GobboNet lets you run an AI chatbot on your own computer, with no accounts, no monthly fees, and nothing sent to the internet. The AI lives on your machine. When you talk to it, your words never leave your home.
This guide assumes you've never set up anything like this before. Take it one step at a time and you'll be chatting in about 20–30 minutes (most of that is just waiting for files to download).
- Llama — working
- Gemma — working
- DeepSeek — working
- Qwen — working [Qwen3 30B-A3B, QwQ 32B]
- gpt-oss — working [20B and 120B, Harmony format]
- Mistral — working [Mistral Small 24B and its finetunes, including Cydonia and Magidonia]
- Granite — working
- Command R — working
- GLM — working
- There's a small launcher (
launch.exe) that starts everything for you. - There's the AI engine — the program that does the actual thinking. It's about 300 MB and downloads once.
- There's the AI model — this is the "brain." It's a big file (usually 4–16 GB). You pick one from a menu and it downloads once.
- There's a second, much smaller retrieval model (~146 MB) that also downloads once. It's what lets a character pull the relevant parts of its own notes and lore into a conversation instead of re-reading all of it every turn.
- There's the chat screen itself, which opens in your web browser like a normal website — except it's running entirely on your own computer.
You download the engine and a model one time. After that, everything runs offline, forever.
- A Windows 10 or Windows 11 PC. (This doesn't work on Mac or Linux as written.)
- A graphics card (GPU) helps a lot. The AI runs much faster on a graphics card. It will still run without a good one, just slowly. The setup checks your hardware and recommends a model that fits.
- Free hard-drive space. At least 10–20 GB free, ideally more. The AI models are large.
- An internet connection for the first setup only. After the one-time downloads, you can unplug from the internet and it still works.
You do not need to be technical, create any account, enter a credit card, or install anything complicated. Just downloading and running the exe from the released versions will do all of this for you.
Go to the Releases page and download GobboNetSetup.exe. It's small (660 KB), it installs just for you, and it doesn't need admin rights.
Double-click it. Windows will probably say "Windows protected your PC" — click More info, then Run anyway. That warning appears because the installer isn't signed with a paid certificate, not because anything is wrong with it.
When it finishes you'll have GobboNet in your Start Menu and on your desktop, along with an uninstaller that asks before it goes anywhere near your downloaded models.
That's the whole install. Skip ahead to Step 2.
If you'd rather not run an installer, click the green Code button at the top of the repo, choose Download ZIP, and extract it somewhere easy to find, like Documents\Gobbonet.
Keep the folder exactly as it comes out of the ZIP. GobboNet isn't a handful of loose files anymore — chat.html loads its styling from the css\ folder and its code from the js\ folder. If those get flattened together or left behind, the chat screen opens as a blank, broken page. Don't pull files out of the repo one at a time either; grab the whole ZIP.
You should end up with this:
Gobbonet\
├── launch.bat <- start here
├── gobbonet.exe <- the server launch.bat hands over to
├── setup-lan.bat <- phone access, run as Administrator once
├── stop-gobbonet.bat
├── teardown-lan.bat
├── fileserver.ps1
├── hardware-probe.ps1
├── hw-recommend.ps1
├── identify-model.ps1
├── engine.sha256 <- which llama.cpp build, and its checksum
├── chat.html <- source of the chat screen (built into the .exe)
├── default-characters.json
├── gobbonet.ico
├── css\ (17 files) <- source, as above
├── js\ (24 files) <- source, as above
├── linux-amd64\ <- the Linux build; Windows users can ignore it
├── docs\ reference notes and the per-change changelog
└── tests\ developer tests; not needed to run anything
A models folder and a few small bookkeeping files (models-list.json, active-model.json) get created for you on the first run — that's normal, you don't need to make them yourself.
Going this route, run launch.bat wherever the rest of this guide says
launch.exe, and setup-lan.bat where it says LAN Setup. Everything else
works the same.
launch.bat is the black-and-green window that walks you through everything —
setting a password, downloading the AI engine, checking your hardware, picking a
model. When it has your answers it hands over to gobbonet.exe, which serves the
chat and keeps running in that same window, so the model loading and the GPU
check appear right where you have been reading. You do not need to start
gobbonet.exe yourself.
