The Prompts area is a workspace library of reusable prompt templates — versioned assets you can build once and reuse across agents and automations.

Overview

A prompt is a message template: an optional system message plus an ordered set of user and assistant turns, with {{variables}} you fill in at run time. This is different from the Prompts section inside an agent, which holds prompts scoped to that agent — the library here is workspace-wide. Use it to standardize prompts your team relies on and keep them version-controlled.

Creating a Prompt

On the Prompts list, use New Prompt and fill in the create dialog:

That creates the prompt; open it to author its messages. The list shows each prompt's Prompt Name, Description, Workspace, Tags, Owner, and Last Updated, with Search prompts to find one; the columns menu also offers hidden Created On / Last Updated By columns you can switch on. If you have none, the empty state reads Add your first Prompt.

Writing a Prompt's Messages

Open a prompt to edit its messages. There's a single System message slot at the top for the system prompt, and you use Add Message to append conversation turns. Each message has a role you can switch between User and Assistant, and its content can include {{variables}} — the editor automatically collects those variables as the prompt's inputs. Add and reorder turns to build the exact conversation shape you want the model to follow.

You don't have to write every message by hand: use Generate to open the Generate Messages modal, choose a Model and write Instructions describing the conversation you want, and the system drafts the system message and turns for you. You can show or hide your existing messages while generating, then Confirm to bring the generated messages into the prompt — a fast way to start a prompt you then refine by hand.

Versioning and Publishing

Prompts version and publish like other assets. As you edit you're working on a draft (the status shows Draft until deployed, then V{n}). Save/Update stores your work; Publish creates a tagged version with optional notes. Open the Version History to view a past version read-only and Restore it if needed. Publish a stable version once the prompt reads the way you want, so anything using it gets a consistent result.

Variable Substitution at Run Time

A prompt is stored as its system message plus the ordered user/assistant turns, together with an input schema collected from the {{variables}} in your text. When the prompt runs, the whole template — system message and every turn — is compiled with the input values you supply through the platform's templating engine, so a variable works anywhere in any message, and the same variable can appear in several messages. The filled-in messages are then sent to the model you selected, and you get back the response plus the tokens used.

Where a Saved Prompt Can Be Used

A saved library prompt is consumable wherever work runs (the prompt editor itself has no test panel — to try a prompt, run it with the Execute Prompt action below):

• Automations — the Execute Prompt action (under Conversational AI by UnifyApps) lets any workflow pick a saved Prompt, a Model from the Model Library, and the variable inputs, and returns the model's response and tokens used. This is the main way to reuse a curated prompt inside business logic.

• MCP — prompts are exposed through the platform's MCP server (list and get), so MCP clients can discover and pull your prompt templates.

• Agents — an agent's own Prompts section holds prompts scoped to that agent; the workspace library here is the shared, reusable pool.