Agent

Overview

The Agent is an AI assistant purpose-built for contract analysis. Unlike general-purpose chatbots, it has direct access to your structured contract data — classified clauses, typed attributes, party scores, and the full taxonomy of provision types across your corpus. It can answer questions about individual documents, run SQL-level aggregations across thousands of contracts, generate exportable data tables, and produce tracked-change redlines.

Starting a conversation

Open the Agent from the left sidebar. You'll see a welcome screen with suggested prompts based on your current context. If you have a document selected, the Agent will offer document-specific suggestions.

Type a question or select a suggested prompt to begin. The Agent streams its response in real time and shows its reasoning process in expandable sections.

You can also use the Agent by email. Send a message to your ContractRabbit agent address, usually agent@contractrabbit.com, from the email address associated with your account. Attach documents when you want them processed, or ask questions about documents already in your workspace.

What you can ask

Document Q&A

Ask questions about a specific document:

• "What is the governing law?"
• "Summarize the indemnification clause"
• "What are the termination conditions?"
• "List all monetary amounts mentioned in this agreement"

The Agent reads individual clauses at full resolution — it finds the relevant sections first, then reads the complete text, rather than relying on truncated summaries.

Corpus search

Query across all your documents:

• "Find an NDA with a duration greater than 5 years"
• "Which agreements have a non-compete clause?"
• "Find all contracts with Acme Corp"

Corpus analytics

Get aggregate insights powered by real SQL aggregations over your extracted attribute data — not LLM summarization:

• "What is the average non-compete duration in our NDAs?"
• "Show me the 95th percentile contract value by document type"
• "What percentage of our service agreements have a limitation of liability cap below $1M?"
• "Compare governing law jurisdictions across our portfolio"

Available operations include average, min, max, sum, count, percentiles (25th through 99th), standard deviation, variance, and histogram distribution.

Data tables

Ask the Agent to build exportable tables from your contract data:

• "Build a table of all NDAs showing effective date, expiration date, governing law, and duration"
• "Show me all service agreements with party, counterparty, contract value, and termination notice period"

Data tables support calculated columns (days until expiration, percentage elapsed, months since effective date), coalesced columns (multiple attribute paths merged into one column), and filters by document type, jurisdiction, date range, or party.

Reports and exports

Ask for a report when you want the answer in a structured format:

• "Send me a CSV of all NDAs with governing law, term, counterparty, and renewal notice period"
• "Create a report of service agreements expiring in the next 90 days"
• "Show the top 20 outlier liability caps in our customer agreements"

In the web UI, large tabular results open as interactive tables that support sorting, filtering, and export. Over email, the Agent includes a compact preview in the reply and attaches the full CSV.

Use case catalog

The Agent routes requests to different workflows depending on whether you are asking it to find documents, answer factual questions, run corpus analytics, draft language, edit a document, compare documents, or align against a standard.

Find and inspect documents

Use the Agent when you know part of a name, party, clause, attribute, or phrase but do not know where the document lives:

• "Open the Acme NDA and show me the confidentiality section"
• "Find documents that mention reverse engineering restrictions"
• "Find agreements with BigCo as a party and Delaware governing law"
• "Show me the versions of this document"

The Agent can search by file name, document text, extracted attributes, parties, and classified spans. When it finds the right document, it can open the file in the viewer or link you directly to the relevant text.

Retrieve exact facts from documents

For named-document questions, the Agent first identifies the target document and then reads the matching spans before answering:

• "What period does the confidentiality restriction apply to in the SRAI non-disclosure agreement?"
• "What is the termination notice period in the current agreement?"
• "Who are the parties, and what are their roles?"
• "List the monetary amounts and where they appear"

If a direct text search does not find the answer, the Agent can fall back to extracted attributes such as dates, governing law, notice periods, party names, and document-specific values.

Build reports and corpus tables

Ask for tables when you need repeatable, exportable work product rather than a narrative answer:

• "Make a table of ten non-solicit provisions from Delaware-law NDAs and include what makes each provision different"
• "Show the ten NDAs with the longest non-solicit durations"
• "Create a report of customer agreements with effective date, expiration date, counterparty, renewal notice period, and days until expiration"
• "Send me a CSV of contracts where the liability cap is below fees paid in the last 12 months"

The Agent can discover canonical attribute paths, combine similar columns, calculate derived values, filter by document type or date, and return the full table as an export.

Compare attributes across the corpus

Use corpus analytics for questions that compare extracted terms across many documents:

• "How many NDAs have forum in a different state than governing law?"
• "Find contracts where the termination notice period differs from the renewal notice period"
• "What percentage of service agreements have an unlimited indemnity?"
• "Show a histogram of non-compete duration by document type"

These questions run against structured contract data, so the answer can include counts, percentages, averages, percentiles, distributions, and the underlying documents.

Draft from precedent and corpus language

The Agent can draft language using documents and classified clause examples as the source material:

• "Write an indemnification provision based on SPAs in the corpus, using language most favorable to Buyer"
• "Draft an NDA based on NDA CORPUS 7, but increase the non-solicit duration to five years"
• "Use the party names from one agreement and the indemnification provision from another"
• "Create a new NDA based on our standard template"

For precedent-backed drafting, the Agent retrieves representative clauses, compares alternatives, and states the assumptions it used.

