> ## Documentation Index
> Fetch the complete documentation index at: https://comis-feature-skill-archive-import.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.
# Media & Voice
> How your agent handles images, voice messages, documents, links, and interactive messages
When someone sends a photo, a voice note, a PDF, or a link to your agent,
Comis processes it automatically. Your agent can see images, listen to voice
messages, read documents, and understand web pages -- all without any extra
effort from you. This page gives you the big picture of what Comis does with
different types of media and how to configure each capability.
You don't need to understand the technical details to use this feature. The configuration examples below are copy-paste ready.
## What Comis Handles
Your agent can work with five types of media. Some processing happens
automatically before your agent even starts thinking -- the results are ready
by the time your agent sees the message. Other capabilities are available as
on-demand tools that your agent can call when it decides more analysis is
needed.
| Media Type | What Happens Automatically | On-Demand Tool | Details |
| -------------- | ----------------------------------------------------------- | ------------------------------------ | ----------------------------- |
| Images | Your agent sees the image directly (when vision is enabled) | `image_analyze` | [Vision](/media/vision) |
| Voice Messages | Audio is transcribed to text so your agent can read it | `transcribe_audio`, `tts_synthesize` | [Voice](/media/voice) |
| Documents | Text is extracted from PDFs, CSVs, code files, and more | `extract_document` | [Documents](/media/documents) |
| Links | URLs in messages are fetched and the content is summarized | -- | [Links](/media/links) |
| Videos | Video content is described using vision AI | `describe_video` | [Vision](/media/vision) |
Automatic processing happens before your agent even starts thinking. On-demand
tools let your agent actively request analysis when it decides to.
## How Media Processing Works
Understanding the flow helps you see where configuration matters. Here is
what happens step by step when someone sends a message with media attached:
A user sends a photo, voice note, document, or link in any connected chat
platform.
The media preprocessor identifies what was sent -- whether it is an image,
audio file, video, document, or URL. Each type has its own processing
pipeline optimized for that kind of content.
Based on your configuration, Comis processes the media automatically. This
might mean transcribing a voice message to text, sending an image to
vision AI for analysis, extracting text from a PDF, or fetching the
content of a linked web page.
The agent sees the original message text plus all of the processed media
results. A voice message becomes readable text. An image comes with a
description. A document's content is available inline. The agent has
everything it needs to respond intelligently.
If the automatic processing was not enough, your agent can use on-demand
tools to analyze further. For example, it can ask specific questions about
an image, re-analyze a document with different settings, or transcribe
audio that was not automatically processed.
## What Happens Without Configuration
Even if you have not configured any media providers, Comis still handles
media gracefully. When media arrives without a configured provider, your agent
does not crash or ignore the message. Instead, it receives a helpful hint
about what was sent so it can still respond appropriately.
Here is what your agent sees for each media type when no provider is
configured:
**Voice messages:** Your agent receives a hint like "Voice message attached --
use `transcribe_audio` tool to listen" so it knows a voice message was sent.
If the on-demand transcription tool is available, the agent can still process
the audio by calling it explicitly.
**Images:** Your agent sees a hint that includes the image URL. If a vision
provider is available for on-demand analysis, the agent can use the
`image_analyze` tool to understand the image contents. Without any vision
provider at all, the agent still knows an image was shared and can
acknowledge it.
**Documents:** Your agent sees a hint about the attached file -- including the
filename, MIME type, and file size. The agent can use `extract_document` to
read the content on demand if the extraction tool is available.
**Videos:** Similar to images, your agent sees a hint about the video
attachment. If Google Gemini is configured, the agent can use the
`describe_video` tool to get a text description of the video content.
**Links:** When link understanding is disabled (the default), URLs in messages
are passed through as plain text. Your agent can still see and reference the
URLs, but Comis does not automatically fetch and summarize the linked content.
This means you can start using Comis without configuring every media provider
up front. Add providers later as you need more automatic processing. Your
agent always knows when media was sent, even without full media configuration.
