Ask your whole security stack. Right in your browser.
An AI analyst inside Cyproxy. It investigates with the integrations you've connected, follows proven CTI methods, shows you the evidence, and keeps what you find. No AI subscription, nothing to install.
An analyst that shows its work.
What you get: answers backed by your own sources, in the form that makes them clearest.
Checks every source you've connected
Reputation, scanning, threat intel and your own OpenCTI or MISP. It finds the right tools itself and says which source each finding came from.
Follows the same skills as your other agents
Ask for a threat actor profile, Sigma rules or an executive brief and it follows the skill from your Cyproxy account. It's the same skill Claude or Cursor would use, with nothing installed anywhere.
Screenshots, attack paths and charts
See the phishing page a scanner captured, an attack path mapped to ATT&CK, or a trend chart, when they make the answer clearer.
Keeps what you find
Ask it to save results to a collection, and they're there next time: in any session, on any device, in any agent.
Looks things up, changes nothing
Tools that submit, change or delete anything in a connected tool are switched off for the agent. It tells you what to run yourself instead.
Every lookup in your audit log
Each tool call the agent makes is logged with who asked, which integration and tool, and the outcome, the same as calls from your own agent.
Collections: keep it,
reopen it anywhere.
What you get: your working lists and tables in Cyproxy, instead of scattered across chats and spreadsheets.
- 01Ask, then decide. Pull results in a conversation. If they're worth keeping, say "save those to a collection called Fernwood lookalikes". The agent never saves on its own.
- 02Lists or tables. A list holds one value per row: IPs, domains, hashes. A table has columns, and gains new ones as you add data.
- 03Edit, import, export. Edit cells, add rows and columns, import a CSV or export one, in the console.
- 04Any agent, any device. Collections work through the same Cyproxy link in Claude, ChatGPT, Cursor and the Cyproxy Agent.
- 05Shared on Teams. The team works from the same lists: blocklists, watched infrastructure, the tech stack.
| Domain | Registered | Hosting | Risk |
|---|---|---|---|
| fernwood-bank-login[.]com | 19 Oct | AS47583 · CY | High |
| fernwoodbank-secure[.]net | 17 Oct | AS13335 · US | High |
| fernwood-verify[.]co | 14 Oct | AS24940 · DE | Medium |
| my-fernwood[.]info | 9 Oct | Parked | Low |
Simple Agent credits,
free to try on Community.
What you get: one credit per Standard question, however many sources the agent checks to answer it.
- 011 credit = 1 Standard question. A quick check or an investigation across many sources, it's one credit on Standard. Choose Deep when you start a conversation and our most capable model answers, for 2 credits a question; the Ask button always shows which. The tool calls it makes count towards your plan's tool-call credits as usual.
- 02Follow-ups are questions too. "Now chart it" in the same conversation is a new question.
- 03Not counted: messages it declines as off-topic, and answers that fail.
- 04Busy month? Top-up packs: 50 for £10, 150 for £25 or 500 for £70. They never expire.
- 05Your own agent is separate. Using Claude, ChatGPT or Cursor through Cyproxy never uses Agent credits.
| Plan | Agent credits a month | Collections |
|---|---|---|
| Community · free | 5, Standard only | 3 |
| Pro · £19.99 | 50 | 10 |
| Teams · coming soon | 50 per user, pooled | 10 per user, pooled |
| Enterprise · unlimited users | Sized to your team | Custom |
Agent credits reset on the 1st; on Pro, up to a month's worth you don't use carries over. Compare all plans →
Built for security work.
What you get: an assistant you can put in front of your sources without a second thought.
Read-only in your tools
Submissions, deletions and changes are switched off for the agent. Collections, your own data in Cyproxy, are the only thing it can write, and only when you ask.
Your keys never reach the AI
Integration keys stay encrypted in Cyproxy and are added to each call inside it. The agent only ever sees results.
Third-party data stays data
Page text, scan results and feed content are treated as evidence, never as instructions to follow.
Not used to train models
Your questions and results aren't used to train AI models.
Security work only
The agent sticks to security: investigations, intel, detections, reports. Off-topic requests are declined and don't use a credit.
Every lookup audited
The same audit log as every other Cyproxy call, with who asked, the integration and tool, and the outcome.
Ask it about your next alert.
// Pro · 50 Agent credits a month and 10 collections