Ask

Ask is the conversational interface built into every project. It lets you ask natural-language questions about your contracts and get answers grounded in the extracted data already in the project.

Where to Find It

Look for the sixth icon in the project sidebar, labeled Ask. It opens a persistent chat thread scoped to that project.

What It Can Answer

Ask reads the project's Analysis Matrix: the structured fields extracted from every contract in the project. Every answer is grounded in that data.

Answers stream in token-by-token. You see the response begin to write itself within a second or two of asking, and tokens arrive continuously until the answer completes. Citation chips populate once the full answer is ready.

Good questions to start with:

• "Which contracts have auto-renewal clauses?"
• "What is the governing law in each MSA?"
• "Do any contracts require written notice for termination? How many days?"
• "Which agreements have a limitation of liability cap above $500,000?"
• "Are there any contracts where the indemnification clause favors the counterparty?"

If a question touches something not captured by your active template fields, Ask says so rather than guessing. Add the missing field to your template, re-run analysis, and re-ask: the next answer will draw from the newly extracted data.

Citations

Every factual claim in an answer is followed by a small citation chip. There are two kinds:

• Term citation: cites a specific extracted value from the matrix. Hover to see the field label and value, along with the contract name.
• Page citation: cites a contract and a specific page number. Hover to see the contract name and page.

Click any chip to open the source viewer: a centered overlay that loads the cited contract's full extracted text and highlights the exact passage the answer is grounded in (for term citations) or scrolls to the cited page (for page citations). Escape, the backdrop, the X, or Open Contract at the footer closes the overlay and returns you to the conversation.

When Ask Does Not Have an Answer

If a project has no contracts yet, or no contracts have completed analysis, Ask returns a message explaining that there is nothing in the matrix to draw from yet. Upload contracts, run analysis, and then return to Ask.

If a question falls entirely outside the contracts in the project, Ask says so rather than inventing an answer. A response like "I don't have evidence for that in this project's contracts" means the information is not in the matrix for any contract you have access to.

Privacy

Conversation content is private to you. No other project member, including Owners and Editors, can read your messages or Ask's responses. App admins see billing metadata only (when, which project, which model, token counts, cost) and never message content.

Clearing the Conversation

Use Clear conversation at the top of the Ask panel to hide past messages. This clears your view. The billing audit trail is retained on the backend for metering purposes, but your prior messages are no longer visible to you or anyone else in the app.

Limits

To keep cost predictable and prevent abuse, Ask enforces the following caps:

• Message length: up to 8,000 characters per question. Longer pastes are rejected with a "content too long" notice.
• Per-user rate: up to 10 messages per minute and 100 messages per hour.
• Per-project rate: up to 2,000 messages per day across all members.
• Answer length: long answers are fine, but very long ones get cut off. If that happens, Ask says so at the end. Break the question into a few narrower ones.
• Project size: very large projects (hundreds of contracts with lots of extracted fields) can be too big for Ask to consider all at once. If so, Ask tells you and asks you to split the project into smaller ones.

When a rate limit is hit, the chat shows a short note with the wait window (for example "Try again in ~30s") and your conversation history is preserved.

Models

Ask is powered by Anthropic Claude, accessed through AWS Bedrock. Your app admin selects the active model in /admin → Chat. Three options are available:

• Claude Sonnet 4.6 (default): good balance of quality and cost for most queries.
• Claude Opus 4.7: more capable; better for complex multi-contract reasoning; higher cost.
• Claude Haiku 4.5: faster and cheaper; suitable for straightforward lookups.

You do not choose the model yourself per question. The model is set at the app level by the administrator.

---

→ For more detail, see The Analysis Matrix. → For admin controls, see Chat Settings.