One cosmetic difference: the installer bundles a font that the source ZIP does not, so a browser console opened on a source install will show a 404 for fonts/atkinson-hyperlegible.woff2. Nothing is broken — the interface falls back to a monospace font your system already has. The font is left out on purpose rather than committed, so that GobboNet never fetches one from a font CDN at runtime.
Double-click the GobboNet shortcut on your desktop, or launch.exe in the install folder.
A black window with green text will open. This is normal — it's the launcher talking to you. Read what it says, because it asks you a few simple questions the first time.
The very first time, it asks you to choose a password. Type one (at least 6 characters) and press Enter, then type it again to confirm.
This password protects your chat. Anyone else on your home Wi-Fi could otherwise open it, so the password keeps it private. You'll only type it again when you connect from your phone.
Your password is stored in a scrambled form that can't be reversed, and it is never written down anywhere in readable form. If you forget it, you can set a new one (see Troubleshooting).
When you sign in from your phone, the password itself travels across your home network unencrypted, the same as everything else here. The sign-in page says so too. That is fine on a network you trust — but it is why you should not reuse a password that matters elsewhere, and why you should not run this on shared or public Wi-Fi.
Next it says the engine isn't installed yet and asks "Download llama.cpp now?" Type Y and press Enter. This is the ~300 MB one-time download. Wait for it to finish.
If you don't have a model yet, it checks your computer's hardware and shows a menu like this:
Detected: 16 GB VRAM, 32 GB RAM, 423 GB free disk
-- SMALL (fits ~6 GB VRAM) ----------------------
[2] Llama 3.2 3B Instruct Q8_0 ~3.4 GB
[1] Gemma 4 E4B IT Q4_K_M ~5.4 GB
-- MEDIUM (fits ~8-12 GB VRAM) ------------------
[4] Qwen3.5 9B Q4_K_M ~6.2 GB
...
[8] gpt-oss 20B MXFP4 ~12 GB [ RECOMMENDED FOR YOUR PC ]
-- LARGE (16 GB VRAM and up) --------------------
[5] Gemma 4 26B-A4B MoE Q4_K_S ~16 GB [ needs ~18 GB VRAM - will be slow ]
...
-- MANUAL ---------------------------------------
[11] Skip - I'll add my own .gguf
One option is marked [ RECOMMENDED FOR YOUR PC ] — that's the best fit for your hardware. Just press Enter to accept it, or type a number to pick a different one. Then wait while it downloads (a big model can take 10–30 minutes — this is the longest part, and it only happens once).
Notice that the 16 GB card above is not offered the 16 GB model. A model needs room beyond its own size — for the conversation it is holding in memory, for the engine's working space, and for whatever your monitor is already using. Every recommendation leaves at least 2 GB spare, which is why the biggest thing that fits is rarely the right pick.
Models that want more than you have are labelled with what they actually need, and picking one anyway will ask you to confirm first. When in doubt, take the recommended one.
Once your model is sorted, the launcher fetches the retrieval model (~146 MB, one time). This is what lets a character pull the relevant pieces of its lorebook into a conversation instead of trying to hold everything at once.
It runs on your processor rather than your graphics card, so it won't take memory away from the model you're actually chatting with. Like everything else here, it never phones home after the download.
Once the model loads, your web browser opens automatically to the chat screen. You can start typing.
Every time you want to use it in the future, you just open GobboNet again. After the first setup, it starts in well under a minute and the AI itself never needs the internet.
Two things reach out, and both are yours to turn on: web search, if you set up a key for it, and any extension you load from a web address rather than pasting in directly. Neither is on unless you switch it on.
You can chat from your phone or tablet as long as it's on the same Wi-Fi as your PC. This is optional — skip it if you only want to use it on the PC.
