Inside Vinteum’s recent retreats on DIY hardware wallets, Floresta, Stratum V2, and Bitcoin security
Open-source Bitcoin infrastructure is built by globally distributed teams, often operating with limited resources and little opportunity to work together in person. Many contributors collaborate daily through pull requests, issue trackers, and chat channels, yet have never shared the same room or worked side by side.
Vinteum has been working intentionally to change this dynamic. Over the past months, we brought together several Bitcoin infrastructure teams for small in-person developer retreats focused on collaboration, knowledge transfer, and long-term alignment. These are not conferences or networking events; they are working gatherings. We provide the means for developers to do what is difficult to achieve remotely: to sit together, review code, discuss architecture, transfer context, and build shared understanding. The goal is to foster depth, continuity, and serendipity for under-resourced teams working on critical layers of the Bitcoin stack.
We have repeatedly seen how in-person time accelerates progress in ways that online collaboration struggles to match. Days spent working side by side can compress months of asynchronous discussion. Equally important are the informal moments that happen around the working hours. Some of the most valuable insights emerge outside scheduled sessions, during meals, walks, or late nights, where contributors connect ideas across projects and challenge assumptions in a low-pressure setting. These conversations are hard to plan for and nearly impossible to reproduce online. A crucial side-effect is to build up trust among people that work on a shared goal.
This is why in-person work has become a core part of Vinteum’s approach to support Bitcoin infrastructure development. In the sections below, we outline the developer retreats we have organized recently, their focus areas, and how this format helps teams move faster, align better, and build more resilient open-source projects.
DIY Hardware Signers Retreat
We organized a week-long, in-person retreat at Casa21 in São Paulo to bring together contributors from three open-source DIY Bitcoin hardware wallet projects: Krux, Specter-DIY, and SeedSigner. While these projects have evolved independently over the years, they all rely on a shared technical foundation: Embit, a minimal Bitcoin library for MicroPython and Python3 with a focus on embedded systems, created by Stepan Snigirev and widely used across embedded signing devices.
Before the retreat, there was little to no collaboration across DIY hardware signer projects, and in some cases, contributors within the same project had never met in person. Each team evolved largely in isolation, developing its own practices and assumptions, with limited visibility into how other projects were approaching similar security, UX, and interoperability challenges.
By bringing together nearly the full active teams behind Krux, SeedSigner, and Specter-DIY for a week of in-person work at Casa21, the retreat created shared context across projects and within teams, laying the groundwork for deeper technical alignment and sustained collaboration across the DIY hardware wallet ecosystem.
The week was structured around deep technical work, with emphasis on Embit. Stepan led multiple sessions walking through Embit’s architecture, design constraints, and trade-offs specific to embedded systems, including memory limitations, MicroPython compilation strategies, PSBT handling, and testing approaches across Python and MicroPython environments. These sessions were not lectures but working discussions, often grounded in real constraints faced by each project in production.
A key concern we wanted to address was Embit’s future as a shared dependency across multiple projects. Despite all teams relying on the library, its development workflow had become fragile, with stalled security improvements and unclear ownership, especially as its original creator, Stepan Snigirev, has stepped away from active Bitcoin development. The group aligned on Embit’s long-term sustainability, maintainership model, and roadmap, moving it from a library implicitly stewarded by a single maintainer to a shared dependency with clearer ownership and cross-project collaboration. This handover marked an important maturation point for the DIY hardware wallet ecosystem.
Beyond Embit, we wanted to pollinate cross-project knowledge among the projects. The teams compared approaches to secure bootloaders, QR encoding and decoding, stateless versus encrypted storage models, Miniscript support, and UX trade-offs in air-gapped signing devices. Several discussions centered on how to reduce duplicated effort across projects, standardize interfaces where it makes sense, and improve interoperability without forcing uniformity.
Equally important were the human-factor outcomes. As one Krux contributor put it, meeting in person made it possible to “build shared dreams,” not just shared code. Another participant described the retreat as a potential catalyst for elevating DIY hardware wallets from a niche approach used by power users to a credible recourse for anyone serious about securing their savings.
The week offered a moment for long-term contributors to reflect on how far the space has come and to intentionally plan what collaboration could look like next. We believe the teams left with clearer technical direction, stronger personal connections, and renewed momentum. Krux, Specter-DIY, and SeedSigner now have a shared context that will continue to shape decisions long after the week at Casa21.
Floresta Developer Retreat
One of the hardest problems in Bitcoin development is turning protocol research into software that real users can run and rely on. Many promising ideas fail not because they are flawed, but because the gap between theory and implementation is too wide. Floresta sits at the intersection of some of the most ambitious efforts to make Bitcoin nodes lighter without compromising on validation or security. By combining Utreexo, proof-of-work fraud proofs, and alternative trust-minimized syncing models, the project operates at the boundary between protocol research and real-world implementation. The Floresta developer retreat at Casa21 was designed precisely to narrow that gap.
