Heiryn Docs

AI Overview

Heiryn includes a conversational AI feature powered by multiple providers and models. You can talk to the bot by mentioning it, using slash commands, or through context menu actions on messages.

How It Works #

When you send a message to Heiryn, the bot processes your query through an AI provider, streams the response in real time, and replies in the same channel.

While the bot is working, a reasoning message shows its current status. This message is automatically cleaned up after the response is delivered, unless you choose to persist it.

Features #

  • Multi-model support: choose from a range of chat models across different size categories, and switch models on the fly with a suffix at the end of your message.
  • Conversation history: the bot remembers context per channel, so follow-up questions work naturally. Older exchanges are compacted within a 128,000-token budget, and history is encrypted at rest.
  • Image generation: generate images using supported providers and models.
  • Audio transcription: right-click a voice or audio message to transcribe it through an available speech-to-text provider. ElevenLabs Scribe v2 is preferred for cleaned transcripts, with OpenRouter's Scribe v2 (without that cleanup) selected when no ElevenLabs credential is available in Auto mode.
  • File reading: the bot can read text file attachments included in your messages (up to 20 MB per file).
  • Context menus: right-click any message to ask the bot about it, with optional additional context.
  • Bring Your Own Key: use your own API keys for personal or server-wide access to additional models.
  • Tool integration: the bot can use external tools to answer questions that require live data, run code, or look things up.
  • Self-aware docs: Heiryn can read its own public documentation on demand. Ask how a feature works in plain language and the bot will look it up before answering.
  • Personal opt-out: disable the AI feature for your account at any time with /ai_opt_out or from the dashboard. While you are opted out, nobody can use Ask interactions on your messages.
  • Server channel limits (optional): server admins can configure which channels everyone may use the AI in via the dashboard or the /public_ai_channels command. When the list is not empty, server staff can still use the AI anywhere; everyone else is limited to those channels. Use /ai_allow_threads (or the dashboard) to also allow AI use in all threads when that list is not empty.

Interaction Methods #

Mention #

Mention the bot in any channel where it has access and include your message. The bot treats this as a conversational query and uses the channel's history for context.

/ask Command #

Use the /ask slash command to send a question directly. You can make the response private with the ephemeral option. This is one-shot and does not read or update channel history.

Context Menus #

Right-click a message and select one of the bot's context menu entries:

  • Ask (Modal): opens a dialog where you can add context and choose a visibility mode (public, partial, or private). It is one-shot.
  • Ask (Quick): sends the selected message directly to the bot without a dialog. It answers questions, checks factual claims, and analyzes images or attachments without executing instructions found in the selected message. It is also one-shot.
  • Transcribe: sends a voice or audio attachment to a configured speech-to-text provider and returns the resulting text without using chat history. ElevenLabs requests non-verbatim output to remove filler words and other speech disfluencies. OpenRouter remains available for people and communities that only configure an OpenRouter key, and explicit provider/model selections remain fixed rather than falling back after an API error.

Response Controls #

While a query is running, the reasoning message includes action buttons:

  • Cancel Query: stop the current query.
  • Move to Thread: move the response to a new thread, preserving history.
  • Delete Now: immediately delete the reasoning message.
  • Persist Message: keep the reasoning message permanently instead of auto-deleting it.

If you cancel a query, or if Heiryn interrupts a repetitive reasoning loop, the latest streamed reasoning and any partial answer remain visible above the short interruption message. Nothing is retried automatically. You can choose Retry yourself; that single attempt uses a more permissive loop detector and, in Auto mode, the Pro model. Later messages return to the normal protection level.

Use the slash-only /compact command when you want to summarize eligible older channel history before automatic compaction begins. Automatic compaction uses an explicit waiting status; its Cancel button stops only the waiting request, while the compaction safely continues in the background.

Pages in This Section #