Q. Could you share a bit about your background in technology and how your love for programming started?
Oliver Kohl: My love for programming goes way back to my teenage years in the 1980s. I got my first Commodore 64 around 1984 or '85, and then moved to the Amiga, where I started programming in Assembly. Back then, personal computers were brand new, and the few kids who were interested naturally formed groups to build intros, demos, music, and graphics.
Looking back, what hooked me was creating unique experiences on a computer—building something you own that has a tangible, visual representation. I’ve spent my entire career working across various operating systems, languages, and enterprise roles, and Mirall is the latest incarnation of that passion.
Q. At what point in your journey did you come across Holepunch and the Pear ecosystem?
Oliver Kohl: I describe myself as a technologist, I’m always watching the space to see what new capabilities are emerging. About five or six years ago, I saw Mathias Buus at a small meetup presenting his early hole-punching framework for connecting computers directly across firewalls and corporate network restrictions. I found that fascinating.
A few years later, I reconnected with Holepunch when I started looking seriously into sovereign computing and data ownership, being German, data privacy is naturally close to my heart. I downloaded Keet, tested it with a few friends, and was blown away by the video call quality and polish. That led me to explore the Holepunch GitHub repositories. When you think about data sovereignty, stacks like Pear and Holepunch are essential for the future of software because relying on centralized clouds is too restrictive—you're always asking someone else for permission.
Q. Was there a particular "Aha!" moment where you knew this was the stack you wanted to build on?
Much of it came from actually using Keet. Being able to restore my identity using a 24-word seed phrase just like a Bitcoin wallet, with zero servers involved, clicked immediately. And seeing high-resolution, ultra-low-latency video calls running machine-to-machine without a cloud bottleneck degrading the stream showed me the raw potential of direct P2P bandwidth.
When I dug into the frameworks—Hypercore, Hyperbee, Hyperswarm—I didn't come from a deep cryptographic background, so I spent time learning how DHT discovery, handshaking, and NAT hole-punching work under the hood. As a product builder, I look for frameworks that solve the hardest infrastructure problems so I can focus on user experience. With the Pear stack, building Mirall felt natural. When I show people that Mirall transfers massive files with zero cloud infrastructure, they’re shocked, and then they try it and see how well it works.
Q. How did you find the learning curve during those initial weeks and months?
Oliver Kohl: For Mirall, I didn't write a single line of code manually, the entire application was built through AI-assisted software engineering. Because of that, my learning curve for adopting the Holepunch stack was practically non-existent; my real learning journey was mastering how to architect, decompose problems, and guide an AI to build production-grade software.
From that perspective, what makes a framework adoptable today is how well an AI can understand it. Having open-source repositories, clear examples, and accessible documentation is critical so the AI can digest the primitives accurately. Because Holepunch is open-source and cleanly modularized, the AI was able to handle the low-level implementation while I focused on enterprise solution architecture, system decomposition, and product quality.
Q. Let's talk about Mirall itself. What was the specific moment that inspired you to build it?
Oliver Kohl: Every October, I attend the Evolution Mallorca International Film Festival with a friend. We both come from enterprise software, so we enjoy mingling with film producers, cinematographers, and editors. Last year, I overheard post-production teams complaining about how painful it is to send massive, unreleased movie files. They hate relying on third-party cloud services like Dropbox, Google Drive, or WeTransfer—both because of size/bandwidth limits and because you never truly know where your pre-release media ends up.
With Keet and the Holepunch stack fresh in my mind, I realized I could solve this directly without middlemen. Having a direct network into that film community gave me a focused niche to build for. Two weeks ago, I went back to Mallorca and demoed Mirall to those same filmmakers, and the feedback was fantastic—they immediately saw the value of zero-cloud transfers.
Q. Was it a conscious decision to target film studios and production houses rather than the broader consumer file-sharing market?
Oliver Kohl: Yes. First and foremost, I wanted to build a tool that genuinely solves a problem. In a P2P architecture, I don't pay for cloud storage or bandwidth. Users use their own drives and connections, and I have zero visibility into who shares what. Charging individual users arbitrary limits for transferring a terabyte over their own hardware didn't make sense to me.
So, Mirall's core desktop app, along with our blind relay infrastructure, is free and open-source so anyone can verify that there is no tracking or hidden backend.
