Images and voice

Your agent works mainly with text, but it can look at images you send it, help you make images with the right tools, and reply with voice once you set that up. Here's what works, and how to check it's really working.

Sending your agent images

In the chat, click the paperclip, drag a file onto the message box, or paste a screenshot. Images show a small preview before you send. You can attach any file type; uploads are saved in your agent's uploads folder, so you can refer back to them later in Files.

Your agent can:

• Describe what's in an image: a product photo, a design, a chart.
• Read text from it: screenshots, scanned documents, signs, receipts.
• Answer questions about it: "Does this mockup show the price?"
• Compare images: "What changed between these two versions of the page?"

Whether it can see images depends on the model. Most current models can. If you've chosen a different model in the model picker and image answers seem off, switch back to Default.

When it says it saw the image, but didn't

Sometimes an agent describes an image confidently and gets it wrong. Usually the image never reached it: it was sent in a different chat, through an app that dropped it, or to a model that can't see images. Confident wording isn't proof.

Run a quick check before relying on image work:

1. Open a New chat.
2. Upload one clear image with an obvious detail, like a screenshot with a heading you can read.
3. Ask it to quote that heading exactly and describe one specific detail, such as a colour and where it is.

If the answer is vague or wrong, fix the upload first (try the dashboard chat if you were using a messaging app) before doing real work.

Making images

Your agent doesn't paint pictures itself. It's best used as your creative director: it knows your project, your brand and your audience, so it writes far better image prompts than you'd get starting from scratch.

There are two ways to work:

• You generate, it directs. Ask for three or four prompt options for a tool like Midjourney, run one yourself, then tell your agent what looked right or wrong and ask for a revised prompt. Save the prompts that work, with version numbers.
• It generates through a connected tool. If an image service is available to your agent, it can create images directly and save them to Files. Look for the service on the Integrations tab, or ask your agent: "What tools do you have for making images?"

Test one image before asking for twenty. If your agent says it can publish or generate directly, ask it to show you the first result before running a long batch.

Voice replies

Your agent can reply with voice messages, most often in Telegram, using a text-to-speech service such as ElevenLabs. It usually sends voice as an audio file or voice note, rather than holding a live spoken conversation.

1. Create an ElevenLabs account and pick a voice. Add it to My Voices in ElevenLabs; voices from the public library often fail until you do.
2. Connect ElevenLabs. Check the Integrations tab first. If it isn't listed, ask your agent to walk you through adding your ElevenLabs key safely, so the key isn't left in a chat message.
3. Tell your agent which voice to use, by its voice ID rather than its display name.
4. Test twice. Ask for a one-sentence voice reply, then a second one with different wording. Two in a row proves it's really working.

If it's still sending text: check the key and the voice ID belong to the same ElevenLabs account, check you haven't run out of ElevenLabs credits, and start a New chat before testing again. Connect Telegram first if you haven't: see connect your agent to Telegram.

Voice and image services usually charge separately for what you use, on their own accounts. Your agent's own AI work around them, like writing prompts, comes from your monthly AI allowance.

Frequently asked questions

Can my agent see images?

Yes. Attach an image in the chat and it can describe it, read the text in it and answer questions about it, as long as the model you're using supports images. Default does.

Can my agent create images?

Not by itself. It can write strong prompts for tools like Midjourney, or create images directly if an image service is connected and available to it. Ask it what image tools it can use.

Can my agent talk to me with voice?

It can send voice messages once you connect a text-to-speech service such as ElevenLabs. It's usually an audio message rather than a live call.

Why did my agent describe an image wrongly?

The image probably didn't reach it, or the model can't see images. Start a new chat, send a clear test image, and ask it to quote exact text from it.

Where do images I send end up?

In your agent's uploads folder. Open Files to see, download or delete them.