We brought the Floresta core team to São Paulo for an in-person retreat alongside key collaborators working on these protocols. Floresta’s maintainer, Davidson Souza, plays a central role in this effort as both the lead developer of Floresta and a coauthor of the Utreexo BIPs alongside Tadge Dryja and Calvin Kim. Calvin is also the maintainer of utreexod, a btcd fork that implements utreexo and serves as a bridge node for Floresta.
The retreat focused on deep mentorship around Utreexo. Floresta contributors spent extended time with Tadge and Calvin reviewing the accumulator’s architecture, design trade-offs, and edge cases that only surface when moving from theory to practice. These discussions went beyond high-level explanations and into concrete analysis of performance, proof sizes, caching strategies, and operational constraints observed in real deployments.
Several sessions focused specifically on utreexod, which has historically been memory-intensive and costly to run. Having Davidson, Calvin, and the broader Floresta team in the same room allowed the group to jointly identify bottlenecks, challenge existing assumptions, and explore alternative designs grounded in hands-on experience.
Having all Utreexo BIP coauthors present proved especially valuable. The group worked through the draft BIPs collaboratively, clarifying ambiguities, debating design decisions, and aligning on how the specifications should communicate intent, security assumptions, and implementation guidance. These conversations were grounded in Floresta’s existing codebase and informed by lessons learned from both Floresta and utreexod.
A major outcome of the retreat was the decision to incorporate Ruben Somsen’s SwiftSync protocol into both Utreexo and Floresta. SwiftSync reframes initial block download as a verification problem rather than a computation-heavy process, significantly reducing disk I/O pressure and overall resource requirements. Aligning on this direction across projects opened a path toward dramatically faster and more resource-efficient IBD for Utreexo-based nodes.
The retreat also created dedicated space to strengthen the Floresta codebase itself. Contributors used the time for joint debugging, architectural reviews, and roadmap planning, benefiting from direct feedback from the protocol researchers.
The Floresta team left with greater confidence in both the protocol foundations they are building on and the direction of the project. The retreat clarified a path forward for the Utreexo BIPs, aligned multiple implementations around shared performance goals, and strengthened the relationship between protocol researchers and implementers working on one of the most promising approaches to lightweight, fully validating Bitcoin nodes.
Fuzzing and Security Retreat
Alongside the 2026 Exploits Edition of Bitcoin++, where Vinteum is co-hosting and acting as a primary sponsor, we organized a small, focused fuzzing and security retreat in partnership with Brink. The goal was to take advantage of the concentration of security-focused developers already gathered around the conference and extend that momentum into a few days of dedicated, in-person collaboration.
Both Brink and Vinteum have become two of the organizations most consistently focused on supporting Bitcoin security work. This includes not only direct contributions to Bitcoin Core, but also sustained investment in testing infrastructure, fuzzing, audits, and the next generation of security contributors.
At Vinteum, this work is represented by Bruno Garcia, who combines active Bitcoin Core security work with his role as lead maintainer of bitcoinfuzz, and Erick Cestari, who leads efforts on differential fuzzing of Lightning implementations. Bruno has also been mentoring developers in India who emerged from programs such as Summer of Bitcoin and Bitshala. At Brink, Niklas Gögge and Marco Leon are both deeply involved in Bitcoin Core security, contributing directly to the codebase as well as working on projects such as Fuzzamoto, a coverage-guided fuzzing framework. Brink has also recently added Eugene Siegel to its team, an experienced security-focused developer previously at Lightning Labs. Brink has also expanded its fellowship and internship programs, while also funding external security audits and other initiatives.
During the retreat, contributors spent several sessions collaborating on improvements to Bitcoin’s fuzzing and testing infrastructure. Fuzzing is a software testing technique that automatically generates large amounts of random, malformed, or unexpected inputs and feeds them into a program to discover bugs, crashes, or security vulnerabilities. Instead of manually writing test cases, a fuzzer explores the input space - often guided by code coverage or feedback from previous runs - to trigger edge cases and unintended behaviors. It is especially effective for finding memory corruption issues, logic errors, and parsing flaws in systems like compilers, network protocols, and applications that process untrusted data such as a Bitcoin full node. Many Bitcoin projects rely on fuzzing to ensure reliability and safety. However, working on fuzzing is not as simple as it seems. Ensuring determinism, writing good harnesses, maintaining a good infrastructure and others, brings much work on it.
