Research, outline and task agents for Claude that check their own work and never send anything without your OK.
Everyday research and content jobs, with checked facts and nothing sent without your OK.
Ask a question and get a sourced brief. Weak, conflicting or out-of-date sources are flagged, not hidden.
Get a blog or newsletter outline built only on checked facts, with any gaps clearly marked.
Draft emails and task lists step by step. Nothing is sent, saved or posted until you approve it.
Get the research, a structured outline and a step-by-step task list, with every source flagged for checking.
We gave the Starter Pack's Research Agent a small business's 2023 tax notes and asked: "Is this still accurate?" Here's an extract of its answer, from our testing in September 2026.
From your files (not independently checked): - UK VAT registration threshold is £85,000 [Your file] (conflicts with published sources) - 31 January Self Assessment deadline [Your file] (correct for online returns only) Key facts safe to use: - VAT registration threshold is £90,000 [T1] - Threshold rose from £85,000 on 1 April 2024 [T2] Do not use: - £85,000 VAT threshold (superseded) Safety Review Check before relying on this: the VAT figure has been updated, but I saw only snippets. Confirm on GOV.UK before acting.
T1 = primary or official source · T2 = reputable secondary source · T3 = other (blogs, vendors, forums)
It caught the out-of-date figure, labelled the claims from the uploaded notes separately from checked sources, and told you exactly what to verify. Example only: always check current official guidance.
See more real examples, including a false premise and a hidden instruction →
Beginner-friendly, self-checking AI agents that run in your own Claude account, for individuals and small businesses.
Three AI agents that work together and check their own work. Builds on the free Research Agent. Built and tested for Claude.
What "self-checking" means: after every answer, each agent reviews its own work against six clear criteria and tells you what to verify. It can't guarantee accuracy, which is why every answer shows its sources.
What it actually does: searches, evaluates and ranks the evidence, keeps every caveat, and carries it through each stage, so you always know what you know, what you don't, what needs checking, and what's left for you to do.
What you get: a download of agent prompts, an n8n workflow and templates that you use with your own accounts. It's not an app or a subscription.
Running costs: the agents work with a free Claude account. The optional n8n workflow uses your own pay-as-you-go Anthropic API key (about 10p per Standard research run in our testing) and a free Tavily search plan.
Buy now — £29.99 · instant download✓ 14-day money-back guarantee · Instant download
| Free Research Agent | Starter Pack (£29.99) | |
|---|---|---|
| Research agent with a Safety Review | ✅ | ✅ Upgraded: ranked sources, status labels, official-source checks, conflict and freshness checks, depth options |
| Content Outline and Task Runner agents | — | ✅ |
| Agent Team: one Claude Project, a simple menu, pauses between stages | — | ✅ |
| Safe handling of attached files and hidden instructions | Basic | ✅ Full |
| n8n workflow with live web search and email approval | — | ✅ Tested |
| Notion tracker, setup guide, quick reference and sample runs | — | ✅ |
| Price | Free | £29.99 one-time, with free updates |
Not sure yet? Try the free Research Agent first →
Instant download: agent prompts, an n8n workflow, a setup guide and a Notion tracker. No subscription.
Set up the Agent Team in a Claude Project in minutes, or import the n8n workflow and follow the step-by-step guide.
Your agents check their own work and ask before sending anything. Track every agent and run in Notion.
Type your question, and the Agent Team shows a short menu: 1 Research, 2 Outline, 3 Tasks or 4 All three, with A, B or C for how deeply to research. Choose 4 (All three), which runs all three agents in turn, pausing after each one. The Research Agent gathers and checks the evidence, the Outline Agent structures it for your audience, and at the Tasks stage you choose "Draft the full report" to get the complete document, built only on the checked facts.
Adding who it's for (e.g. "for our board") shapes the tone and structure at every stage. A full run is the heaviest option, so it works best on a paid Claude plan.
One self-checking research agent prompt. Paste it into a Claude Project or chat — no coding, live in 5 minutes. No credit card required.
A download containing the Agent Team instructions and three agent prompts (Markdown), an n8n workflow file (JSON), a step-by-step setup guide, a quick reference, two sample runs, and a link to the Notion tracker template, with CSV files as a backup. Everything is plain text, so you can edit it freely.
The Starter Pack is built and tested for Claude, on free and paid plans. The n8n workflow uses Claude through the Anthropic API. We haven't tested it with other AI apps, so we don't claim it works with them.
No. The agents are set up by pasting them into a Claude Project, and the n8n workflow is imported and connected by following the guide. No code is involved.
Start with the free Research Agent if you want to try one self-checking agent first, with no purchase needed. Choose the Starter Pack when you want an upgraded researcher, two more agents that work with it, an Agent Team that runs all three from one Claude Project, an n8n workflow with live web search and approval, and a Notion tracker. The comparison table above shows the differences at a glance. You don't need both.
You're not paying for a smarter AI; you're using your own Claude. You're paying for a tested research method, ready to use: source ranking, conflict checks, staged research, outlining and task planning, approval steps, self-review, an n8n workflow and templates, so you don't have to design all of that yourself.
After every answer, each agent prints a six-point Safety Review, checking things like whether it stayed on task, invented anything, relied on weak or outdated sources, or took any action without your approval. If something affects the main findings, it warns you at the top. Self-checks reduce errors but can't remove them, which is why every answer shows its sources.
Yes, including web search and file uploads. Free plans have usage limits that reset every few hours. In our testing, a free session fitted about two full three-stage runs (research, outline and a drafted task) at Quick scan depth. A single research question uses much less; deep dives and large files use more. The agents pause after each stage, so you can stop early and save your allowance.
Using the agents in Claude is covered by your Claude plan, free or paid. The optional n8n workflow uses your own accounts: n8n Cloud has a free trial and then a monthly subscription (or you can self-host n8n for free), and in our testing a Standard research run cost about $0.12 (around 10p) in Anthropic API credit, with Tavily's free search plan covering the searches. The setup guide shows you how to set a spending limit.
No. Every agent works on its own in Claude. The n8n workflow is for when you want research with live web search and emails that wait for your approval. It runs when you ask it a question; it doesn't run by itself in the background.
Yes, for planning, mapping evidence and outlining. It searches the open web, not academic databases or paywalled papers, so check every source in the original before citing it, and follow your institution's rules on AI use.
The email step has approval built in. When the agent wants to send a report, n8n pauses and shows you the exact subject and report, with Approve and Decline buttons. Nothing is sent until you click Approve.
The agents were tested on Claude Sonnet 5 and Sonnet 5.5 (free plan) and Claude Opus 5.5 (paid plans) in September and October 2026, and the n8n workflow on n8n 2.40.7 with Claude Sonnet 5. The Notion tracker was tested on a free Notion account. Other Claude models should work, but results may differ, and smaller, faster models may follow the agents' rules less closely. AI tools change quickly, and when they do, we update the pack.
Prices are in GBP, and prices shown include UK VAT. Our payment provider, Gumroad, handles the tax: buyers outside the UK may pay a slightly different total depending on their local tax, and you'll always see the full price before you pay.
A 14-day money-back guarantee for every customer: if you're not happy, email hello@agnetic-ai.com within 14 days of purchase for a full refund, even after downloading. See our Terms & Conditions for details.
Yes. When we improve the pack, you get the new version free through your original download link, for as long as we sell it.
No. Personal and internal business use only. You may not resell or redistribute the files themselves.
We collect the minimum needed to deliver your purchase and support you. See our Privacy Policy for exactly what, why, and who it's shared with. When you use the agents, your data goes to the services you choose (such as Claude, n8n, Tavily or Gmail), under their own policies; we never see it.