-
Open LAN Setup from your Start Menu (or
launchLAN.exein the install folder). It tells you exactly what it's about to change before Windows asks for admin permission — say yes to the prompt. This opens the door for your phone to connect. You only ever do this once. -
Look at the launcher window. Under phone / LAN it prints the addresses to try, with the name of the network adapter each one belongs to:
phone / LAN: http://192.168.1.24:9066/ (Ethernet) http://10.0.0.57:9066/ (Wi-Fi) -
On your phone's browser, type the top one. Enter the password you chose. Done.
If there's more than one address, that's normal — it means your PC has more than one network adapter, and only your phone knows which one it shares. Work down the list. The line whose adapter name matches the Wi-Fi your phone is on is the one that will work. Anything marked virtual adapter or VPN or tunnel is listed for completeness and is almost never the answer.
The LAN Setup window also prints a http://your-pc-name.local:9066 address at the end of its run. That one is worth bookmarking where it works, because it survives your PC's network address changing and so keeps your saved chats attached to the same address. It relies on a feature called mDNS, which iPhones and iPads handle well and some Android browsers do not — if it doesn't load, use a number address from the list above instead. Nothing is wrong; that's just Android.
Gobbonet is more than a plain chatbox. Here are the parts you'll actually use, in plain terms:
-
Just type and chat. Type in the box at the bottom, press Enter. That's the basics.
-
Characters. It comes with a few built-in personalities (a terse coder, a wordy lore-keeper, a riddle-speaking oracle). You can switch between them or make your own — give it a name, a description, and a style, and the AI will play that role. You can also bring in character cards you already have, and send yours back out (the common
.pngcards used by other AI chat apps), so your existing collection works here too. Most cards carry over cleanly, though a few may need small tweaks after importing. -
lorebooks. A character can carry a lorebook — lore, world details, backstory, notes, whatever you want it to know. Rather than cramming all of that into every single message, GobboNet works out which bits are actually relevant to what you're talking about right now and quietly slips those in. It means a character can "know" far more than would ever fit in one conversation.
-
Threads. Each conversation is saved separately in the sidebar, like chat history. You can rename them, pin favorites, and sort them into folders.
-
Web search (optional). There's a search button that lets the AI look things up online. It needs a free key — see below. Off unless you turn it on.
-
Adding models (optional). Config → Add a Model shows a catalogue you can download from without leaving GobboNet or working out HuggingFace's interface. GobboNet downloads the file, checks it, and puts it where it belongs. The list itself is a small public file fetched from goblincorps.com so it can be kept current — a plain download with nothing about you or your machine attached, cached for a day, and switchable off with
model_catalog_remote = false. If it can't be reached, the list that shipped with GobboNet is used instead and downloading still works. -
Switching models. If you've downloaded more than one model, you can switch between them from a dropdown at the top — no need to restart.
-
Using a server you already run (optional). If you already have llama.cpp or Ollama running, GobboNet can talk to that instead of starting its own engine. Open
config.toml, setserver_exe = ""and pointllm_urlat your server:server_exe = "" llm_url = "http://127.0.0.1:11434" # Ollama's default; llama.cpp is usually 8080
The model dropdown then lists whatever your server reports, and picking one just changes which model the next message goes to — there's no restart, because GobboNet isn't the one running it. Everything else works exactly the same: characters, lorebooks, threads, RAG, web search. Any
.gguffiles you also have inmodelsstill show up in the same dropdown alongside them. -
Saving files. If you ask the AI to write something like code or a document, it can give you a Save button to download it.
-
Settings. A gear/settings area lets you tweak things, change the look, and manage your data.
The search button needs a free key from Ollama:
- Go to ollama.com, make a free account, and get an API key.
- In Gobbonet, open Settings and paste the key into the API-key box.
- Now the search button works. (Searches go through Ollama; everything else stays on your machine.)
If you never set this up, the chat still works perfectly — you just won't have the live-search button.
This tool was built to be private, but a few honest notes:
- The AI itself is 100% offline. Your conversations are not sent anywhere. There are no accounts and no tracking.
