COBBLESTONE SOFTWARE

Contract Insight — User Guide

Natural Language Processing (NLP)

03. Admin User Manual › VISDOM & Risk Management › Natural Language Processing (NLP)

Note: Each procedure in this guide begins at the Contract Insight homepage, so any section can be followed on its own. Custom fields are configurable, so the exact field list on a contract category record may vary between environments.

1. Overview

VISDOM's natural-language processing (NLP) is not exposed as a separate "NLP" screen; it is the language capability behind VISDOM's extraction, chat, and analysis features. In Administration > VISDOM Configuration > Models & providers, three services make up the practical face of NLP: LLM Agent (Chat) ("The conversational VISDOM chat assistant"), Language Risk ("Flags risky / non-standard contract language"), and Sentiment ("Scores clause sentiment (favorable/unfavorable)"). Together these services read and understand contract text well enough to extract fields, flag unusual wording, score clause sentiment, and answer questions in plain English through the VISDOM Assistant chat.

2. Key Elements

• No dedicated "NLP" settings screen exists; NLP capability is delivered through the LLM Agent (Chat), Language Risk, and Sentiment services.
• All three are listed under Per-Service Routing in VISDOM Configuration > Models & providers. LLM Agent (Chat) can be routed to its own provider, model, and fallback; Language Risk and Sentiment are marked "Model chosen by VISDOM Services" and are managed by CobbleStone.
• Model choices are bounded by the license tier (for example, VISDOM+) and the server-enforced routing policy. A service left at (inherit) uses the Tier Defaults provider and model.
• VISDOM Status shows whether VISDOM Services are connected and whether Assistant chat is on, so it is the first place to check if NLP features are not responding.

3. Viewing the NLP Services in Models & Providers

The Models & providers tab shows which AI provider and model each VISDOM service uses. The license tier is set by CobbleStone, and the choices available here are bounded by it.

1. From the homepage, open Administration (the settings icon at the bottom of the left navigation bar).
2. Select VISDOM Configuration.
3. Click the Models & providers tab. The AI Profile Manager shows the Effective Tier, Default Tier, Routing Policy, and the Tier Defaults (default provider and model) used by any service that does not define its own.

4. Scroll down to Per-Service Routing to find the LLM Agent (Chat), Language Risk, and Sentiment rows.
5. To route the chat assistant differently, set the Provider, Model, and (optionally) Fallback Provider and Fallback Model on the LLM Agent (Chat) row, then click Save Routing Profile. The Language Risk and Sentiment rows are read-only ("Model chosen by VISDOM Services").

4. Asking Questions with the VISDOM Assistant

The VISDOM Assistant is the conversational front end of the LLM Agent (Chat) service. It answers plain-English questions, including questions about your own records.

1. From any page, click the VISDOM Assistant button (the purple chat icon in the bottom-right corner).
2. Type a question in the Message VISDOM... box and send it, or click one of the suggested prompts (for example, "Which contracts expire in the next 90 days?").
3. Review the answer. Matching records appear as cards showing key fields, each with a Go to record link. Expand Show how I looked that up to see how VISDOM found the answer.

5. Checking VISDOM Status

VISDOM Status is a live snapshot of the VISDOM Services connection and the AI capabilities enabled for the environment. The page reads configuration only and does not call VISDOM, so no quota is used.

1. From the homepage, open Administration.
2. Select VISDOM Status (or click Connection & health at the bottom of the VISDOM Configuration page).
3. Confirm the banner reads Connected & Live and that, under Capabilities, VISDOM enabled, Assistant chat, and LLM field extraction are on.
4. If the access token has lapsed, click Re-authorize VISDOM under Connection & Provisioning to approve the connection and get fresh tokens.

6. Quick Reference Summary