Show HN: Ttar 2.4 KB freestanding TAR archiver written in C, no Libc

Show HN: Open-source Git-like Markdown docs for humans and agents

This is nothing special but it has been very very nice to kick off cleaning jobs remotely via MCP.  Working towards a genuinely smart home and this was a useful step in the right direction.

Show HN: SharkClean MCP

Hey HN. I&#x27;m a high schooler in Fremont, CA. Tuesday morning I got a county-wide AC Alert text telling everyone in Alameda County to prepare a go-bag for an East Bay Hills Red Flag Warning that starts tonight at 11 PM. The text went to ~half a million phones.The actual NWS warning polygon only covers East Bay Hills (NWS zone CAZ515). Most people who got the text don&#x27;t need a go-bag tonight. Some in the hills don&#x27;t realize how close they are. So I built this tool - <a href="https:&#x2F;&#x2F;redflag-check.info&#x2F;" rel="nofollow">https:&#x2F;&#x2F;redflag-check.info&#x2F;</a> mit licensed public github - <a href="https:&#x2F;&#x2F;github.com&#x2F;vedant-f-is-ma&#x2F;redflag-check" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;vedant-f-is-ma&#x2F;redflag-check</a>It does a few things - tells people if they are in the flagged zone, and also provides a way to check if a buddy is in flagged zone and send them a text. Everything without installing an app.I heard back from Oakland Firesafe Council director about a gap in my understanding (and the tool). To my surprise, and through feedback, I realized that you cannot assume that only the flagged area is at risk. Adjacent areas are at risk too! Fires do not follow zone boundaries! I fixed the tool.I built this in 48 hours to close that specific gap: type your address, get a yes&#x2F;no on whether the NWS polygon covers it, your Genasys evacuation zone, tonight&#x27;s wind + humidity at your point, a plain-English action checklist, a per-school decision view for East Bay districts, and a one-tap iMessage buddy-check template for a hill-neighbor at 10:30 PM.

Show HN: I built a Red Flag Warning zone-check tool for the East Bay in 48h

Last April I shared about my Resonate project here (<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=43694157">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=43694157</a>)A lot has happened since: the work I presented in much more detail at last June&#x27;s International Computer Music Conference (ICMC) got best paper award. I also gave a talk at the Audio Developer Conference in Bristol last November, the video is on YouTube).This year&#x27;s work, which I recently presented at this year&#x27;s ICMC, starts with known techniques from the phase vocoder literature to build self-tuning filter banks that extract very efficiently the frequency components that are actually present in the input signal. Overview on the project website, more details in the papers, including applications to super-resolution spectrograms and re-synthesis experiments.As many people have pointed out, none of the techniques I have used are new (some of them even have different names across different fields), but I haven&#x27;t seen them applied together in this way, and to me the results are incredibly satisfying and sometimes look magical. See for example this demo: <a href="https:&#x2F;&#x2F;youtu.be&#x2F;LasdoIJJkw8" rel="nofollow">https:&#x2F;&#x2F;youtu.be&#x2F;LasdoIJJkw8</a>Of course the best way to experience in person is through the free demo app: <a href="https:&#x2F;&#x2F;alexandrefrancois.org&#x2F;Oscillators" rel="nofollow">https:&#x2F;&#x2F;alexandrefrancois.org&#x2F;Oscillators</a>Looking forward to feedback from the community!