- The password keeps other people out. Anyone on your Wi-Fi could otherwise reach the chat, so don't disable it.
- The connection is not encrypted. This is fine on a home network you trust. Don't run this on public or shared Wi-Fi (like a coffee shop), and don't reuse an important password here.
- Your chats are saved in your browser. If you clear your browser data, or switch browsers, your history may not carry over (though the app tries to back it up and offer to restore it).
Replace one file: gobbonet.exe. That's the whole update.
Download the new ZIP, open it, and copy gobbonet.exe into your GobboNet
folder, replacing the one already there. Your models, conversations, characters
and password are all kept — none of them live in that file.
The chat screen is built inside gobbonet.exe, so you cannot end up with a
new chat screen and an old program, or the other way round. That used to be the
most common way an update went wrong: people copied chat.html, js\ and
css\ across, which changed how the app looked while everything that runs
behind the scenes stayed on the old version. Features would appear in the menus
and then do nothing.
Windows won't let you replace it while it's running. Close GobboNet first (see Shutting it down, below). If Windows offers to Skip the file because it's in use, don't — that leaves you half-updated. Close GobboNet and copy it again.
Copying the rest of the folder across as well is harmless. Nothing else is
required, and a leftover web\ folder from an older version is ignored — the
launcher window says so when it starts, and you can delete it.
Checking it worked. Open the ≡ menu, choose ABOUT, and look at the version. If something is out of step, that panel says so and what to do about it.
Did you install with GobboNetSetup.exe? Then run the new
GobboNetSetup.exe instead; it replaces the file for you and keeps your
models.
On Linux? Replace the whole linux-amd64 folder, or install the new
.deb/.rpm over the old one.
To stop everything: find the launcher window (the black window with green text — it shrinks to your taskbar after a few seconds) and close it, or click it and press Ctrl + C.
That window quietly watches the AI in the background and restarts it if it ever hiccups, so it's normal for it to stay open while you're using the chat.
The chat says it can't connect to the AI / it's very slow.
The model may be running on your processor instead of your graphics card, which is slow. Look at the GobboNet window: it says GPU acceleration confirmed once the model has loaded, or prints a warning with the likely causes if it could not confirm it. The engine's own loading output appears there too, marked [llama]. Make sure your graphics drivers are up to date (search your card maker's site: AMD, Intel, or NVIDIA). If it's still slow, pick a smaller model next time.
The chat screen is blank, unstyled, or looks broken.
The css and js folders aren't where the app expects them. This only happens on a manual install — re-extract the ZIP and keep the folder structure exactly as it comes, or just use the installer.
My phone can't connect. Make sure your phone is on the same Wi-Fi as the PC. Then run LAN Setup from your Start Menu one time.
If it still won't connect, try every address under phone / LAN, not just the first. A PC with both an Ethernet cable and Wi-Fi has an address on each, and only the one matching your phone's network will work. The adapter name in brackets after each address tells you which is which.
If none of them work, run this from the install folder and read the LAN ACCESS section it prints:
gobbonet doctor
It says whether the firewall rule exists, whether it's on the right port, and whether something is blocking the app despite the rule.
My saved chats disappeared.
This usually happens when you opened the chat at a different web address than before (for example, the PC's network number changed). The app keeps a backup and will usually offer to restore it. To avoid this entirely, always use the .local address and bookmark it.
I forgot my password.
You can set a new one. Delete the hidden file named .gobbonet-secret from your GobboNet folder and relaunch — it will ask you to set a fresh password. If you installed the manual way, you can instead open the launcher folder, hold Shift and right-click in empty space, choose the terminal/command option, and run:
launch.bat reset-password
Either way, restart your computer after this or it will not take properly.
The download failed. Run the launcher again — it will pick up where it left off. If the engine or a model repeatedly fails, your internet may be blocking it; try again on a different connection. The launcher also prints manual download links if you'd rather grab the files yourself.
A model is too big and crashes or errors during chat.
