Documentation - Cadence IP

Connecting an AI provider

Navigate to Settings → AI Settings to configure your AI provider. Cadence IP supports three providers:

Anthropic (Claude)

Models: Claude Sonnet, Claude Opus. Recommended for complex legal reasoning and long document analysis.

OpenAI (GPT)

Models: GPT-4o, GPT-4o mini. Good general-purpose option with fast response times.

Google (Gemini)

Models: Gemini Pro, Gemini Flash. Large context window for processing lengthy patent specifications.

1

Select a provider

Choose your AI provider from the dropdown. Each provider requires its own API key.

2

Enter your API key

Paste your API key from the provider's dashboard. The key is encrypted and stored securely. Cadence IP never sends your key to any third party.

3

Choose a model

Select the specific model to use. More capable models produce better results but cost more per request. You can change models at any time.

API costs

AI usage is billed by your chosen provider based on token consumption. Cadence IP does not add any markup. Typical matter interactions cost between $0.01 and $0.10 per request depending on the model and context length.

Global AI guidance

The AI Guidance field under Settings → AI Settings lets you provide organisation-wide instructions that shape how the AI behaves across all matters. This is a free-text field where you describe your firm's tone, conventions, and preferences.

Example global guidance

You are an assistant for a patent and trademark firm based in Sydney, Australia. Our firm is "Patentec".

When drafting emails: Use formal but approachable tone. Address recipients as "Dear [First Name]" for existing clients and "Dear [Title] [Surname]" for new contacts. Sign off with "Kind regards" followed by the attorney's name.

When discussing deadlines: Always reference the specific deadline date and the consequence of missing it. Use Australian date format (DD/MM/YYYY).

When referencing legislation: Cite the Patents Act 1990 (Cth) for Australian patents, the Trade Marks Act 1995 (Cth) for Australian trademarks.

Be specific

The more specific your guidance, the better the AI performs. Include your firm name, jurisdiction, preferred terminology (e.g., "specification" not "patent application document"), and formatting conventions.

AI chat on matter pages

Every matter page includes an AI chat panel. The AI has full context about the matter, including:

Matter name, type, current step, and all field values
Contacts assigned to each role (Applicant, Inventor, Attorney, etc.)
Recent emails sent and received
Tasks and their status
The knowledge base entries for this matter type
Your global AI guidance instructions

This context-awareness means you can ask questions naturally without repeating matter details:

You

Draft an email to the applicant reporting the examiner's objections from the latest office action. Recommend amending claims 1 and 3.

AI

I've prepared the following actions for your review:

Send Email

To: John Smith (Applicant) · Re: Examination Report - Widget Fastener

Change Step

Written Opinion → Report to Applicant

Create Task

Follow up with applicant on claim amendments · Due: 2 weeks

AI chat on a PCT Application matter page

@mention guidances in chat

Guidances are specialised AI instruction documents stored in your knowledge base. You can reference them directly in chat using the @ prefix to activate specific expertise.

When you type @ in the chat input, a dropdown appears listing available guidances for this matter type. Select one to include its instructions in the AI's context for that message.

Using @reporting_office_action guidance

Your knowledge base has a guidance called @reporting_office_action that contains your firm's standard approach to reporting examination results to clients: structure, tone, required disclaimers, and fee estimates.

In chat, you type:

@reporting_office_action Draft the report email for the latest examination report. The examiner raised novelty over D1 and inventive step over D1+D2.

The AI follows your firm's reporting template exactly, including the standard fee estimate for responding and the deadline for filing a response.

Multiple guidances

You can mention multiple guidances in a single message. For example, @reporting_office_action @fee_estimate_examination combines reporting instructions with your fee schedule guidance.

Propose and implement

The AI does not take actions directly. Instead, it follows a propose → review → implement pattern that keeps you in control:

1

AI proposes actions

When the AI determines that actions should be taken (send an email, change a step, create a task), it presents them as proposals with full details for your review.

2

You review each proposal

Each proposed action shows exactly what will happen: the email content, the step change, the task details. You can edit the proposal before implementing.

3

Implement selectively

Click Implement on individual proposals to execute them. You can implement some and discard others. For example, send the email but skip the step change.

The AI can propose these action types:

Send Email

Draft an email with subject, body, and recipients. You can edit before sending.

Change Step

Move the matter to a different workflow step, which may trigger automations.

Create Task

Add a task to the matter with a title, description, assignee, and due date.

Update Fields

Set or change field values on the matter, such as updating an application number.

Safety by design

The propose/implement pattern ensures the AI never sends an email or changes data without your explicit approval. This is critical in legal practice where accuracy and client communication standards are paramount.

File attachments in chat

You can attach files to your chat messages for the AI to analyse. Drag and drop or click the attachment button to upload PDFs, Word documents, images, or text files.

Common use cases in IP practice:

Upload an examination report PDF and ask the AI to summarise the objections
Attach a prior art document and ask how it differs from the claimed invention
Upload a foreign associate's letter and ask the AI to extract action items and deadlines
Attach a draft specification and ask for a review of the claims for clarity

Analysing an examination report

You receive a 12-page examination report from IP Australia on matter M-42. Instead of reading through it manually to extract the key points, you upload the PDF to the AI chat and ask:

"Summarise the objections in this examination report. For each objection, identify the relevant claim(s), the cited prior art, and the examiner's reasoning."