Show HN: Resonate – Low-latency, high-resolution spectral analysis

Show HN: Open-source API Key server written in Go by Ory

Hi HN. My name is Andrey. On a regular business day, I&#x27;m a software engineer working at AWS. Outside of work hours, I spend time on my hobby - writing code.I was once building a pet project that allowed customers to spin up fully synchronized blockchain nodes within just a few minutes. The backend was split into a control plane and a data plane, each with its own AWS account. Later I added two more AWS accounts. One for shared RPC nodes. One for the Analytics Service.Since I love to visualize things, I used drawio to visualize the architecture.With time, I noticed a pattern. I&#x27;d write some code, add a few lambda functions, update my drawio diagram, write more code, introduce a few more resources, test things, see that everything works fine and go to sleep with a smile on my face. Next week I&#x27;d check my diagram, and shockingly, it&#x27;s missing some of the resources! This kept happening for a few more weeks until I decided to fully abandon the project until my infrastructure diagrams could stay in sync with my cloud account.That&#x27;s how Atlasphere.io was born. I&#x27;ve been working on it for the past 6 months and I think the product is ready for some feedback :)A few notes:- Atlasphere uses a ReadOnly IAM role to scan your AWS account (my account reaches your account through a trust relationship).- The number of services is currently limited (WIP)- It&#x27;s a macOS app- It&#x27;s NOT an Electron app, i use Rust + WebviewWhat am I looking for? All I really need is for someone to try the app and tell me what they like about it and what they absolutely hate about it, haha!The website is <a href="https:&#x2F;&#x2F;atlasphere.io&#x2F;" rel="nofollow">https:&#x2F;&#x2F;atlasphere.io&#x2F;</a>

Show HN: Atlasphere – Live Infrastructure Diagrams

Show HN: macOS menu bar gauges for your Claude Code quota

What works now: user signups, org creations, private&#x2F;public repos, and importing GitHub repositories (both as read-only mirrors and full migrations). So basically, you can create, push and pull to a repo, but we don&#x27;t have many features quite yet (issues, PRs, CI).What is a bit unique is: 1) we built it in Rust and 2) the website is a little odd. Its design is inspired by CLIs (e.g., fzf, broot, vim) instead of web apps, and as such, lacks some affordances that you might typically expect in favor of keyboard-driven instant navigations (we have the very ambitious goal of an FCP of 100ms). In case you&#x27;re curious, here&#x27;s how we we built it: <a href="https:&#x2F;&#x2F;gitdot.io&#x2F;designs">https:&#x2F;&#x2F;gitdot.io&#x2F;designs</a>We recognize that we&#x27;re making some bold claims here and are also well aware that we have much to learn. Building software is still hard, and that&#x27;s a fact we seem to relearn everyday.But we wanted to share what we built so far nonetheless.Cheers, thank y&#x27;all for reading, and till the next
—paul &amp; mikkel.

Show HN: Gitdot – A better GitHub. Open-source, written in Rust

I&#x27;m a network engineer that likes to think about the future of the internet and this is what I&#x27;ve built over many nights and weekends. One reputation graph over IPs, ASNs, domains, and entities, exposed as a JSON API. Try it:<pre><code> curl https:&#x2F;&#x2F;api.tunnelmind.ai&#x2F;v1&#x2F;check&#x2F;1.1.1.1
</code></pre>
Every answer is a signed receipt with an attestation tier so you can see what was produced and how your agents can use it. The protocol is opensource. Try it out let me know what you think and yes I am still working on the radar section of the site. Also What would make this useful for you?

Show HN: TunnelMind – reputation API for IPs, ASNs, and ad-tech supply chains

Show HN: I applied Lyapunov stability theory to detect when LLM agents spiral

Show HN: ZeroFS – Make S3 your primary storage

Hey HN, cofounder of Artie here. We’ve built a real-time data replication tool that captures every row-level change in your source database and streams it to your warehouse in under 60 seconds.The last time I posted here, people had to book a call with us in order to access Artie. Today, that’s no longer the case. You can now connect your source and destination and start streaming immediately.I spent years of my career building large-scale data pipelines and experienced how difficult it was to get real-time data firsthand. I believed there must be a better way to stream data into our warehouse, which resulted in Artie being born. And now with AI agents, reducing data latency has become more and more crucial as agents need to make decisions off of fresh data.When I first started building Artie, I quickly learned that the components meant to keep CDC running smoothly are very much bolted on with tons of edge cases. Unfortunately in practice, they were not built to work together. We ended up dealing with schema drift, backfill race conditions, Kafka offset commits, and TOAST columns. I’d love to know if others have hit these same issues while building in-house.artie.com, would love feedback!

