49 posts filed under engineering.
Three months after publishing our open source rationale, here is the honest count: 14 external issues and 4 human pull requests this quarter, an all-time merge record written in two days by one contributor, and the requests we have not said yes to.
Why VisorCraft keeps splitting its catalog into small, focused, single-purpose projects instead of merging them into one suite, and the one place a suite is actually the right answer.
We run our entire desktop catalog on KDE Plasma 6 over Wayland every day, so this is the honest state report: what works, what still bites, and which defaults need flipping.
The honest version of what VisorCraft open sources and what stays closed: the rule fits in one sentence, and the interesting part is where the sentence bends.
Four VisorCraft repos are forks of upstream projects, and each has a different relationship to its upstream, because a fork is a tool with several legitimate uses rather than a declaration of war.
MongrelDB encrypts every page, WAL frame, cache entry, and index checkpoint with AES-256-GCM, and Roamarr stacks field-level encryption on top, because encrypted at rest only means something when the defaults cover the boring files too.
Every VisorCraft app ships GPL-3.0-only, and the choice was made deliberately against LGPL and against dual licensing, because copyleft is the whole point of publishing.
Streaming keeps repricing, fragmenting, and delisting the things you wanted to watch, so we run the library on our own hardware with Stash as the server and Sinema as the Android TV client, and in 2026 that stack is finally boring enough to recommend.
The lock-in of hosted tools is a pile of small papercuts, so we built Tailshell for the terminal and a tiny Mailspring plugin for the inbox.
Hosted travel apps ask for an account because your itinerary, passport scans, cards, and real-time location are unusually valuable once correlated, and the account is the correlating key, which is why Roamarr runs as one container on your own hardware with no account at all.
Roamarr, Foxden, SpiderTypes, Sinema, and Tailshell all ship without accounts on our infrastructure, and the interesting part is what that absence does to the data model once the users table stops being the vendor's.
Every VisorCraft project, including the closed-source MongrelDB, ships with no telemetry, no event hooks, and no opt-out switch, because there is nothing to opt out of.
Every VisorCraft app works fully offline, with no account, no cloud, and no telemetry, and this is what that constraint actually buys you once the cloud is just one option among many.
MongrelDB Kit's three language surfaces are pinned to the same JSON fixtures, so the same query produces the same row or the same typed error in TypeScript, Rust, and Python, and this is how we keep three SDKs honest without writing the same test three times.
Grexa and Arte-Ogre each ship as one Cargo workspace holding every crate the app needs, and the layout decisions behind that, shared version metadata, single dependency pins, and path-dependency layering, are what keep a Rust desktop app maintainable as it grows.
Three VisorCraft projects store their data as plain files in three different ways, and the reasons line up with three different access patterns, which is the actual decision framework for when the filesystem is the right database.
LinSight's remote monitoring path is an mTLS byte pipe between two processes you control, with no relay server and no account, and the honest answer is that most people should still use ssh -L instead.
LinSight's daemon serves its GUI and CLI over a postcard-framed Unix socket instead of HTTP, and the reason is that a monitoring hot path has no business parsing headers when it could be moving a number every 50 milliseconds.
Every VisorCraft app ships a CLI that exposes the full GUI feature set with stable exit codes and JSON output, and the reason is that GUI-driven testing is where CI budgets go to die.
LinSight loads metric sources as runtime .so plugins behind a stabby FFI vtable and an explicit ABI version, and the interesting part is the set of choices that keep the host alive across releases.
LinSight and LinSync write their QObjects in Rust and their UI in QML, glued by cxx-qt, and the interesting part is why the migration boundary sits exactly there.
Comparing Quest Toad's specialist database editions against Mongrel's unified workbench for engineers managing multiple engines and surrounding systems.
Eight of our desktop apps and libraries are Rust now, and the reasons have less to do with the language's reputation than with what quietly matured between 2020 and today, plus an honest list of where it still hurts.
Compare Aqua Data Studio's database IDE with Mongrel's broader systems workbench for databases, terminals, containers, Kubernetes, and APIs.