It's running out of graphics memory. Use a smaller model from the download menu, or ask whoever set this up to lower CTX_SIZE at the top of launch.bat.
Sometimes an error code shows up. Here's what the common ones mean and the quickest fix for each.
The chat won't load, or you see a 500 error.
Your model is too big for your graphics card's memory. The "context limit" (how much the AI can read at once) plus the model size is asking for more memory than you have. Use a smaller model, or have someone lower CTX_SIZE at the top of launch.bat.
The model won't load, or you see a 502 error.
A 502 means GobboNet is running but the AI engine behind it is not answering. Check the front page of the chat first — where it normally says the model is online, it will now say what actually went wrong, in the engine's own words. That is usually the whole answer.
If it says something about memory, the model is too large for your graphics card; pick a smaller one. If there is no explanation to be had, the most common cause is another AI program called Ollama starting up on its own and taking the port GobboNet needs. Close Ollama (check your system tray near the clock, right-click its icon, and quit it), then run the launcher again.
The web address won't load on my phone.
The phone address can change when the PC or its network restarts. First, double-check you're typing the exact address the launcher window currently shows — it may have changed since last time. Bookmarking the .local address (instead of the number address) avoids this.
Switching models gave me a 404 error.
The model file is probably in the wrong place. Make sure every .gguf file is inside the models folder, directly — not in a subfolder and not loose in the main folder.
Switching models gave me a 502 error.
Open Task Manager (press Ctrl + Shift + Esc), find anything labeled "Windows PowerShell," and end those tasks. Then run the launcher again. This clears out a stuck background process.
I was mid session and everything broke! The LAN access for mobile access has changed on the variable address. There is a more stable address listed right above it.
We like to be upfront about what's broken. We'll update this section as we squash them.
Logit bias is broken: (this is what powers "banned words"). The banned-words feature — the one that's supposed to discourage specific words from showing up — relies on logit bias under the hood, and logit bias isn't working right now. Banning a word won't reliably keep it out of replies. For the moment, treat the feature as non-functional rather than just hit-or-miss. (You'll see this flagged in the feature list too.)
Mistral Small models talking nonsense — fixed in 1.7.2. Cydonia, Magidonia and other Mistral Small 24B descendants would ignore their character entirely and answer in French, or start writing academic papers. The cause was one character: GobboNet was framing the conversation with a space where those models expect none, which was enough to stop them recognising that a conversation was happening at all. They now use their own built-in formatting. This was previously listed here as "Tekken tokenizer models misbehave".
Same honesty policy as the bug list: here's what's actually queued, and roughly when we expect it to land. No dates promised — this is a list of intent, not a contract.
Between 1.5.8 and 1.6
- PDF handling — attach a PDF and have the AI actually read it, rather than being limited to plain text.
- Mac / Linux / Android port — GobboNet is Windows-only today. This is the work that changes that.
- Model list moves to a GoblinCorps web call — the download menu currently ships with a fixed list baked into the files. Pulling it from our site instead means new and better models can show up in your menu without you reinstalling anything. We’ll be using an ‘add model’ button in the config menu that’ll let you download more models directly from huggingface.
Between 1.6 and v2.0
- Security update — a general pass over how the app protects itself and your data.
- Cloud complement — optional support for cloud models, shipped as a mod rather than baked into the app, because the whole service should stay private unless you explicitly say otherwise.
- Temperature / top-k on the carousel — save response settings alongside each system prompt, so swapping prompts also swaps the feel of the replies. More variance between them without hand-tuning every time.
- Canvas side panel — a proper side panel for long code blocks and documents, with syntax highlighting, instead of everything being crammed into a chat bubble.
- Auto-swipe on a bad entry — if a reply trips a banned word, reroll it automatically instead of making you do it by hand.
- Optional flags on all fields — per-field toggles across the app. Honest note: this one may turn out redundant, and we'll drop it if it does.
- Smoother alternate greetings — make moving between a character's different opening lines feel less clunky.
- Undo button — take back the last thing you did.
- Group chat — more than one character in the same conversation.
