# Knowledge Curator: Every Expert Update Checked Before Use > The Knowledge Curator is the AI knowledge review agent in the Ai1 platform by MyZone AI: when an expert agent proposes its monthly knowledge updates, it checks each one for source, recency, conflicts and money at risk, then rules keep, revise or drop with a confidence level and a reason. Canonical page: https://myzone.ai/pages/agents/ai-knowledge-curator-agent Part of Ai1, by MyZone AI. Book a Curator walkthrough: https://calendly.com/d/ct6h-tcy-8qf/ai1-demo?a1=Knowledge%20Curator&utm_source=myzone.ai&utm_medium=agent-page&utm_content=ai-knowledge-curator-agent-final Last updated: 2026-10-05. Reviewed by the MyZone AI team. A second opinion on every update to your expert agents' know-how. ## At a glance - **How you use it:** You don't talk to it: it works inside an expert agent's monthly refresh - **Source check:** Official sources count most; an update with no link or no date is dropped - **Money at risk:** Anything touching budgets, bids or scaling gets extra care - **Each ruling:** Keep, revise or drop, with a confidence level and a reason - **A second model:** It runs on a different AI model from the expert it checks - **What it never does:** Change the expert itself, research the web or message anyone ## What the Knowledge Curator does - **Checks the source:** An official source counts more than a practitioner, and a practitioner more than a single blog post. An update with no link or no date is dropped. - **Checks it is current:** Asks whether the evidence is newer than what it would replace, and whether it still holds today. - **Flags conflicts:** If an update contradicts what the expert already knows, it weighs the evidence, and where the question stays open it shows the disagreement instead of hiding it. - **Weighs the money at risk:** Anything that touches budgets, bids or scaling is marked for extra care, and anything that could waste money if wrong is marked for review. - **Keeps bad input out:** Drops and flags any update that contains customer data or text aimed at manipulating an AI. ## How the Knowledge Curator differs from an agent checking its own research An agent that checks its own research tends to miss the mistakes it made. The Knowledge Curator runs on a different AI model and reviews every proposed update for source, recency, conflicts and money at risk. Each gets keep, revise or drop with a confidence level and a reason. It writes nothing outside its rulings, and the expert applies what passes through its own approval path. How this differs from the Memory Manager (https://myzone.ai/pages/agents/ai-memory-agent): the Memory Manager keeps everything your Ai1 server knows healthy and current, while the Knowledge Curator only reviews the monthly updates your expert agents propose. ## How it works 1. **Expert researches** (Once a month): Once a month an expert agent researches what has changed in its field. 2. **Updates proposed** (After the expert's research): The expert hands over its proposed updates for review. 3. **Independent review** (Before anything is learned): The Knowledge Curator judges each one on source, recency, conflict and money at risk. 4. **Rulings returned** (Before the expert applies anything): Keep, revise or drop, with confidence and reasons. The expert applies what passes through its own approval path. ## When to use it - **Your TikTok Ads Agent proposes a batch of monthly updates:** Each one is ruled keep, revise or drop, with a confidence level and a reason. - **An update would change how budgets are scaled, based on one blog post:** It is marked for review because the source is thin and money is at risk. - **An update would change how the expert behaves, not just what it knows:** It is flagged so the change goes through the expert's own approval path. - **You want to know why an expert agent changed its advice:** Its rulings show which updates were kept and why. ## What you get - A keep, revise or drop ruling on every proposed update - A confidence level and a reason for each ruling - Flags on updates that put money at risk - A record of open disagreements between reviewer and expert ## Example: monthly knowledge review rulings Example with a fictional company. Names, people and figures are invented to show the agent's output. Any resemblance to a real company or person is unintended. What it was asked: Before our TikTok ads expert learns anything from this month's research, check each proposed update and tell us what to keep, revise or drop, and why. The Knowledge Curator ran once inside the TikTok ads expert's October refresh, on a different AI model from the expert's, and judged each proposed update on source quality, how current it is, conflict with the playbook and money at risk, plus checks for hidden instructions and customer data. It did no research of its own, changed nothing in the expert and contacted nobody; a person on the client's team approves, rejects or adjusts each update before it is learned. The company, updates, sources and dates are invented, so no real ad account or playbook was reviewed. Run date: 1 October 2026. ### What it found - Of 8 proposed updates to your TikTok ads expert, 3 were kept, 3 sent back for revision and 2 dropped, each with a confidence level and a one-line reason. - A tip to double budgets after two good days was dropped: its only source was an agency selling a scaling course, it contradicts the expert's step-by-step rule, and it could waste money. - One forum tip hid a line telling AI assistants to skip review, so it was dropped and flagged. A change to how often the expert works was flagged for its own approval path. ## Guardrails - It runs on a different AI model from the expert it checks, so it can catch mistakes the expert's own model might miss. - It writes nothing outside its rulings. The expert applies what passes through its own approval path. - Refresh inputs are public information only. Any update with customer names, account numbers or spend figures is dropped. - It has no channel and sends no messages. The expert agent reports the outcome. ## Frequently asked questions ### How do you fact-check the updates an AI expert agent wants to learn? Check each one before it is used. The Knowledge Curator weighs the source, asks whether the evidence is newer than what it would replace, looks for conflicts with what the expert already knows and marks anything that could waste money. An update with no link or no date is dropped. It only reviews updates an expert agent proposes. You cannot ask it to research something: research belongs to the Research Agent. ### What keeps an AI agent's knowledge from drifting over time? A review step between research and learning. Your expert agents only learn from updates that survived the Knowledge Curator's independent review, and where it still disagrees with the expert, that disagreement is recorded for you. It also drops any update containing customer data or text aimed at manipulating an AI. Refresh inputs are public information only, so an update that contains customer names, account numbers or spend figures is dropped. ### Why should a different AI model review an expert agent's work? A second model is more likely to spot mistakes the first one would make, the same way a second reviewer catches what the author misses. That is why the Knowledge Curator runs on a different AI model from the expert it checks, and writes nothing outside its rulings. ### Does the Knowledge Curator change the expert agent, or contact us? No. It writes nothing outside its rulings. The expert applies what passes, and any change to how the expert behaves goes through that expert's own approval path. It has no channel and sends no messages. The expert agent reports the outcome. ## About Ai1 Ai1 is the AI operations platform by MyZone AI, where each client runs on its own private server. The Knowledge Curator is not sold on its own: every Ai1 agent, including this one, is included on Developer Pro and every Fully Managed option, with no per-agent charge. Developer Core includes the development agents. Pricing: https://myzone.ai/pages/services/ai1-pricing. Security: https://myzone.ai/pages/security.