Compare Mongrel's unified workbench against a multi-tool stack of DBeaver, Postman, Termius, and Lens across workflows, safety, and operations.
Three of our Linux desktop apps run the same Qt 6 / Kirigami QML interface on KDE Plasma, GNOME, and XFCE without a fork in the codebase, and the interesting part is why that stopped being a gamble.
Comparing Lens Desktop's Kubernetes focus against Mongrel's cross-system investigation workbench, including database context, safety controls, and integrated tooling.
Mongrel brings HTTP, GraphQL, WebSocket, and gRPC requests into a unified workbench alongside databases, terminals, containers, and file operations.
underskrift signs and verifies PDFs at every PAdES level from B-B through B-LTA in pure Rust, and the hard part was PDF archaeology and conformance testing, not the cryptography.
Mongrel consolidates SSH, databases, containers, and APIs in one cross-platform desktop workbench, contrasting with MobaXterm's Windows-focused remote-access breadth.
Comparing Termius's cross-device session management with Mongrel's integrated desktop workbench for databases, APIs, and surrounding systems.
Five vendor tools cover five engines well, but polyglot stacks multiply windows, settings, and mental models. Mongrel brings 30+ engines into one workbench.
Compare Beekeeper Studio's approachable SQL client with Mongrel's broader workbench for databases, terminals, containers, Kubernetes, APIs, and team connection workflows.
Mongrel and DbVisualizer both connect across database engines, but their architectures, safety models, and system coverage diverge significantly across operational workflows.
Comparing Mongrel and TablePlus for database workflows, focusing on scope, safety controls, and when consolidation across databases, terminals, and APIs makes sense.
Comparing Navicat Premium's mature database suite against Mongrel's integrated systems workbench for developers managing databases and surrounding infrastructure.
Compare DataGrip's database IDE depth with Mongrel's database, terminal, API, container, and Kubernetes workbench for cross-system investigations.
Comparing Mongrel and DBeaver: database clients, architecture tradeoffs, workflow depth, and when a full-stack workbench makes sense.
How Mongrel unifies 30+ databases, terminals, APIs, containers, and Kubernetes while preserving engine semantics, safety, and platform limits.
Every production database lives behind a bastion, and most database GUIs treat SSH tunnels as an afterthought, so here is how tunneling actually works, why the port-forward incantation rots, and what it looks like when the tunnel manager and the database client are the same app.
MongrelDB-PHP is a pure-language HTTP client that targets PHP 8.5, uses no C extension, and leans on property hooks, readonly classes, and the pipe operator, because writing modern PHP in 2026 is genuinely pleasant and nobody is forcing it.
MongrelDB-Kit gives TypeScript, Rust, and Python the same schema-aware query builder, migrations, and constraint enforcement from one Rust core, and the interesting part is why one shared core beats three independent ORMs that slowly drift.
LinSync runs third-party document and image extractors inside a layered Landlock, seccomp, and bubblewrap sandbox so a compromised plugin cannot compromise the host.
LinSight is a system monitoring daemon that samples metrics on demand over a postcard-framed Unix socket, with a fully synchronous hot path, no Tokio, and no executor, and the absence of an async runtime is the design rather than the oversight.
IcefallDB stores tables as plain Parquet with JSON metadata and runs DataFusion on top, adding row-level mutations and warm aggregate caches to a file format that was never designed to support either.
grexa-db stores records as files and relational joins as directories of symlinks, turning every file-reading CLI into a database client without anyone knowing a database exists.
Arte Ogre's cut-and-paste correctness is pinned by a byte-identical round-trip test in CI. Here's why we test the data-model invariant and not the user-visible result.
Eight new pure-language HTTP clients shipped this week for MongrelDB. The total is now twelve. Here's the design call behind shipping them as HTTP, and what the C ABI means for the next wave.
We keep dozens of public repositories under GPL-3.0, Apache-2.0, and MIT. Here is some of what we open source, and the one thing we don't.
