Recon automation
AI for recon automation in scope
Recon automation with a model means feeding it the output of tools you already ran against assets you are authorized to test, and getting back a deduplicated, ranked list with notes on what to verify first. The model assists with reading and sorting; scope is validated separately and a person approves anything that runs next.
curl https://api.pentestllm.ai/v1/chat/completions \ -H "Authorization: Bearer sk-ptl-live-…" \ -H "Content-Type: application/json" \ -d '{ "model": "auto/security", "messages": [{"role": "user", "content": "Rank these scanner findings by severity and say what to verify first."}] }'What the model does with recon output
- Summarise
- Turn thousands of lines of scanner and enumeration output into a short account of what was found and where.
- Deduplicate
- Merge the same host, service or finding reported by different tools under different names.
- Rank
- Order the results by what looks most worth a person’s time, with the reason for each position.
- Scope check
- Compare every host and domain in the output with your scope file and flag anything that is not in it.
- Notes
- Keep a running inventory of assets and observations that the report can be written from later.
- Gaps
- Point out in-scope assets that the tool output says nothing about, so you can decide whether to look at them.
Scope and authorization
Recon is where scope mistakes start. A wildcard that looks generous, a shared hosting address or a third-party service in the output can each lead a tester onto a system nobody authorized. Put the scope file and the exclusions from the rules of engagement in the system prompt, and ask the model to sort every item into in scope, out of scope or unclear. Unclear items go back to the client, not into the next run.
The sort is a review aid, not a scope check. The model can mislabel an item or suggest a host that is not in scope, so scope validation is done separately, against the signed rules of engagement, and a person approves every target and every next step. PentestLLM is for systems you own or have written permission to test, and the acceptable use policy applies to every request. The rules of engagement tool helps write the scope down before the engagement starts.
A person approves what runs
An agent that can call tools can also call them wrongly. Keep a human approval step between the model’s plan and any command, which terminal agents such as OpenCode and Claude Code offer as a setting. Log what was approved and when, so the activity record matches the report.
Rate limits and test windows in the rules of engagement are part of the approval. A model does not know the client’s change freeze unless you tell it.
Working alongside your existing tools
The model does not replace the tools you already use. It reads their output: exports from Nuclei, Burp Suite or the utilities on a Kali workstation, pasted as text or passed as files. For tools that run through the Model Context Protocol, the Kali MCP guide connects a security build to the Kali project’s own MCP package with an approval step in the client.
Large exports are the main cost. Strip progress lines and banners before sending, and send one tool’s output per request where you can, so the model’s ranking has less noise to work through.
Which builds to use
Send auto/security and the router picks a fast, balanced or frontier security build per request at that build’s price. Deduplicating a short list fits a fast build; ranking a large inventory with reasons benefits from a stronger one. Where a client needs a fixed id, Gemma 4 26B A4B Cybersecurity (gemma-4-26b-a4b-it-cybersecurity, 262,144 tokens of context) is the build the setup guides use.
Choose by context window when the output is long, since a window too small means splitting the input and losing cross-references between tools. Choose a build with tool calling for agent loops that read files or call MCP tools. Then compare price: recon is input-heavy, so the input column in the table below matters most.
What is kept and what it costs
Request and response bodies are not retained by default; the usage log keeps the model, token counts and charge. A key can opt in to seven days of retention for replay, which is worth leaving off for client data. Billing is per token from a prepaid balance from $10 by card, WeChat Pay or Alipay, and a request that would overdraw the balance is refused, not billed.
16 security builds, priced per token
Read from the live catalog on 2026-11-20. Prices are US dollars per million tokens, paid from a prepaid balance that starts at $10. The id auto/security has no price of its own: each request is billed at the build that answered.
| Wire formats | |||||
|---|---|---|---|---|---|
| Qwen3.8 27B Cybersecurity | Security | $0.40 | $2.40 | 262K | |
| Uncensored Qwen3.8 27b Cybersecurity | Security | $0.40 | $2.40 | — | |
| Gemma 4 26B A4B Cybersecurity | Security | $0.4224 | $1.3376 | 262K | |
| Gemma 4 26B A4B Cybersecurity (Uncensored) | Security | $0.48 | $1.52 | — | |
| GLM-5.3 Flash Cybersecurity | Security | $0.5093 | $1.6975 | 1049K | |
| GLM-5.3 Flash Cybersecurity (Uncensored) | Security | $0.5093 | $1.6975 | — | |
| Red Team (Flash) | Security | $0.90 | $3.00 | — | |
| Red Team (Pro) | Security | $1.50 | $6.00 | — | |
| OrcaCyber MiniMax M3.1 Flash | Security | $2.00 | $4.00 | 262K | |
| OrcaCyber MiniMax M3.1 Flash (Uncensored) | Security | $2.00 | $4.00 | — | |
| OrcaCyber Qwen3.8 Flash Next (Uncensored) | Security | $2.00 | $4.00 | — | |
| OrcaCyber Qwen3.8 Flash Next | Security | $2.00 | $4.00 | 262K | |
| Uncensored Qwen3.8 Flash Next Cybersecurity | Security | $2.00 | $4.00 | — | |
| Cyber Frontier | Security | $5.25 | $15.75 | 1049K | |
| OrcaCyber Zero 1.5 | Security | $7.50 | $18.75 | 999K | |
| OrcaCyber Zero 1.5 (Uncensored) | Security | $7.50 | $18.75 | — |
Questions
Does the model run the recon tools?
Not on its own. It reads output from tools you already ran in scope. In an agent setup it can propose a command, and a person approves it before anything runs.
Can the model expand the scope?
It should not. Give it only in-scope output and the scope file, ask it to flag anything outside that file, and check those items with the client before anyone tests them.
Which tools does it work with?
Any tool whose output is text or a file, including common scanners and the utilities on Kali. The model reads the output; it does not need an integration with the tool.
Which build is best for long scanner exports?
One with a large context window, so the whole export fits in one request. Send auto/security or pin a build from the table on this page.
What does it cost?
Per token from a prepaid balance with a $10 minimum top-up. On 2026-11-20 input prices for security builds start at $0.40 per million tokens.
Read next
One key, from $10
PentestLLM is for systems you own or have written permission to test. The acceptable use policy says what is not allowed.