However, film studios and enterprises need fleet management, compliance, and auditability. That’s why I introduced Mirall for Studios, a management console where studios can generate license keys, push locked configuration templates for backups and self-hosted relays, and maintain audit logs of who shared what and when to resolve production disputes.
Q. Looking under the hood, how does Mirall’s architecture work, specifically how files are chunked, verified, and served directly from disk?
Oliver Kohl: Architecturally, Mirall runs on a Bare backend worker paired with an Electron + React frontend across Windows, Linux, and macOS. All the P2P heavy lifting happens in the Bare worker, while React simply renders state changes.
Most core problems like identity, handshaking, and state consensus were solved straight out of the box using Holepunch primitives like Hyperbee and Autobase. In a serverless app, the hardest problem is getting all peers to agree on room state. By syncing append-only logs across the swarm, the sender sees live download speeds, receivers see accurate ETAs, and everyone sees who has completed a download in real time.
However, I had to take a custom approach for file transfer. By default, the Pear framework chunks each file into blocks and copies them into its own internal storage before seeding. When a filmmaker wants to share a 500GB or 1TB file, cloning it locally first is a non-starter.
During my research, I found Hyper Overlay, built by Fleeky in the Pear community. Instead of duplicating files on disk, Hyper Overlay reads the source file in place, chunks it into a lightweight in-memory hash map, and shares that map with peers. During transfer, blocks are read directly from the original file on disk, verified against the hash map over the wire, and reassembled on the receiver's drive. I worked closely with Fleeky through several iterations to refine it for massive payloads.
Q. How do you handle state synchronization and edit conflicts across peers without a central coordinator?
Oliver Kohl: Everything is built around encrypted, append-only logs. When you first open Mirall, there’s no registration or email required; your profile simply lives in a public Hyperbee. When you create a shared Space, it’s backed by an encrypted Hyperbee. Inviting someone sends them a discovery hash to find you on the DHT, and once approved, they receive the space key to decrypt the log and see the shared files.
From there, I treat state sync as an event-driven architecture split into two tiers. Critical events, like adding a new file to a space or changing permissions, require guaranteed consensus and are persisted in the append-only logs so every peer is guaranteed to receive them eventually, even if they were offline. On the other hand, real-time telemetry like transfer progress bars and current download speeds are sent as lightweight, fire-and-forget ephemeral updates. If one of those packets drops, it simply catches up on the next tick without bloating the permanent log.
Q. How do you keep local records and metadata secure against physical tampering or stolen devices?
Oliver Kohl: That actually took a dedicated week of security hardening. Using AI-assisted security auditing, we identified roughly 50 to 60 potential attack vectors across the app and systematically locked them down.
Today, all local application state, including your spaces, contacts, and file metadata, is encrypted at rest. Your root identity key is stored securely inside the operating system's native keychain (such as macOS Keychain or Windows Credential Manager). Even if someone steals your laptop, they cannot open Mirall or read your space metadata without access to that system keychain. We also built an encrypted backup and recovery phrase system so users can safely restore their identity and spaces onto a new machine.
Q. Looking at the broader ecosystem, where do you see Mirall fitting into the movement toward sovereign, serverless developer tools?
Oliver Kohl: Beyond human UX, I believe the future of software is heavily agentic, and AI agents will soon be the primary operators of workflow tools. To prepare Mirall for that, I’m decoupling the Bare backend worker from the Electron UI so it can run standalone as a headless system daemon, complete with a CLI and an MCP (Model Context Protocol) server.
In a film studio, an editor will soon be able to tell their AI assistant, "Send the latest render to the colorist in Montreal." The agent will talk to the local Mirall daemon via MCP, spin up an encrypted space, invite the remote studio's agent, and execute the P2P transfer automatically without human clicks or cloud uploads.
On the product side, I’m also working on multi-writer shared folders, allowing multiple peers in a space to contribute files into a single synced directory. That gives teams a true serverless alternative to Dropbox or Google Drive.
Q. What is immediately next on the roadmap for Mirall, and what would you say to people thinking about trying it out?
Oliver Kohl: We’re shipping updates every week or two, with a major focus on the Studio Console backend, headless agentic workflows (CLI/MCP), and multi-writer folders before the end of the year.
Right now, my biggest goal is getting Mirall into the hands of more users. User feedback has already shaped so many of our features. If you regularly move large files and are tired of paying for cloud storage, hitting file-size caps, or handing your data over to big tech platforms, give Mirall a try. It’s free, open-source, secure, and built to stay out of your way.



