Why the agent needs a library
An AI model knows a lot about hacking, but only what it learned in training, and in general terms.
Ask a model about a classic SQL injection and it answers well. Ask it for the exploit of a CVE published last month, the endpoints of an in-house product, or your team's procedure for a client, and it has nothing to recall. It may even invent something that sounds right.
A human pentester solves this with their bookmarks and notes: they look the detail up at the moment they need it. Tradecraft gives the agent the same habit. Pick a need below and compare.
What you can add
Knowledge lives in very different places, so Tradecraft understands seven kinds of source and reads each one the way that suits it.
You do not need to know which kind a site is. Paste its address and RedAmon recognises it, or pick one of 105 ready-made sources. Click a kind to see what it is and how RedAmon reads it.
From a site to a library
When you add a source, RedAmon ingests it in the background: it reads the site and builds an organised map of it, ready for the agent.
You can keep working while it runs, and the source's card shows which step it has reached. Press Play to watch a knowledge wiki go through the eight steps, and switch between Live and Stored to see the one step that differs.
Nothing breaks while it runs. A refresh builds a whole new version in the background, and the agent keeps using the old one until the new one is complete. A refresh you cancel leaves the agent with exactly what it had before, and so, by default, does a refresh that fails.
Live or Stored
For each source you decide how it is kept: like a bookmark, or like a copy saved for offline reading.
A good rule: Stored for the sources you rely on most, Live for huge sites, or for sites whose terms do not allow copies. RedAmon recommends one for each site when you add it. A PDF is always Stored; a CVE exploit archive is always Live.
How the agent uses it
During an engagement the library works in four ways, from a quiet reminder to a full research assistant.
The agent's instructions list your sources and what they cover, and what they do not, so it knows when to look and when to rely on itself or the web.
Before each step, RedAmon looks for pages about what the agent is about to do, and puts the best ones in front of it.
The agent searches the library itself and reads the one section it needs, not a whole page. Searching and reading use no AI call.
For a vague need, a helper with its own AI model reads the promising pages and answers with up to three of them and their exact words, all checked.
A fictional lab target. The flow follows how RedAmon really works; the texts are shortened.
Skills and the library work together. RedAmon's attack skills are playbooks: they tell the agent what to do for a kind of attack. The library supplies how exactly: the payload, the command, the steps of the exploit, from a source you trust.
You stay in control
The library is yours, and so is every decision about what goes into it.
A big wiki covers everything. A focus keeps only the pages about the topics you test, such as web payloads and bypasses. You can preview the result before saving.
See every page the agent can read. Block an outdated page, add one the crawl missed, rename a page with a poor title. Your corrections survive every refresh.
Every project uses your library. Turn a source off in one project with a switch, and the chat shows how many sources the next message will use.
A paid knowledge base or a private repository can carry its own login. It is sent only to that site, only in your own chats, and never shown back.
The library's reads go only to its own sites, never to the system under test. Whatever a page says reaches the agent marked as untrusted information, not as instructions.
Sites change. Refresh a source when you like; the old version keeps working until the new one is ready.
When it helps most
The library helps exactly where the model's memory ends. Pick what is in your library and which model drives the agent.
It is not free of cost. Searching and reading use no AI call, but the list of your sources and Library now's suggestions are added to what the agent reads at every step, so a run uses more of the model's input. If a library adds little for your targets, you can trim it or turn Library now off.
Where to go next
Set up your first source in a few minutes, or read how each part works.
Plain-words tour, quick start, and the full guide to Global Settings → Tradecraft.
Tradecraft Lookup →What the agent does with the library: search, reading, the librarian and Library now.
Inside RedAmon Agent →The whole agent, interactively: how it thinks, acts, remembers and asks for approval.