## Channel × Media Capability Matrix
Not every channel can deliver every media type — that is a platform
limitation, not a Comis decision. The table below shows what each connected
channel can actually receive and send.
| Channel | Images in | Voice in (audio attachment) | Voice out (TTS reply) | Video in | Documents in | Link previews |
| --------------- | ----------------------- | --------------------------- | ---------------------- | ---------------- | --------------- | ------------- |
| Telegram | Yes | Yes | Yes (opus) | Yes | Yes | Yes |
| Discord | Yes | Yes | Yes (mp3) | Yes | Yes | Yes |
| Slack | Yes | Yes | Yes (mp3) | Yes | Yes | Yes |
| WhatsApp | Yes | Yes | Yes (mp3) | Yes | Yes | No |
| Signal | Yes | Yes | Yes | Yes | Yes | No |
| LINE | Yes | Yes | Yes | Yes | No | No |
| iMessage | Yes | Yes | Yes | Yes | Yes | No |
| IRC | No | No | No | No | No | No |
| Email | Yes (inline/attachment) | Yes (audio attachment) | Yes (audio attachment) | Yes (attachment) | Yes (full MIME) | No |
| Microsoft Teams | Yes | Yes | No | Yes | Yes | No |
The "voice out" column reflects whether the channel adapter can deliver an
audio reply when [auto-TTS](/media/voice#auto-reply-modes) is enabled. The
default audio format per channel is documented in
[Voice → Per-Channel Audio Formats](/media/voice#per-channel-audio-formats).
## Cross-Platform Support
Media processing works across all connected chat platforms. The same
configuration applies everywhere -- you set up vision once, and it works
whether someone sends a photo in Telegram, Discord, Slack, or any other
connected channel.
However, each platform handles media attachments differently behind the
scenes. Comis abstracts these differences so you do not need to worry about
them:
* **Telegram** sends voice messages as `.ogg` audio files
* **Discord** uses various audio formats depending on the client
* **WhatsApp** sends voice notes in a specific format that Comis converts
automatically
* **Slack** hosts files on its own CDN with authentication tokens
Your configuration stays the same regardless of which platform the media
comes from. Comis handles the platform-specific details internally so your
agent gets consistent results no matter where the message originated.
Some media features are only available on certain platforms. For example,
rich messages with buttons currently render on Discord, Telegram, Slack, and
Microsoft Teams (as Adaptive Cards); LINE and WhatsApp button rendering is not
yet wired through `sendMessage`, and Signal, iMessage, IRC, and Email do not
support buttons. See each
capability page for platform-specific details.
## Configuration Overview
All media settings live under `integrations.media` in your `config.yaml`.
Each media capability has its own section with sensible defaults, so you only
need to configure the features you actually want to use. Here is a minimal
example showing the main toggles:
```yaml theme={null}
# config.yaml
integrations:
media:
vision:
enabled: true
transcription:
provider: auto # Keyless-first default (no API key needed)
tts:
provider: edge # Keyless default (Microsoft Edge, no API key)
autoMode: off
linkUnderstanding:
enabled: false
documentExtraction:
enabled: true
```
Each section in the configuration corresponds to one of the five media
capabilities:
* **`vision`** -- Image and video analysis (see [Vision](/media/vision))
* **`transcription`** -- Speech-to-text for voice messages (see
[Voice](/media/voice))
* **`tts`** -- Text-to-speech for voice replies (see [Voice](/media/voice))
* **`linkUnderstanding`** -- Automatic URL content fetching (see
[Links](/media/links))
* **`documentExtraction`** -- File text extraction (see
[Documents](/media/documents))
Each capability page below shows the full configuration options. You only need
to configure the features you want to use -- everything has sensible defaults.
## Explore Capabilities
How your agent sees and understands images and videos
Speech-to-text transcription and text-to-speech auto-reply
PDF, CSV, code files, and more
Automatic URL content understanding
Buttons, cards, embeds, and polls across platforms
## Related
See all media tool parameters and usage
Full config.yaml reference