Two personalities. One protocol. On Discord and on your own devices.
One AI personality suite in two places. On Discord, a dual-personality bot system that is live today: AI chat, economy, XP, moderation, 37 toggleable modules, and a full web dashboard. On your desktop and your phone, personal companion apps that carry the same two minds, still in development. One calm, one chaotic, the same wherever you talk to them.
Order and Chaos, Working in Harmony
The Collective
Meet the models
Placeholder character models. Final in-app renders are on the way.
More Than a Command Parser
95 text commands across the two bots. Under them sits a small pipeline that decides what each message means before a model ever sees it.
A message is sorted first: ordinary chat, a search, an image, a question about something it was told, a summary. Each one goes to its own handler, with the bot's current mood and what it remembers about your server folded in. That is why the reply sounds like the character and not a template.
Two checks read every incoming message, and both have to agree before anything is blocked. A blocked message is dropped and never reaches the model. If the classifier goes down, the pattern check still blocks on its own.
One request answered on the hosted hardware costs one Cinder. Your server plan grants a pool of them each month. The balance is a running total of an append-only history, so it is never a number the bot can quietly edit, and anything arriving from this site is signature-checked before it counts.
Both bots register three slash commands. /privacy lists every category of data held about you and how long it is kept. /data-optout stops new collection. /data-delete erases what is already there, behind a confirmation step. XP, economy balances, profiles and warnings are scoped per server.
The Same Two Minds, On Every Screen
Halcyon listens, drafts, and explains the why. Rascal is decisive, curt, and gets out of the way. Preview them below and flip between the two brand themes. These are design previews of apps that are still in development, not finished software.
Pocket Oracle, a phone companion in the making.
One wake phrase, speech transcribed on the phone rather than sent to Google, and intent routed to what is wired today: Home Assistant and web search. The model answers, the integration fires, the response lands in the same channel. More bindings to follow.
- Two personalities: pick Halcyon for diplomacy, Rascal for speed.
- Server Manager mode for Discord admins: triage escalations from your phone.
- Phrase router: exact matches fire instantly on the cheapest tier.
- GDPR-grade controls. Wipe everything from one screen.
No App Store or Play Store listing yet. There is nothing to install today.
A desktop where the AI is a character, not a sidebar.
The design puts the companion center stage, with modular capability panels orbiting it. Equip what you need: chat, compute contribution, providers, marketplace, notes, Home Assistant. Halcyon and Rascal each carry their own tone, tier, and routing budget.
- Local-first inference. Optional pool contribution from idle GPU.
- Tweaks drawer, no nested settings pages.
- Plugin loader for capabilities you write yourself.
- One brand, both themes. Toggle Halcyon and Rascal at any time.
No build to download yet for Windows, macOS, or Linux. This is a preview of the target design.
What the Wider Protocol Adds
Beyond the Discord modules, the desktop and mobile apps (both still in development) are designed around a shared set of capabilities.
In Discord today the bot can speak its replies in a voice channel, with mood-reactive prosody and a choice of an online or a fully offline voice. On the apps, transcription is built to run on your own machine. The audio itself is never uploaded. That side is in development.
Discord live, apps in developmentNotes are encrypted on your device with AES-256-GCM. The key is derived from your passphrase, and the passphrase itself is never stored. Export is plain Markdown, so your notes are not locked in. Part of the desktop client, so it arrives when that does.
In developmentConversations are not written to disk unless you switch that on. Local Mode blocks every outbound call, for when you want the machine sealed. The client needs a Doombringerz account to get past its sign-in screen, and if your local engine is not running, chat falls back to a hosted endpoint on the Doombringerz servers. Local-first is the default, not a promise you cannot turn off.
In developmentPoint the companion at a folder of local model files, or at a model server on your own machine, and pick which one answers. While a local model is the one answering, the prompt does not leave your machine.
In developmentCapabilities load as plugins. Install one, switch it off, remove it, or write your own. Each gets its own window. They run with the same access as the app itself, so only install ones you trust: a permission model is planned, not built.
In developmentPut idle GPU or CPU to work answering your own Discord server's requests instead of the hosted ones. It runs your server's jobs and no stranger's. Built for a machine you own and trust: inference runs in the node's own process, not a sandbox. Earning Cinders back for jobs served is built and switched off. It ships inside the desktop client, so there is nothing to install on its own yet.
In developmentAll 37 Modules, Toggleable Per Server
Enable only what you need. Each module can be turned on or off per guild. AI Chat and Image Generation are core, always on, and not module toggles.
Designed for Independence
Free Bot. Server Plans Unlock The Rest.
