Veriprompt Docs

Appearance

Chat Terminal ​

The Chat Terminal is where you try prompts interactively. It helps you pick the right model and provider for each type of content, without writing code.

What you can do ​

  • Chat with different models and providers
  • See which provider is used for each message
  • Switch categories to match the content type
  • Run stored prompt chains for multi-step workflows
  • Manually enable PII sanitization before a message is sent

Choose a category (content type) ​

Each chat session has a category (for example: General, Internal, or Compliance). Categories help the system choose the right routing rules.

  1. Open the category dropdown at the top of the chat.
  2. Pick the category that matches your content.
  3. The chat will follow the routing and safety rules for that category.

PII sanitization toggle ​

The chat terminal includes a manual PII sanitization toggle above the message input.

  • When enabled, VeriPrompt sanitizes detected PII before the message is sent to the provider.
  • The response is re-mapped before you see it.
  • Whether the toggle is available depends on your subscription package and the active routing policy.
  • If the routing policy disables manual sanitization, enabling the toggle has no effect and the request stays unsanitized.

Keep an eye on usage ​

You will see a simple usage bar in the header:

  • Session usage: how much you have used in this chat
  • Context usage: how much fits in the model window
  • Budget usage: how close you are to your team limits

If you approach a limit, the chat suggests what to do next.

Conversation Summaries ​

When your conversation approaches the context limit, VeriPrompt offers to create a summary. Summaries compress your chat history while preserving:

  • The main goal you're working on
  • Key decisions and facts
  • Code snippets, URLs, and technical details (verbatim)
  • Open questions and next steps

How it works ​

  1. Warning appears when you reach 80% of the context limit
  2. Click Save Summary to compress the conversation
  3. The summary is stored and can be continued later
  4. Maximum 3 summary chains - after that, start a fresh chat

Continuing from a summary ​

  1. Open a conversation that has a summary
  2. Click Continue from Summary
  3. A new chat opens with the summary context injected
  4. Your classification and routing rules carry over

This feature is useful for long technical discussions that span multiple sessions.

Blacklist enforcement and reports ​

Chat categories can include blacklist rules that block messages containing restricted terms. The scan covers the prompt text and common document attachments (PDF, Word, CSV, TXT, and other text-based files).

When a blacklist match occurs, the event is logged with the time, user, prompt text, and matched terms. Super Admins and Security Custodians can download a report for a date range and optionally archive or delete older records in:

  • Admin → Chat Blacklists → Blacklist Violation Reports

Run a stored prompt chain ​

If your team has saved a prompt chain (multi-step workflow), you can run it directly from the chat. This is useful for tasks like intake -> review -> summary.

For setup details, see:

Optimize your prompt with Guru ​

If your draft is more than a few words, a wand icon appears in the composer toolbar next to the prompt picker. Click it (or press Cmd/Ctrl+Shift+O) to send your draft to Guru's prompt optimizer.

  1. Type a draft message of ten words or more.
  2. Click the wand icon, or press Cmd/Ctrl+Shift+O.
  3. A review window shows your original text and Guru's rewritten version side by side. Edit the rewrite if you want.
  4. Click Use this to replace your draft with the optimized version, or Keep mine to close the window and keep typing.

Nothing is sent to a provider during this step — optimizing only rewrites the text sitting in your composer. Your message still goes through the normal send flow afterward, including the Shield pre-send review if it's active for this conversation.

The wand always uses Guru's own optimizer, independent of whichever model/provider quality tier you have selected for this chat, so optimization quality doesn't change if you're chatting on a cheaper tier.

The same one-click optimizer is also available on saved prompts in the Studio prompt editor's toolbar ("Optimize with Guru"), where it shows a full diff instead of a side-by-side preview — useful when you want to see exactly what changed before saving a new prompt version.

Best practices ​

  • Use the right category for sensitive data.
  • Turn on PII sanitization when you are discussing personal, contractual, or customer-specific data.
  • Keep reusable prompts in the repository and call them from chat.
  • Use synthetic tests for prompts that will go to production.