Fuzzing a distributed system like Bitcoin Core has even more challenges, especially if we want to simulate real scenarios and interactions between the components. Sometimes, to properly fuzz the P2P layer of a Bitcoin full node requires a refactor of this component and since it is not trivial to do, it may not be worth it. That is one of the reasons why Niklas created Fuzzamoto, a framework and fuzzing engine for coverage-guided fuzzing of Bitcoin full node implementations, which was one of the topics of the retreat.
They worked together and discussed some enhancements to Fuzzamoto, such as incremental snapshots; better support for testing BIP153 - by developing a feature that would allow users to load seed files to, for example, prepopulate the mempool before taking a snapshot; improvements to compact block reconstruction by including more transactions beyond just the coinbase; some performance improvements, and others. Beyond Fuzzamoto-specific work, the group discussed related areas such as hypervisors, property-based testing tools, performance improvements for fuzzing, and the creation of dashboards to better track Bitcoin Core’s fuzzing efforts.
The retreat also created space for contributors to share their individual workflows and approaches to fuzzing, testing, and vulnerability research. Participants compared how they structure fuzzing campaigns, triage crashes, and prioritize targets across different parts of the Bitcoin codebase. Another topic that surfaced repeatedly was how large language models might assist security and development research. While LLMs are not a replacement for careful review, contributors discussed how they can be used to leverage their work.
Security work rarely produces visible features, but it plays a critical role in ensuring that Bitcoin’s infrastructure remains robust as the system grows. Retreats like this help strengthen the network of contributors responsible for maintaining that foundation.
Stratum V2 Retreat
Another retreat focused on the Stratum V2 ecosystem and open mining infrastructure. While the protocol has matured significantly, deploying it in practice still requires coordination across multiple layers of the mining stack, including pool software, mining firmware, protocol implementations, and the Bitcoin node itself.
One of the main discussions centered on improving the developer and user experience of running Stratum V2 infrastructure. A presentation on the SRI UI project outlined a roadmap to simplify deployment. Today, running the reference implementation often requires managing multiple binaries and manual configuration. The proposed direction is a phased approach: first a Docker-based single-command deployment, then integration with environments such as Umbrel and Start9, and eventually a single sv2-client binary that bundles the Job Declaration Client (JDC) and Translation Proxy (tProxy).
The UI prototype demonstrated during the retreat included a configuration wizard to guide users through setup decisions such as solo mining versus pool mining, pool selection, and network configuration. One important UX issue discussed was how pools batch share acknowledgements to reduce bandwidth. Because shares can appear as submitted but not yet accepted, miners may misinterpret this as rejected work. Improving how this is presented in the interface was identified as a priority.
Another discussion explored how Stratum V2 could integrate with the growing ecosystem of home mining devices. Bitaxe hardware was highlighted as a promising platform for early adoption, with Stratum V2–compatible pools such as SRI Pool and Blitzpool serving as endpoints for home miners experimenting with the protocol.
The retreat also included technical sessions on the internal architecture of the Stratum V2 implementation, including the Rust protocol crates responsible for message framing, Noise encryption, codecs, and buffer management. A separate session reviewed the current fuzzing infrastructure and discussed expanding fuzzing targets as well as introducing differential fuzzing across implementations.
Sjors Provoost joined the retreat to discuss the new Bitcoin Core IPC mining interface, which allows external software such as Stratum V2 components to interact with Bitcoin Core through a modular Cap’n Proto–based architecture. The discussion explored how the Job Declaration Server (JDS) interacts with Bitcoin Core and the challenges of maintaining a mempool mirror to validate mining jobs locally.
Two residents from the Bitcoin Dev Launchpad also participated in the retreat: Johnny Santos, contributing to P2Pool v2, and Jayr Motta, who recently began working on Mujina. In both cases, the goal was to help them better understand the Stratum V2 ecosystem and explore how the protocol could be integrated into the projects they are contributing to.
Taken together, the sessions highlighted how much of the remaining work around Stratum V2 lies at the boundaries between components of the mining stack: node interfaces, pool infrastructure, firmware, and testing. Aligning contributors working across these layers is essential for turning the protocol into production-ready mining infrastructure.
Looking Forward
These retreats reinforced a simple observation: many of the hardest problems in Bitcoin development are not purely technical, but organizational. Ideas, protocols, and implementations improve significantly when the people building them have time and space to think together.
Vinteum will continue to support this kind of work by bringing together developers across different layers of the Bitcoin stack. Whether the focus is hardware signing devices, node implementations, mining infrastructure, or security research, our goal remains the same: help under-resourced teams collaborate more effectively and build the open infrastructure Bitcoin depends on.
Acknowledgments
These retreats would not have been possible without the support of the Human Rights Foundation (HRF), which helped fund several of them, including the DIY Hardware Signers, Floresta, and Stratum V2 retreats.