Tax
Without vitamins"Drop your return at a CRA drop box."
With vitamins"Those drop boxes closed on 29 May 2026. File online instead."
Included on Developer Pro or higher
Your agents get a regular, checked dose of fresh outside knowledge, compressed so they read less and know more.

The world keeps moving after setup. Your agents don't. Rules change, platforms update and best practices move on, and most teams never find out their agents need refreshing until a customer gets an out-of-date answer.
We tell you where your agents' knowledge is missing or out of date.

Here's what changes when your agents know what changed.
Example answers
Without vitamins"Drop your return at a CRA drop box."
With vitamins"Those drop boxes closed on 29 May 2026. File online instead."
Without vitamins"FedNow payments are capped at $1 million each."
With vitamins"FedNow's per-payment limit has risen to $10 million."
Without vitamins"Run the older page-speed audits."
With vitamins"Lighthouse 13, Google's page-speed tool, replaced those audits. Use the new ones."
Without vitamins"Indiana and Kentucky's new privacy laws aren't in force yet."
With vitamins"New privacy laws are in force in Indiana and Kentucky from 1 Jan 2026."
It isn't one kind of knowledge. Laws, tax, payments, privacy, software, AI tools and ad platforms all keep moving.

The EU AI Act: who enforces it and when the high-risk rules apply.
Wrong dates mean wrong advice.

Canada's tax agency closed all 45 of its drop boxes on 29 May 2026.
Old steps send customers to the wrong place.

FedNow's per-payment limit rose from $1 million to $10 million, and new ACH fraud-monitoring rules arrive in 2026.
Limits and rules decide what's possible.

New US state privacy laws in force in Indiana and Kentucky from 1 Jan 2026.
Missing one is a compliance gap.

Lighthouse 13, Google's page-speed tool, replaced older page-speed audits.
Old checks give old scores.

Each new ChatGPT and Claude model generation brings new prompting guidance.
Yesterday's prompts underperform.

Knowledge packs for Google, Meta, TikTok, Amazon and Reddit ads.
Ad platforms change their rules often.
Not sure which of these your agents are missing? We'll tell you.
Book a callResearched, double-checked, made safe, crystallized and delivered weekly, so you don't have to.
Research. At each field's own pace, monthly for most agents, we research what changed in that agent's field. We use public sources only. Your business data never goes into the research.
Check. Every proposed update must cite a dated public source. A second AI, from a different AI company than the one that proposed it, then challenges each update: keep it, revise it or drop it.
Safety rules. A plain, cited fact, like a new platform limit, a rule change or a new feature, can apply on its own. Anything touching money, regulated advice (legal, tax, medical or financial), client accounts, passwords or access, or contacting people is held for a person to review.
Crystallize. Our crystallization agent condenses the checked research into a compact pack of one-line, sourced facts, rules and known gaps. Research skips what the agent already knows and looks only for what changed.
Apply. The agent's knowledge guide is updated, with a dated "what changed" note. Every change can be rolled back exactly.
Deliver. Each week, whatever changed is delivered to your agents by MyZone, with a short "what's new" note. Each pack is checked before it loads into your agents.
A deep research run can read hundreds of thousands of tokens of web pages. Our crystallization agent boils all of that down to a pack of about 1,500–2,500 tokens that your agents can actually use. It also decides how each agent's brain is layered, so new knowledge lands in the right place.

One GPT deep research report: about 35,000 tokensMore than 10,000 words and about 200 source links. One we ran with OpenAI's deep research model.
One vitamin pack: about 1,500–2,500 tokens of one-line, sourced facts
About 15–20 times less for your agent to read
Your agent reads the pack, not the report, every time it looks something up.
Report size estimated from its length at about 4 characters per token.
Measured, October 2026: packs of 1,450 to 2,600 tokens, with 14 to 33 sourced facts each.
Each update is a single cited line of at most 280 characters, listed in a dated "what's new" note.
Our Industry Analyst agent does this today: a knowledge guide, refreshed monthly on a set schedule, on AI adoption by industry, how companies use ChatGPT and Claude at work, rolling them out, competitor AI moves and the rules on business AI use. Here's roughly what running that research yourself would take.
From its latest guide: UK Office for National Statistics, 8 Oct 2026: 30% of UK businesses used AI in September 2026, and 51% of those with 250 or more employees (official statistics in development).
| What you'd have to run yourself | Est. per agent, per month |
|---|---|
| Deep research on 4 topics | $4–12 |
| Check every source with a second AI | $2–5 |
| Condense it, merge it with what the agent knows, drop outdated facts | $2–4 |
| Extra reading on everyday tasks without a compact pack | about $7 |
| Total | about $15–28 |
Plus what never shows up on an AI bill: setting it up, scheduling it, fixing it when it breaks, and knowing you need it at all.
Doing it yourself, for a fleet of
Illustrative estimate based on published AI research and model prices as of October 2026. Your own costs depend on topics, models and how often your agents look things up.
Included on Developer Pro or higher.
See plans & pricingMost AI agents are frozen the day they're installed, while rules, platforms and best practices keep changing. We tell you where your agents' knowledge is missing or out of date.
Agent Vitamins is included on Developer Pro, Fully Managed and Enterprise, and plan prices stay exactly as they are. We pay for the research and checking; your own AI usage covers your agents reading the packs.
The research and checking happen once, centrally, at MyZone, not on your install. Your agents read a compact, already-checked pack only when a task needs it, instead of searching the web on every job.
No. The research uses public sources only, and your business data never goes into it.
Every update must cite a dated public source, and a second AI from a different AI company challenges it before it's kept. Every change can be rolled back exactly. As of October 2026, no update has had to be rolled back.
No. Plain, cited facts can apply on their own. Anything touching money, regulated advice, client accounts, passwords or access, or contacting people waits for a person to review.
Each expert knows one product or area of law in depth. Here's what its vitamins keep fresh, on Developer Pro or higher.

QuickBooks Online, its API and app rules, and Intuit's AI features

Sage 300 inventory, purchasing and sales data, its Web API and releases

Canada's SR&ED tax credit: rates, deadlines and how CRA reviews claims

HubSpot plans, Breeze AI, API versions and what's being retired

Jobber plans, payments, integrations and its developer platform

Red Hat Enterprise Linux release and support dates, Red Hat AI and agents

Federal and all 50 states' rules on wages, overtime, leave and hiring

Federal, provincial and territorial employment standards and rights
More experts are on the way. We're giving more of our agents a dedicated expert, in waves.
See plans & pricing →Agent Vitamins is included on Developer Pro, Fully Managed and Enterprise. Book a call and we'll tell you where your agents' knowledge is missing or out of date.