not make you want to come back tomorrow. Webnite makes you want to come back tomorrow. Sir Alaric, a medieval knight pulled forward through time into a world of cloud misconfigurations and rogue AI, learns in real time, and teaches you what he learns.

Longer write-up on the product and the curriculum in Why Story-Driven Learning Beats Flashcards.

Voxlit and Patchbay are the open parts of the studio. Artefacts of our engineering practice, shared because the community should be able to read, fork, and rebuild them. Webnite is the commercial side. It funds the next wave of projects and gives us the surface area to learn what shipping a real product to real learners actually requires.

• Sharp scope. We cut features until the product can be described in one sentence.
• Honest defaults. The first run should already be the right configuration for most users.
• Ship when ready. Not when the roadmap says so, not when a marketing calendar says so.

More on the operating model in How We Ship at EagerHQ.