Every tool, any model

In most AI apps, what you can do depends on the model you picked. A model without image generation can’t make images; a model without speech can’t talk. In StickyPrompts it works the other way round: every StickyPrompts capability is available to every chat model. Pick the model whose thinking and writing you like best, and ask for what you need.

app.stickyprompts.com Screen recording

How it works

StickyPrompts exposes its features to the model as tools - the same tools listed in the MCP tool reference. When you ask for something a tool can do, the model decides to use it, writes the input (for an image, the image prompt) and StickyPrompts runs it.

• Models that speak MCP call the tools directly.
• Models that don’t go through StickyPrompts’ function-call bridge and other bridging layers, which translate the same tools into a form that model understands. You don’t need to know which is which; it is chosen for you.
• Models with a built-in ability use it. Ask GPT-6 Astra for an image and it uses its own native image generation. Ask Claude Opus 5.5, which has none, and it calls StickyPrompts’ image generation instead - writing the prompt itself.

app.stickyprompts.com/conversations/qLjRNKst

So the model you choose becomes the director: it plans the work, writes the prompts and checks the results, while the specialist models do the rendering.

What you can ask for from any chat

Ask for	What happens
“Three banner images for the autumn campaign”	The model writes three prompts and generates them, several at once if you ask.
“Make her sweater forest green and the light golden hour” (with an image attached)	Image editing on the picture you attached.
“An 8-second vertical video of this scene”	A video is generated in the background; the chat tells you when it’s ready.
“A 30-second upbeat instrumental for the launch video”	Music generation.
“Read this paragraph out in a warm voice”	Text to speech, attached to the reply.
“Transcribe this recording and list the action items” (with audio attached)	Transcription, then the model works with the transcript.
“What does our refund policy say about annual plans?”	Knowledge base search, with citations.
“Top 5 customers by revenue last quarter”	A read-only query against your connected database. See Databases.
“Email the summary to Sarah”	StickyPrompts finds Sarah among your teammates and sends the email. See Using integrations from chat.

Every step is visible: expand the actions line on a reply to see which tools were called, with what input, in what order.

Ask for several at once

Tools can run more than once in a reply, so you can ask for a set rather than one-offs:

Create four variations of a square social ad for the Northwind launch:
yellow background, bold headline "Every store, one screen", a laptop
showing a dashboard. Vary the layout, keep the colours.

The model writes four distinct prompts and returns four images in the same reply.

Keep going in the same chat

Because the results land in the conversation, you can iterate the way you would with a designer: “the second one, but with more space for the headline”, “now a 9:16 version”, “turn it into a five-second video”. The model keeps the context of what it made and why.

Where the same tools work

• Chat, with any model.
• Agents - give an agent a job such as “support triage” and it can use the same tools while doing it.
• Workflows - a scheduled instruction such as “every Friday, turn the pipeline into a PDF and email it to Sarah” runs the same tools unattended.
• Your own code, through the MCP server and the REST API.

Good to know

• Credits. Generation tools use credits like the apps do; video is the heaviest. The model’s own thinking and the tool runs are both counted.
• Guardrails still apply. Your workspace’s redaction rules and model access apply to tool calls too.
• The apps are still there. When you want full control - a specific video model, exact resolution and length, reference images - the dedicated apps under Apps give you every setting. See All apps.