Creating a template turns a good prompt into a shared, reusable assistant. The template editor assembles the ingredients a chat needs — instructions, model, knowledge, capabilities, tools and parameters — and saves them as one card.
Open the editor
- Open Templates in the module rail.
- Select the add (new) button on the Templates page.
- The template editor opens.
Note:
Who can create templates depends on your group permissions. If you cannot see the add button, ask your administrator. See Permissions.
Editor sections
Basics
- Name — the title users see on the card.
- Description — one or two sentences explaining when to use this assistant; this is what people search on.
- Icon — upload a custom image so the template is recognisable in the library.
System Prompt
The core instructions: the assistant's role, tone, output format, and hard rules. Good system prompts are specific and short.
You are a legal-operations assistant. Summarise contracts for a
non-lawyer audience. Always end with a "Key risks" section.
If a clause is ambiguous, say so rather than guessing.
Model
Choose the default model for this assistant. Only models your administrator has enabled are listed. Leave the default unless you have tested a better fit.
Capabilities
Toggle what the assistant is allowed to do. Each capability appears in conversations started from this template:
| Capability | Allows the assistant to… |
|---|---|
| Vision | Understand attached images. |
| Web Search | Look up current information online (when your organisation permits internet use). |
| Code Interpreter | Run code to analyse data or generate outputs. |
| Image Generation | Create images from text. |
| Video Generation | Create videos from text/images. |
| File Upload | Accept file attachments. |
| Citations | Show numbered source references from knowledge retrieval. |
| Form | Render structured input forms for the user. |
Note:
Administrators can lock capabilities platform-wide; options your admin disabled will not be selectable here. See Admin Panel → Capabilities.
Knowledge
Attach one or more knowledge bases (collections of processed documents) so answers are grounded in your material. Knowledge bases are managed in the Work Station's Vector Database section — see Knowledge Bases & RAG. You can create a new knowledge base directly from the editor.
Tools, Actions, Filters and Guardrails
- Tools — external actions available to the assistant via connected tool servers (email, ticketing, lookups…). Your admin registers tool servers; you select which this template may use. See Tools.
- Actions / Filters — pre/post-processing steps applied around the model call (for example formatting or routing behaviour), where enabled.
- Guardrails — policy checks applied to inputs/outputs for safety and compliance, defined by your administrator. See Guardrails.
Advanced Parameters
Fine-tune generation for this assistant:
| Parameter | Effect |
|---|---|
| Temperature | Randomness: low = deterministic/factual, high = creative. |
| Top-P | Nucleus sampling cutoff; an alternative to temperature. |
| Max tokens | Maximum length of the answer. |
| Frequency / Presence penalty | Discourage repetition of words/phrases. |
| Stop | Sequences that end generation immediately (comma-separated). |
The full explanation of each is in Model Parameters.
Test before publishing
After saving, open your template and select Start Chat. Run the two or three requests you had in mind and refine the system prompt, knowledge or capabilities until the answers match expectations. Iterate freely — changes apply to new chats.
Share with your team
Once published (subject to your permissions), the template appears in the Templates library for your group. Point colleagues at it by name; see Using Templates.
Related
- Template Library / Work Station
- Prompt Library — lighter-weight saved prompts
- Template Editor Options — full option reference