Make targeted edits

When you ask for a focused change, the Agent finds the exact text span before proposing or applying a revision:

• "Change the non-solicit duration to five years"
• "Replace the indemnification section with the Buyer-favorable version we discussed"
• "Insert a notice requirement after the termination clause"
• "Accept and apply the proposed revision suggestions"

The Agent is conservative with edits. If a request could affect many spans, ambiguous defined terms, or the wrong document, it asks for confirmation instead of applying a broad change.

Review, mark up, and align

Use full-document workflows when you want the Agent to analyze an agreement against a standard, playbook, reference document, or precedent:

• "Review and mark up the entire agreement against our NDA playbook"
• "Align this contract with our standard NDA template"
• "Analyze this agreement against the reference document I uploaded"
• "Write a share purchase agreement based on the DSPA precedent and the DTS term sheet"

In the web UI, these workflows can produce linked source evidence, recommendation lists, and tracked-change revisions. Over email, the Agent returns a concise summary and attaches generated documents when available.

Compare documents and pasted text

The Agent can compare stored documents or text you paste into the chat:

• "Compare NDA CORPUS 6 with NDA CORPUS 7 and summarize the differences"
• "Compare these two pasted sections and say which version is better for the discloser"
• "Make a chart comparing section, topic, original text, revised text, difference, and party favorability"

Comparison results can include added, removed, and modified clauses, party-favorability notes, and links back to the source documents. Detailed structured comparison artifacts stay available in the persisted chat.

Review for precedent and market risk

The Agent can use your corpus taxonomy to identify issues that are hard to see from one document alone:

• "What clauses are unusually rare in this agreement?"
• "What important clauses are missing compared with similar agreements?"
• "How common is this provision in our corpus?"
• "What provision usually follows this clause in similar agreements?"

These workflows are useful for spotting missing provisions, unusual drafting patterns, and terms that differ from comparable agreements.

Advanced actions

Review for issues

Ask the Agent to review a document against your corpus:

• "Review this agreement for issues"
• "What clauses are missing compared to similar NDAs in our corpus?"

The Agent identifies missing clauses (provisions that appear in a high percentage of similar documents but are absent), unusual provisions (clauses found in fewer than 10% of comparable documents), and terms that deviate from your corpus norms. Each issue includes a frequency percentage showing how common or rare it is.

Optimize for party

The Agent can suggest tracked changes to make a document more favorable for your side:

• "Optimize this for our party"
• "Suggest improvements to the indemnification section"

This generates specific revision recommendations that appear as inline tracked changes in the document viewer, which you can accept or reject individually.

Align with reference document

Upload a reference document or select an existing standard, then ask the Agent to align your current document:

• "Align this contract with our standard NDA template"

The Agent compares clause-by-clause, generating recommendations with importance scores and category labels. Each recommendation shows the specific text change and which reference clause it's based on.

Compare documents

Ask the Agent to compare two documents side by side:

• "Compare this SPA with the one from Acme Corp"

The comparison identifies added, removed, and modified sections, clauses, parties, and text spans. Results include corpus analytics showing how common each differing provision is across your portfolio.

Generate documents from templates

• "Create a new NDA based on our standard template"

The Agent can copy an existing document as a starting point for a new agreement, immediately opening it in the editor.

Ways to interact with the Agent

The web UI and email use the same agentic workflow and the same persisted chat history. Email is best for quick requests from your inbox; the web UI is best when you want to inspect source text, use interactive artifacts, apply revisions, or continue deeper review.

Every Agent email reply includes a Continue in Contract Rabbit link. Use that link to open the persisted chat session, inspect richer artifacts, and keep working in the web UI.

Using the web UI

Use the web UI when you want the Agent to work with visible context:

1. Open Agent or open a document and use the split-panel Agent.
2. Ask a direct question or choose a suggested prompt.
3. Review linked clauses, generated tables, comparison views, or revision recommendations.
4. Accept, reject, export, or continue the conversation from the same chat.

The Agent automatically tracks the current document when a document is open. It can also open additional documents beside the chat when a request needs comparison, alignment, or inspection.

Using email

Use email when you want to start or continue work without opening the app:

1. Send an email to your ContractRabbit agent address, usually agent@contractrabbit.com.
2. Put the request in the email body. Attach documents if they need to be uploaded or reviewed.
3. For reports, say what columns or filters you want. The Agent will return an inline preview and attach the full CSV.
4. Reply in the same email thread to clarify, refine, or continue the request.

Email replies are intentionally concise. Full structured outputs, source links, document viewer actions, and comparison artifacts are preserved in the linked chat session.

Clarifications

The Agent tries to execute with best-effort defaults and states assumptions in its response. It asks clarifying questions only when execution would likely be wrong, such as:

• The target document is missing or ambiguous
• Multiple documents or standards match equally well
• The requested edit scope could affect the wrong text
• The intent is too low-confidence to proceed safely
• A targeted edit would produce a larger-than-expected set of revisions

Document context

When you have a document open in the split-panel viewer, the Agent automatically knows which document you're looking at. You don't need to specify the document name — just ask "What's the termination notice period?" and the Agent reads from the currently open document.

The Agent can also open documents in the viewer alongside the chat, creating a true side-by-side workstation where you can read the contract while discussing it.

Email attachments become document context for that email-backed chat. If you reply later in the same thread, the Agent can continue from the same documents and conversation history.

How the Agent differs from generic AI

Managing chats

• Chat history is accessible in the sidebar — click any previous conversation to resume it
• Add participants to a chat for team collaboration
• Upload reference documents for comparison or alignment context
• All conversations are persisted and searchable