Messaging

Chat

The Chat page is your direct interface to CorvinOS from the browser. Every conversation is a session — past sessions appear in the left panel and can be revisited at any time. New sessions start clean with no shared memory from previous ones.

Session list

The left sidebar shows your past sessions, most recent at the top. Each entry is titled automatically from the first exchange. Examples you might see include sessions like Playwright Audit Graph Test, E2E Full Audit Demo, or various code and engineering tasks. Clicking a session restores the full conversation. Sessions are scoped to the browser tab — opening a new tab gives you a fresh context.

Suggestion cards

When a new session is started, the center area shows a "What's on your mind today?" prompt with four suggestion cards to help you get started quickly:

• What is the probability of rain in Berlin tomorrow?
• Write me a short Python function that filters prime numbers up to n.
• How do I add a new messaging channel?
• Which files were last changed under core/console/?

Click any card to pre-fill the input and send it immediately.

Input bar

The input bar at the bottom of the screen accepts text and voice. Key controls:

Engine selector

Each chat can use a different AI engine. The engine selector (shown as e.g. Claude Code) appears in the input bar area. Click it to switch engines for the current session. Available engines depend on what is installed and detected — see the AI Engine page for details.

New session

Click the New button to start a fresh session. This resets the conversation context, clears any session-scoped skills or tools generated during the previous session, and emits a session.reset event to the audit chain before any cleanup occurs.

Channels

The Channels page shows all messaging bridges that CorvinOS can connect to. Each bridge corresponds to a platform that Corvin listens on and responds through — Discord, Telegram, Slack, WhatsApp, Email, and more.

Configuring a bridge

Each bridge requires a platform token to connect. Tokens are stored in the secret vault (~/.config/corvin-voice/secrets.json) and are injected into the bridge process at start time via environment variables — they are never written to the audit log or exposed to the AI context.

1. Obtain a bot token from the relevant platform (e.g. create a Discord application and generate a bot token in the Developer Portal).
2. Add the token to the secret vault via System → API Keys in the console, or by editing ~/.config/corvin-voice/secrets.json directly (mode 0600).
3. Restart the relevant bridge: bridge.sh restart. The bridge process picks up the token at boot.
4. The channel indicator on the dashboard and on this page turns green once the bridge has successfully connected.

Bridge settings

Each bridge has a settings.json file under operator/bridges/<channel>/settings.json. This file controls the whitelist, rate limits, chat profiles, and voice summary mode for that channel. Settings are read by the bridge on every incoming message — you do not need to restart the bridge to apply most changes.

Per-chat profiles

The chat_profiles key in settings.json lets you assign different AI personas, engine preferences, or permission sets to specific chats or channels within a bridge. Profiles are matched by chat ID. If no profile matches, the bridge uses the default open profile.

Voice Profile

The Voice Profile page controls how CorvinOS speaks. It combines TTS provider selection with an audience configuration that shapes the tone, detail level, and register of spoken responses.

TTS provider

CorvinOS supports multiple text-to-speech providers. The provider is selected here and controls which API is used to synthesize spoken responses. The default provider uses the OpenAI TTS API; the key is read from the secret vault at ~/.config/corvin-voice/.env (mode 0600).

Voice audience profile

Voice responses are synthesized from the same text replies the AI produces, but the audience profile shapes how that text is written before synthesis. Seven voice_audience_* fields in ~/.config/corvin-voice/profile.json let you tune responses for the listener:

• Expertise level — beginner to expert; controls how much the AI explains versus assumes.
• Preferred register — formal, neutral, or casual tone.
• Response length — brief, balanced, or detailed.
• Language — the language the AI responds in for voice output.
• Domain focus — technical, general, creative, etc.
• Pacing hints — pause preferences and sentence length for natural-sounding TTS.
• Accessibility notes — any listener-specific requirements to keep in mind.

Use the console commands /voice-user-show, /voice-user-set, /voice-user-clear, /voice-user-preview, or /voice-user-help to manage the profile from any connected channel.

People

The People page shows every user that has interacted with CorvinOS across all connected messaging bridges. Each entry includes the user's display name, bridge identifier, role, and consent status.

Roles

CorvinOS uses a four-tier role hierarchy. Roles descend in permissions:

• Owner — unlimited quota, full administrative access, can grant and revoke any role. There is one owner per installation.
• Admin — up to 500 messages per day, can manage members and observers. Cannot self-promote to owner (the "kerze rule").
• Member — up to 100 messages per day, standard access to AI capabilities.
• Observer — read-only presence; can receive messages but is not processed by the AI without explicit grant. /join self-registers a user as an observer.

Role assignments are stored per channel and chat (<global>/roles/<chan>__<chat>.json) with a default TTL of 7 days. Commands: /grant, /revoke, /role, /roles, /leave.

Quota

Daily message quotas are enforced per user per bundle role. The check and the record are split operations: if an AI run fails, the quota is not consumed. The owner role has unlimited quota. Quota resets at midnight (server time). Use /quota to check remaining budget, or /audit to see recent activity.

AI disclosure

The first time a new user is seen on any channel or chat, CorvinOS automatically sends a one-time disclosure card explaining that they are interacting with an AI system. This card is structural — it cannot be disabled via configuration. The card is capped at 1 500 characters and is available in German and English. After the card is sent, the user's disclosure state is recorded and the card is not shown again for that user on that channel.