Show HN: Artie – Real-time data replication to your warehouse, now self-serve

Show HN: Geiger – A blast radius triage tool for any credential

Show HN: Vilvona AI – Self-Hosted AI Assistant with Tamil and Hindi UI

Show HN: Nenya – A lightweight, highly secure AI API Gateway/Proxy written in Go

Hi HN, I&#x27;ve been building Nucleus, a lightweight Linux container runtime focused on two workloads: ephemeral AI-agent sandboxes and declarative NixOS services. It&#x27;s a single Rust binary, no daemon.It is not a Docker replacement and not a strict subset of Docker either. I dropped the entire image-and-distribution half (no Dockerfile, no layers, no registry, no pull&#x2F;push, no persistent storage layer) in exchange for going deeper on isolation and reproducibility. The rootfs is either a directory copied into tmpfs (agent mode) or a Nix-built closure mounted read-only (production mode). If your mental model is &quot;run my image instead of docker run,&quot; this won&#x27;t fit. If it&#x27;s &quot;run untrusted or ephemeral workloads with stronger, auditable isolation on a single host,&quot; that&#x27;s the target.Things that I think are interesting:<pre><code> - Defense-in-depth defaults. All capabilities dropped, ~100-syscall seccomp allowlist (vs Docker&#x27;s ~300), up to 8 namespaces including time&#x2F;cgroup, Landlock LSM path ACLs per service.
 - Deny-by-default egress. Outbound traffic is denied unless you allow specific CIDRs or DNS-resolved domains. Enforced with namespace-local iptables rules.
 - Externalized, hash-pinned security policies. seccomp (JSON), capabilities (TOML), and Landlock (TOML) live as separate SHA-256-verified files, decoupled from the rootfs build. There&#x27;s a nucleus seccomp generate that records syscalls in trace mode and emits a minimal profile.
 - gVisor as a first-class integrated runtime, not an add-on. Explicit network modes including a gvisor-host mode that&#x27;s intentionally separate from native host networking.
 - Nix-native production path. nucleus.lib.mkRootfs builds locked-down closures; rootfs attestation verifies a per-file SHA-256 manifest at startup; first-class NixOS module.
 - Formal verification. TLA+ specs for the isolation&#x2F;resource&#x2F;filesystem&#x2F;security&#x2F;gVisor subsystems, checked with Apalache, plus property-based tests that drive the Rust implementation against the specs.
</code></pre>
Honest tradeoffs:
 - Linux x86_64 only. No macOS&#x2F;Windows&#x2F;BSD, no plans.
 - No CNI, no overlay networks, no cluster orchestration. nucleus compose is a single-host TOML DAG over systemd, not Swarm&#x2F;K8s.
 - Ephemeral-by-default storage. Persistence is opt-in via explicit --volume binds.
 - Agent mode applies several mechanisms best-effort by design (warn-and-continue on seccomp&#x2F;Landlock failure). For fail-closed isolation on ephemeral workloads use --service-mode strict-agent; for long-running services use production mode.Cold-start is ~12ms in the native runtime. Postgres 18 pgbench numbers under Nucleus are within noise of bare metal in our harness (full results in benches&#x2F;).

Show HN: Nucleus – A security-hardened, Nix-native container runtime

Show HN: Squishy & Friends – Claude Fabel 5 coded a game and it is good

Hi, in the last ~month I&#x27;ve learned a lot about ECS [1,2].I&#x27;m currently developing a robotics simulator from scratch (python+raylib) and, due to lack of game dev experience I went &quot;full OOP&quot; on it. A SceneObject with a lot of inherited interfaces (e.g. Collidable, Movable etc.). These are inherited and fixed at run time.The main loop inevitably became:<pre><code> for scene_object in sim.scene_objects:
 scene_object.update(...)
 for scene_object in sim.scene_objects:
 scene_object.draw(...)
</code></pre>
Well, it turns out that this can become a bottleneck if you have many scene objects because computers love contiguous memory for caching, physics&#x2F;math vectorization and so on.Having recently learned more about ECS[1,2], I started doing a bunch of experiments in a sandbox with turning the update() function into ECS. The idea is that the data is stored in columnar numpy arrays (components) + a lot of data structure optimizations for querying scene objects and fast access e.g.<pre><code> qr = scene.query(HasMotion, HasPosition) # query result acting like a np array of (N, ...) shape
 qr.position += ... # operate like numpy &#x2F; vectorized
</code></pre>
In any case, the standalone library only needs python and numpy. Raylib is only for rendering, but the raw data structures don&#x27;t need it.I&#x27;d love some feedback on it, e.g. what is it missing or what are gotchas I&#x27;ll find out later on during the simulator development.As an anecdote: I used Claude as an &#x27;engineering manager&#x27;, I wrote the code myself, it did the code review, tasks management and tests (super useful for corner cases).[1] <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=qglU107_DA4" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=qglU107_DA4</a> hytale&#x27;s ECS video (great for beginners)[2] Casey Muratori&#x27;s latest video about the first ECS in the game industry (<a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=73Do0OScoOU" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=73Do0OScoOU</a>)

