Documentation

Prompts

Build reusable, multi-turn prompt templates with typed variables — each becomes a slash command in your linked chat channels.

Prompts are reusable message templates your project exposes through its MCP server. In any linked chat channel, each prompt becomes a slash command people can run by name — so a common request turns into a one-word shortcut. A prompt can be a single message or a full multi-turn exchange, and any {{variable}} you use is filled in when the prompt runs.

Why use prompts

Turn repeat work into one command. Capture a request you send again and again once, then run it as /command — no re-typing, no forgetting the details.
Consistent results every time. Everyone triggers the exact same wording and structure, so your AI answers the same way no matter who asks.
Share expertise with your whole team. A well-crafted prompt written once is instantly available to every teammate in the channel — no prompt-engineering skill required to use it.
Guide the model with typed inputs. Variables are validated by type and marked required, so callers can’t run a prompt with a missing or malformed value.
Steer the conversation. Multi-turn User/Assistant templates set the tone and give the model examples to follow, producing sharper, more predictable replies.
Edit without redeploying. Update a prompt in the dashboard and the new version is live in every linked channel immediately — nothing to ship.

Create a prompt

Open the Prompts page and select New prompt. A panel opens on the right with empty Title, Description, and Messages fields.

The Prompts page showing the empty state, with the New Prompt panel open and blank Title, Description, and Messages fields

Name it and write a message

Give the prompt a title — Ganju turns it into the slash command shown just below (here, test becomes /test). Add a short description so teammates know what it does, then write your first message.

Drop {{variables}} anywhere you want a value supplied at run time. In this example the user message is hi user {{name}} your age is {{age}}, so name and age become inputs the caller provides.

The New Prompt panel with the title "test", slash command /test, a description, and a user message containing {{name}} and {{age}}

Configure the variables

Ganju auto-detects every {{variable}} in your messages and lists them under Variables. For each one you can:

set a type — String, Number, and Boolean;
add a short description to guide whoever runs the prompt;
mark it required or optional.

Here {{name}} is a required String (“user’s name”) and {{age}} a required Number (“user’s age”).

The Variables section listing {{name}} as a required String and {{age}} as a required Number, each with a type and description

Add more messages

Real prompts are often a conversation, not a single line. Select Add message to append another turn, and switch each message between User and Assistant to shape the exchange. This second message is an Assistant turn, this is a new {{message}}.

The Messages section with two messages: a User message and an Assistant message containing {{message}}

New variables you introduce are picked up automatically — {{message}} now appears alongside {{name}} and {{age}} in the Variables list, ready to type and describe.

The Variables section now showing {{name}}, {{age}}, and {{message}}, each configurable

Edit as JSON

Prefer to work with raw structure? Toggle Visual / JSON at any time to edit the messages directly as an array of { role, content } objects. It’s the fastest way to paste in a template you already have, and it stays in sync with the visual editor.

The Messages editor in JSON mode showing an array of role/content objects

Save and run it

Select Create and the prompt is saved as a card showing its slash command, description, and messages. From any linked channel, type the command (like /test), fill in the required variables, and Ganju runs the template — variables substituted — as if you’d written the whole thing by hand.

Next: give your prompts something to work with — add resources.

Last updated July 6, 2026