- User-controlled information hierarchy — you decide what gets sent to the model first, and what gets pushed down the list when space runs short.
- TTS / STT — speak to it, and have it speak back.
| I want to… | Do this |
|---|---|
| Install it | Download and run GobboNetSetup.exe from Releases |
| Start the chat | Open GobboNet from the Start Menu or desktop |
| Use it on my phone | Run LAN Setup from the Start Menu once, then use the address shown |
| Change my password | Delete .gobbonet-secret and relaunch (or launch.bat reset-password) |
| Add another AI model | Put a .gguf file in the models folder, or use the launcher's menu |
| Turn on web search | Paste a free Ollama key into Settings |
| Shut it down | Close the green launcher window |
Everything GobboNet can do, grouped so it's easy to scan.
Chatting
- Streaming responses — replies appear word-by-word in real time as the AI thinks.
- Stop button — cut off a reply mid-sentence whenever you want.
- Reroll — not happy with an answer? Regenerate it with one click.
- Copy, edit, delete, or reroll any individual reply.
- Copy button on code blocks — grab code in one tap.
- Token counter — see how much of the AI's "memory" you're using.
- Chain-of-thought (reasoning) view — on models that support it, you can watch the AI's step-by-step thinking.
- Output token limitations + smart cutoff cutoff feature
- Auto-stop on stuck reasoning — if the AI's thinking gets caught in an endless loop, it's cut off automatically instead of running forever.
Conversations & organizing
- Landing page — a home dashboard you start from each time.
- Conversation threads — every chat is saved separately, like a history list.
- Rename any thread; copy, edit, or delete threads.
- Search your conversations — find an old chat by what was said in it.
- Folders, tags, and pins — sort chats into folders, label them with tags, and pin your favorites to the top.
- Branching conversations — fork a chat at any point to explore a different direction without losing the original.
- Memory + summarization — older parts of a long chat are automatically summarized so the AI keeps remembering what matters.
- Drag-and-drop file attachments — drag a file straight into the chat to attach it.
Characters & personas
- Default characters — comes with ready-made personalities to chat with right away.
- Character cards — build your own detailed characters (name, description, personality, style, settings).
- Import and export character cards (V2/V3) — bring in the standard
.pngcharacter cards used by other popular AI chat apps, and export your own back out in the same format. Cards are cross-compatible both directions, so you don't have to recreate a library you already own. Honest note: most cards carry over cleanly, but some may need a little tweaking after import. - Lorebook import — embedded lorebooks come across with their underlying information preserved, and card details like creator, tags, version, and notes are kept and handed back when you export. Now fully exposed. Honest note: an imported lorebook is flattened into always-on lore rather than staying keyword-triggered, so all the content is there, it just arrives as one block instead of firing on cue.
- User-controlled RAG lorebook — give a character a lorebook and GobboNet pulls in only the parts that matter to the current moment, using two approaches at once: meaning-based matching (via the small retrieval model) and weighted tag matching for lore you've structured yourself. Honest note: it degrades gracefully — if the retrieval model isn't running, tag matching carries on by itself and chat is never blocked.
- Lorebook + RAG lorebook work together to auto-update new information from extended conversations.
- lorebooks travel with the card — export a character and its lorebook rides along inside the file, so sharing a character shares everything it knows.
- Copy, edit, or delete characters.
- Custom code for individual character cards — attach a small piece of JavaScript to a single character. It runs only while that card is active and is torn down the moment you switch away. Honest note: it is not sandboxed. It runs with the same access as the page does, exactly like the mod controls below — your machine, your card, your code.
- Alternate greetings — give a character several different opening lines.
- User personas — create a profile for yourself so the AI knows who it's talking to.
- Avatar, background, and text-color customization — make each character and the chat look the way you want.
Writing & macros
{{char}}and{{user}}placeholders — automatically fill in the character's and your name as you chat.{{current_DAT}}— lets the AI know the current date and time.- Built-in macros like
{{continue}}and{{fast forward}}to nudge a story or task along. {{auto_continue_x}}— keeps a story or task running on its own so it makes progress while you step away.- Custom macros — create your own shortcuts for text or actions you use a lot.