Show HN: MicroECS – entity component system library in Python/NumPy

crowfly is a single-html-file golf round GPS tracker.Hit your shot, tap &quot;log shot&quot;, walk to your ball, repeat. Shots and rounds are stored in localStorage. Distance markers. CSV export. Zero account, zero backend.Short-term roadmap:- Mandatory login- AI club recommendation- Paywall everything- ERC-20 issuance- MCP servers- Even darker modeLong-term roadmap:- Humanoid agent plays the round for you- Configurable celestial body (Moon, Mars, Titan etc)- Round summary as a podcast- Rewrite in Rust

Show HN: Crowfly.golf – Zero-backend GPS round tracking (localStorage)

Hey HN!Lathe is an experiment in using LLMs to teach me something new, instead of doing the work for me. It generates a hands-on, source-backed tutorial for any technical topic you want to learn. Then you work through it yourself by reading and typing the code by hand (gasp) in a local UI built for exactly that.It&#x27;s a Go CLI plus LLM agent skills (Claude Code &#x2F; Cursor &#x2F; Codex). You prompt something like &quot;&#x2F;lathe build a 3D slicer in Erlang&quot;, run `lathe serve` to spin up a local webapp, and read it in your browser. Every tutorial comes with the things that have made self-learning a pleasant experience for me in the past:- table of contents that follows along as you scroll
- side-notes that nudge you to think
- exercises for the reader
- sources backing up the content that you can use to take you deeperTo help make up for the lack of human brainpower behind the tutorial, you can also ask questions about the content, have another LLM verify the tutorial actually compiles and runs, or extend it with another part (no more &quot;Part 4 of 6&quot; that hasn&#x27;t seen an update since 2021).I didn&#x27;t build lathe to replace human-written tutorials. I built lathe because I _love_ human-written tutorials, but wanted to learn technical domains where no good human-written tutorial exists yet (building a 3D slicer from scratch, making embedded Zig approachable, etc). There&#x27;s a longer story in the README about how I got started with programming through PSP homebrew tutorials, and why losing that to LLMs bugged me enough to build this.I&#x27;m not here to sell you anything (there&#x27;s nothing close to a VC-backed startup here :D). It&#x27;s an LLM, and its output is usually good but not perfect by any means. So far, my experience is that because you&#x27;re the one typing and actually engaged, you catch the weird stuff (and I&#x27;m finding that pushing back on it is its own kind of learning). And yes, it&#x27;s vibecoded, because it&#x27;s low scope, low risk, and scratching a personal itch. I run it on Claude Code + macOS personally, other setups should work but I haven&#x27;t been able to verify them yet.If you can find resources to learn something that was written by a human, read that first. But Lathe is here to fill in the gaps when that isn&#x27;t the case, and I hope it serves as an example where LLMs can help us think better, rather than less.Repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;devenjarvis&#x2F;lathe" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;devenjarvis&#x2F;lathe</a>Would love your feedback if you decide to check it out!

Show HN: Lathe – Use LLMs to learn a new domain, not skip past it

My frustration with distributing java apps didnt show up recently. 
I remember having implemented my first network jar downloaded back in the 2000&#x27;s because i needed applet like feature support with desktop full control.
Years after, the problem is the very same. Webstart didnt really took off and the only mean i had in my projects was the ugly fatjars, including the (for me) uglier spring-boot repackaging that changes the application classloading behaviour and hence giving me by time some headackes i was not prepared for.So basically nuts started as a response to this frustration 9 years ago, but from now i think its mature enough (used in production) to be shared, and forecebly i am more keen to need suggestions and help from fellow contributors.