Add the bots to any server for free. Paid tiers below are server plans, one plan per Discord server, applied to every member of that server.
- + AI Chat and Image Generation (core, always on)
- + AFK Status
- + Auto-Role
- + Birthdays
- + Color Roles
- + Confessions
- + Economy / Wallet
- + Event Scheduling
- + Free Games Notifier
- + Giveaways
- + Highlights
- + Polls
- + Quotes
- + Reddit Notify
- + Reminders
- + Reputation / Thanks
- + Role Menus
- + Scheduled Messages
- + Social Notify
- + Starboard
- + Suggestion Box
- + Ticket System
- + Timezone Utility
- + User Profiles
- + Welcome / Leave / Ban
- + XP & Leveling
- + 6 modules: Auto-Mod, Audit Log, Moderation (Warns), Alt Account Detection, Verification Gate, Thread Watcher
- Everything in Spark
- + 3 more modules: Server Analytics, Voice TTS, Voice STT
- Everything in Ember
- + 3 more modules: Custom Commands, Media Search, SIMKL Watchlist
- + This unlocks every plan-gated module
- Everything in Blaze
- + The largest monthly allowance on the ladder
- + Priority when the hosted servers are busy: served ahead of lower tiers, and held on that hardware instead of being pushed to an outside provider
- + Nothing is ever refused, only made to wait. Idle servers mean priority changes nothing
Bot service tiers are separate from creator support on Patreon, they're independent subscriptions for different things.
One request answered on the hosted servers costs one Cinder. A plan grants a pool of them each month, scoped to the one server it is attached to. Someone else's busy community can never spend yours.
Allowance is for the month it was granted in. Whatever a server has not spent is retired when the next month's grant lands, so it does not stack up. Cinders you earned by serving your own jobs, and any you bought outright, are treated differently: those are yours and do not expire.
The tier decides which modules that server can switch on. The Free column above lists what needs no plan at all. Adding the bots to a server costs nothing either way.
When the hosted servers are quiet, priority changes nothing and costs nothing. When they are busy, Inferno requests are handed the next free slot ahead of lower tiers, and are not pushed off that hardware onto an external provider. Nothing is ever refused, only made to wait.
Discord usage and personal usage are kept in two separate pools. A busy Discord server's load is driven by its member count and should not eat a personal allowance. The separation is structural: the two balances sit under different keys, so a query against one cannot reach the other. Today only the Discord side is wired up.
Metering is built but dormant, so nothing is being deducted from any pool today. Earning Cinders back by serving your own server's jobs is built and switched off. The personal pool for the desktop and phone apps is built but not yet connected to those apps, and carries no allowance, so personal use is not metered at all.
Where Your Requests Are Answered
Your own compute is tried first, then the hosted pool. The outside provider is built but switched off, so nothing is being sent there at all. That is what the bots run today, not a roadmap. The numbers are the router's own tier index, so the gaps are deliberate: the slots in between have no shipping code behind them.
| Tier | Name | Description | Cost | Consent |
|---|---|---|---|---|
| 1 | Your Compute Module | Tried first when your server runs a licensed Compute Module node that is online. The node pulls your server's own jobs and answers them on your hardware. Skipped automatically when no node is online. The module is a one-off licence rather than part of a server plan, and it is not on sale yet: you can see what it covers in the meantime. | Your hardware | You run the node |
| 2 | Hosted Server (Ollama) | The default. Self-hosted Ollama on Doombringerz machines, no third-party API in the path. This is the tier a plan's monthly Cinders pool is for. At Inferno it is also the tier that is held for you when the servers are busy. | What the plan pool covers Nothing deducted today | None |
| 3 | External API Offload Off | Built, and switched off. It exists for the case where the hosted servers cannot serve, and it is the one path where prompts would leave the Doombringerz network. Because metering is off, this tier is dropped from the list for everyone, so no request reaches it today. Free servers and the owner's own traffic are excluded from it separately, and stay excluded whatever metering does. | Not in use | Would need a paid plan, consent switched on, and a balance |
The outside provider is configurable and defaults to Gemini. Nothing advances there silently. Today a request that cannot be served on Doombringerz hardware waits or says the servers are busy; it is not sent outside.
Privacy First
Your server data stays yours. Here's exactly what is stored.
The HR Protocol Ecosystem
One AI personality suite. Every platform you use.
Downloads
Nothing to download yet. The desktop client and the mobile app are both still in development, so there is no build for any platform. This list is the build status, not a download list. Real links are published here on release.
Ready to meet Rascal and Halcyon?
Check out the documentation or explore each bot's personality.