Waterbrother is being built as a local coding runtime that can extend into messaging, model providers,
MCP tools, and remote-control gateways without losing deterministic local control.
The integration model is explicit: terminal first, provider-flexible runtime underneath, and channel
adapters on top. Messaging is DM-first before collaboration.
3Messaging tracks: Telegram live, Discord live, Signal staged
6Provider families: OpenAI, Anthropic, OpenRouter, Ollama, xAI, and compatible APIs
CLIContracts, sessions, approvals, MCP, web, and vision
Messaging Apps
Remote control surfaces
Waterbrother is following the practical order that keeps remote control sane: one user first, then
observation, then approvals, then collaboration. The messaging layer is explicit about what is live,
what is staged, and what still needs a bridge.
Live nowTelegram
Telegram
The first real messaging surface for Waterbrother: paired bot control, live TUI bridging, shared-room
commands, task flow, and group execution gating.
Single-user remote control is live
Shared-room status, invites, tasks, and operator lock
WhatsApp is planned as another messaging surface once the Telegram room model is proven in real-world testing.
No setup yet
Integration Map
TelegramLive now
Single-user remote control, live TUI bridge, shared-room commands, and group execution gating.
DiscordLive now
DMs, guild channels, native slash commands, room control, tasks, role routing, and verifier workflow are live now.
SignalComing soon
Onboarding and config exist now; live bridge follows after lighter bot APIs.
WhatsAppComing soon
Planned on the same messaging-room model once Telegram proves the collaboration surface.
Model ProvidersLive now
Bring-your-own-model runtime across OpenAI, Anthropic, OpenRouter, Ollama, xAI, and compatible APIs.
MCP ToolsLive now
Project-level MCP servers, tool discovery, and structured local workflow extension.
Provider Layer
Bring your own model
Runtime profiles, provider switching, and session persistence are the base layer that make messaging and
remote control viable instead of one-provider hacks.
OpenAI
Anthropic
OpenRouter
Ollama
xAI
OpenAI-compatible APIs
Telegram Flow
Current live path
waterbrother config set-json channels '{"telegram":{"enabled":true,"botToken":"YOUR_BOT_TOKEN","pairingMode":"manual","allowedUserIds":[]}}' --scope user
waterbrother config set-json gateway '{"enabled":true,"controlMode":"single-user","startupChannels":["telegram"]}' --scope user
waterbrother gateway status
waterbrother gateway run telegram
waterbrother gateway pairings
waterbrother gateway pair telegram 123456789
Unknown Telegram DMs create a pending pairing request. Approve locally with
waterbrother gateway pair telegram <user-id>, then the linked remote session is available. Pending requests expire automatically after 12 hours if ignored.
If the TUI is open, Telegram prompts route into that live terminal session and the paired Telegram user operates with the same permissions as the terminal operator. Otherwise Waterbrother falls back to the remote session runner.
Terminal
Primary control surface
Interactive slash commands, durable sessions, exact shell routing, approvals, and checkpointed local work
stay the source of truth.
MCP
External capability without losing structure
MCP is part of the integration story too: structured tool extension instead of pushing every problem
through raw shell commands.
Operator Layer
Supervised surfaces, not sidecars
waterbrother ps, up, down, and restart now manage the live Telegram and Discord surfaces like services, while the TUI stays the source of truth.
Advanced Runtime
Integrations sit on top of a learning local control plane
Regression memory
Build failures, user corrections, and scorecards can feed future work through regressions, correction learning, and /evolve proposals instead of vanishing into terminal history.
Voice in the loop
Streaming dictation, dual-model correction, and xAI speech output let Waterbrother operate as more than a text-only shell assistant.
Resilient long sessions
Provider-aware errors, auto-retry, doom-loop detection, and context overflow recovery are part of the runtime underneath Telegram, Discord, and the terminal.
Next
What follows this page
Signal bridge work, richer remote approvals, and deeper shared-project collaboration once the current multi-surface room model is proven.
Roundtable
The collaboration model
Shared projects now have a dedicated collaboration model with room state, room modes, and a single active
operator lock. Start with the Roundtable page before rolling shared Telegram rooms out to a team.