Show HN: Nuts – pip/NPM for Java with first-class workspaces, JDK provisioning

I started off with a simple goal: port the game Soldat to Typescript. I ended up not only with the game, but with a full system to train AI bots to play my own game, building a neural net to train new bot AIs from tens of thousands of simulated matches with different strategies. It also runs live all the time on my website and has a news dashboard and you can replay any game exactly. It all was a very fun way to dig into Fable&#x27;s capabilities.The bill if I had done this with API billing? Nearly $2000.

Show HN: I had Fable clone a game, it turned it into an arena combat AI trainer

As a developer of typia, have to alive in the new typescript-go era, I made a new toolchain for tsgo.Making the ttsc, I understood that the new plugin toolchain system can dramatically reduce lint rule checking time by integrating into the typescript-go. So made `@ttsc&#x2F;lint`, and its time is 600x to 1600x faster than eslint.From now on, when you compile TypeScript, lint violations would be caught in the same time. No more need to run both `tsgo` and `eslint` independently.&gt; This repository is not v1 yet because the typescript-go has not been released yet, but if you wanna experience superfast lint, or keep using typia in typia era, you can do it right now.

Show HN: 800x faster Linter and TypeScript-Go toolchain for plugins like typia

Show HN: Chirp, My constraint-driven metaprogramming language

Hello HN, I am the founder of Agora Cosmica.This started about three years ago. On a walk I asked a chatbot to interpret the cave dream from Cormac McCarthy&#x27;s book &quot;The Road&quot; as C.G. Jung. It gave me a perspective I had not thought of. But for my own dreams the policies of the big providers felt wrong for so personal conversations, as zero data retention is not available. So I started building.The project evolved to a German nonprofit and we published the code (AGPL-3.0) last month. The content is still copyright, but will be opened to CC-BY 4.0 in the next 6 to 12 months.Agora Cosmica is a library to learn from 30 historical figures. Each one has 12 narrated stories about their teachings &#x2F; life wisdom, speech to speech conversation. Four learning modes and a council where you can gather the figures to discuss or reflect on a topic. Each figure is an AI Echo, an interpretation grounded in primary works, historical context, with a factcheck per figure to show what&#x27;s verified versus recreated.
On privacy: The speech is self hosted on Hetzner GPU servers, Qwen3-TTS for German, Kokoro TTS for English, Faster-Whisper for transcription. 30 free messages per day (EU-hosted for GDPR), BYOK, or you can run it in a full local self-hosted mode.No conversation is stored, no tracking cookies, no profiling, no signup.The app is slow on purpose. Cosmic, no dopamine rush.The mission is to be a doorway, a first step, an introduction to get people interested and outgrow the app to move to primary texts and human teachers.Live at: <a href="https:&#x2F;&#x2F;agoracosmica.org" rel="nofollow">https:&#x2F;&#x2F;agoracosmica.org</a>

