Turn on packaged AI agents from a catalog. Each one reads a scoped slice of your firm data, runs overnight, and puts read-only insights and one-click suggested actions in an in-app AI Insights feed. The agents suggest. A person approves. Nothing acts on its own.
Beta · Pricing coming soon · You approve every action
The AI Insights Feed
Last overnight run · 6:00 AM
CRA notice for Maple Leaf Cafe Inc. sets a response deadline of May 2.
Suggested: link email to client · create reminder · flag deadline
Sam is carrying 14 open steps this week while Dana has room to spare.
Suggested: reassign 3 steps from Sam to Dana
River Valley Dental payroll filing is 2 days out and still unassigned.
Suggested: create a filing-step task · flag deadline
Northside Logistics T2 has a blocked step with no assignee for 5 days.
Suggested: create a filing-step task · reassign work
Every card is read-only until you accept it. When you click a suggested action, the real change runs under your identity, with full audit.
How It Works
Pick from the agent catalog and switch on only the ones you want. AI runs only after your firm gives explicit consent through an AI data-processing disclosure.
Each night, every agent reads its scoped slice of your data and writes read-only insights and suggested actions into the AI Insights feed. Nothing changes while they run.
Open the feed, read the insights, and accept the suggestions that make sense with one click. The real change runs under your identity through the normal MyCPACRM services.
The Agent Catalog
Turn on only the agents you need. Each one reads a scoped slice of your firm data and drops its findings into the AI Insights feed. More agents are on the way.
Runs nightly over the shared inbox. Sorts and classifies each email, such as CRA notice, documents received, question, scheduling, payment, or FYI. Writes a one-line summary, links it to the right client, and pulls out deadlines.
Suggests
Can send an optional morning digest.
Looks across staff assignments, open steps, logged time, and deadlines to show who is overloaded or behind versus who has room to spare, and recommends reassignments. Built for managers.
Suggests
Aimed at managers watching team capacity.
Reads CRA notices, identifies the notice type, explains what it means and the next step, and pulls out response deadlines so nothing slips past you.
Suggests
Runs overnight in Beta. Real-time on-arrival handling is coming later.
Surfaces the filings and steps most at risk of slipping, whether close to a deadline, behind, or unassigned, with the reason, in priority order. Gives a firm-wide view and a per-owner view of what is at risk.
Suggests
Firm-wide and per-owner "what is at risk" views.
Spots stalled steps, missing assignees, and out-of-order or blocked steps, then proposes the next action for each filing so the work keeps moving.
Suggests
Keeps every filing's steps unblocked and assigned.
The catalog is built to grow. New agents can be added over time, and each one is opt-in.
How It Stays Safe
The agents produce read-only insights plus suggested actions. Nothing acts silently, and v1 has no automatic actions. When you accept a suggestion, the real change runs under your identity through the normal MyCPACRM services, with full audit.
Suggest, then approve
No automatic actions in v1
Step 1 · The agent suggests
Overnight, the agent reads its scoped slice of data and writes a read-only insight with a suggested action into the feed. Nothing has changed yet.
Step 2 · You approve
You click accept. The real change runs under your identity through the normal MyCPACRM services, and it shows up in the audit trail like any other action you take.
You choose the AI provider. Claude is the default, and a private or self-hosted model option exists for firms that cannot send data to a third party. AI runs only after your firm gives explicit consent through an AI data-processing disclosure.
Claude by default
The latest Claude models power the agents right out of the box.
Private model option
Prefer to keep everything in house? Run a private or self-hosted model instead.
Consent first
No agent runs until your firm turns it on and accepts the AI disclosure.
Questions
It is a paid Beta add-on. Your firm turns on packaged AI agents from a catalog. Each agent reads a scoped slice of your firm data, runs overnight, and surfaces read-only insights and one-click suggested actions in an in-app AI Insights feed. The agents suggest, and a person approves.
No. Agents only suggest. Every action is one-click and approved by a person, then runs under that person's identity with full audit. There are no automatic actions in v1.
The v1 catalog has five agents: Email Organizer, Workload, CRA Notices, Deadlines, and Filing Steps. All of them run overnight in Beta. Real-time handling, such as processing a CRA notice the moment it arrives, is planned for a later phase.
Agents can only see your own firm's data, never another firm's, and honor personal-mailbox privacy. AI runs only after your firm gives explicit consent through an AI data-processing disclosure. Claude is the default provider, and a private or self-hosted model option is available for firms that cannot send data out.
Pricing is coming soon while the AI Assistant is in Beta. Reach out and we will let you know as soon as plans are available.
See how the AI Assistant surfaces what matters each morning and lets you approve the right action in one click. Join the Beta and we will walk you through it end to end.