The AI's behavior (controls)
- System prompt editing — change the underlying instructions that shape how the AI behaves.
- System prompt carousel — save several system prompts and swap between them quickly.
- Response controls — fine-tune things like temperature and top-k to make replies more focused or more creative.
- Banned words — discourage specific words from appearing (best-effort, not guaranteed, appears to be bugged at this time).
- Model selector — choose which AI model you're talking to.
- Model hot-swap — switch to a different downloaded model without restarting anything.
Web search (optional)
- Search the web — let the AI look things up online (needs a free Ollama key).
- Privacy protection — your GobboNet session cookie and login token are dropped before a search leaves your machine, so the search provider never sees them. What does go is the search text as you typed it, along with your Ollama key, because that is what a search is. Chat messages are never included.
Saving & getting things out
- Save AI output as a
.txtor.jsonfile. (Other file types are intentionally left out for now to keep things simple and safe.) - Export, import, or purge all your data — back up everything to a file, restore it later, or wipe it completely.
Customization (advanced)
- Internal "mod" controls — add your own custom JavaScript extensions and override the stylesheet to change how the app looks and works.
Scheduling
- Scheduled tasks — set prompts to send themselves at specific times.
- Copy, edit, or delete scheduled tasks.
Devices
- PC-to-phone connection — use the same chat from your phone or tablet over your home Wi-Fi.
The folder used to open on 57 files with the launcher buried among twenty changelogs. Docs and developer tooling now sit in their own folders. Nothing was deleted, and nothing you run has moved.
launch.bat |
The launcher. Still the first thing you run: it asks the questions, fetches the engine and the model, then hands over to gobbonet.exe in the same window. |
gobbonet.exe |
GobboNet itself — the server, with the chat screen inside it. This is the one file an update replaces. |
setup-lan.bat, stop-gobbonet.bat, teardown-lan.bat |
Phone access on/off, and shutdown. |
gobbonet, linux-amd64/ |
The Linux program and its bundled engine. |
web\ |
The chat interface GobboNet is actually serving. Written by gobbonet.exe on startup, and replaced when you update. Edit it and reload to see changes. |
chat.html, js/, css/ |
The source those files are built from, kept so you can read and diff it. Editing the source changes nothing until it is recompiled — edit web\ instead. |
docs/ |
Reference material, plus docs/changelog/ — one write-up per fix. |
tests/ |
Developer tests and browser preview pages. Not needed to run anything. |
internal/, cmd/ |
The Go server's source. |
installer/, installer-linux/, installer-fedora/ |
How the Windows .exe and the Linux .deb/.rpm are built. |
Want to change how the chat looks? Edit the files in the web\ folder and
reload the page. GobboNet writes that folder itself on startup and serves it, so
what is on disk is exactly what you get — nothing is hidden inside the program.
One catch worth knowing: web\ is replaced when you update, because that is
how a new version reaches the chat screen. The launcher window says so when it
happens. To keep changes across updates, copy them somewhere else and point
GobboNet at that folder instead:
gobbonet config set web_root "C:\Users\you\my-gobbonet-theme"
Only the files you put there override the built-in ones, so a folder holding one CSS file changes that one thing and leaves the rest alone. Remove the setting to go back to normal.
README.md, TROUBLESHOOTING.md and SECURITY.md stay at the root, because
that is where you would look for them and because the launcher points at the
last two by name.
I'm one highly paranoid goblin and I ship a .exe that non-technical people install, so I need to understand every line that goes out. In practice that means I often reimplement a fix in my own code rather than merging the patch especially for security issues. You'll be credited as co-author on the commit and named in the release notes. If I close your PR, it usually means it shipped, not that it was rejected. Bug reports and security findings are hugely welcome even if you don't write a line of code.
GobboNet is brought to you by the GoblinCorps. No corpo money, no venture capital, no masters.