Show HN: Learn from 30 historical figures, open source, nonprofit, self-hosted

Hi HN,Over the last few days I&#x27;ve been building USTPS (UDP Speedy Transmission Protocol Secure), an experimental encrypted transport protocol built on top of UDP.The primary goal of USTPS is low-latency video streaming. A server can take a video source and expose it through a USTPS endpoint, while Linux and Android (Termux) clients receive the stream and expose it locally to applications such as VLC, mpv, and FFmpeg.Although streaming is the main focus, USTPS is not limited to media delivery. It can also be used for other reliable encrypted UDP-based applications, which is why I built USSH on top of it.Some of the main design differences compared to TCP-based transports are:- USTPS is reliable but unordered.
- If packet N is lost, later packets can still be accepted and processed immediately.
- Missing packets are recovered through selective retransmission.
- Ordering is handled by the application layer when needed.This means the transport layer itself does not introduce Head-of-Line Blocking. The tradeoff is that applications which require ordering must implement reordering themselves. I consider this a reasonable tradeoff because it avoids forcing every application to pay the cost of transport-level ordering.For media player compatibility, the default USTPS client creates a local TCP endpoint at 127.0.0.1:1238.The client maintains a small reordering buffer (350 ms by default) to give retransmissions time to arrive before forwarding data to the local TCP stream. This allows existing software such as VLC, mpv, and FFmpeg to work without modification.USTPS currently provides:- Reliable delivery using ACKs and selective retransmissions
- X25519 key exchange
- AEAD encryption (AES-GCM and ChaCha20-Poly1305)
- Optional unordered live output mode
- Stream position metadata
- Multi-client support
- Local TCP compatibility output
- No congestion control (currently intentional)While developing USTPS, I also built USSH, an SSH-like remote shell running entirely over USTPS.USSH uses the same unordered transport underneath, but the client reconstructs and orders terminal data before presenting it to the user. This prevents terminal corruption while still allowing the transport layer itself to remain unordered.USSH includes:- Interactive terminal sessions
- PTY support
- Password authentication
- Host key verification (TOFU)
- End-to-end encrypted communication through USTPSI&#x27;m currently using USSH from my Android phone through Termux to manage my VPS.The project is very young (less than a week old) and is primarily experimental and educational. I&#x27;m interested in feedback from people working on transport protocols, streaming systems, SSH implementations, QUIC, SCTP, and networking software.USTP-Secure:
<a href="https:&#x2F;&#x2F;github.com&#x2F;x1colegal&#x2F;USTP-Secure" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;x1colegal&#x2F;USTP-Secure</a>USSH:
<a href="https:&#x2F;&#x2F;github.com&#x2F;x1colegal&#x2F;USSH" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;x1colegal&#x2F;USSH</a>Internet-Drafts:USTPS Draft:
<a href="https:&#x2F;&#x2F;datatracker.ietf.org&#x2F;doc&#x2F;draft-x1co-ustps&#x2F;" rel="nofollow">https:&#x2F;&#x2F;datatracker.ietf.org&#x2F;doc&#x2F;draft-x1co-ustps&#x2F;</a>USSH Draft:
<a href="https:&#x2F;&#x2F;datatracker.ietf.org&#x2F;doc&#x2F;draft-x1co-ussh&#x2F;" rel="nofollow">https:&#x2F;&#x2F;datatracker.ietf.org&#x2F;doc&#x2F;draft-x1co-ussh&#x2F;</a>Questions, criticism, and suggestions are welcome.

Show HN: Ustps (UDP Speedy Transmission Protocol Secure) and USSH

Show HN: I built a lead list tool to find businesses you can sell to

Anthropic&#x27;s usage analytics dashboard is only available to Team and Enterprise org admins. On a Pro or Max plan all you get is &#x2F;usage and the claude.ai usage page, which show where you stand right now but not where you&#x27;re heading. I looked at various open-source projects but none quite matched what I had in mind: an all-round control panel for Claude Code that&#x27;s also a single binary, with no dependencies and no install steps.The other thing I cared about was forecasting the usage limits. Existing tools do burn-rate projections (ccusage) or percentile heuristics (Claude-Code-Usage-Monitor), which felt too simplistic for what I wanted - I was after a calibrated statistical model with proper credible intervals.I built claumon over the last few months in Go. It runs on Linux, macOS and Windows, with a Homebrew install. It has the usual consumption gauges, cost breakdowns and conversation history, plus two tabs for memory management: after a while I had memories scattered across several projects and wanted to see them in one place and prune the stale ones.In the last few weeks I focused on the forecast model. I started from an empirical-Bayes linear regression with Brownian noise, but ended up with a Gamma process for the path noise: token usage can&#x27;t decrease over time, so Brownian motion, however mathematically convenient, was the wrong choice. The intervals are calibrated against your own recorded history, and there&#x27;s tooling that scores the forecasts out-of-sample, so the coverage is checked rather than assumed.I wrote a formal, versioned spec for the model, and the implementation follows it: <a href="https:&#x2F;&#x2F;github.com&#x2F;fabioconcina&#x2F;claumon&#x2F;blob&#x2F;main&#x2F;internal&#x2F;forecast&#x2F;MODEL.pdf" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;fabioconcina&#x2F;claumon&#x2F;blob&#x2F;main&#x2F;internal&#x2F;f...</a>Everything runs locally - nothing leaves your machine. It&#x27;s open source, MIT. I&#x27;d welcome feedback on the model especially.