AI for red teams and pentesters
Mainstream models stop at the offensive part. Nef 4.0 and Sifu 3.5 are built for authorized red team and penetration testing work, from recon to the report, in your terminal or the chat.
A second pair of hands for in-scope work, with you deciding what runs.
Sort scan and enumeration output, spot what stands out and plan the next step within your rules of engagement.
Read requests and responses, reason about auth and access control, and draft test cases for in-scope targets.
Build and adapt proof-of-concept code for in-scope findings, so your client sees real impact instead of a theoretical risk.
Audit source, infrastructure code and configs for the bugs and misconfigurations behind your findings.
Write and adapt the tools and scripts an engagement needs in your own coding agent, with you deciding what runs.
Turn notes into clear findings with impact, evidence and fixes your client’s developers can act on.
Nef 4.0 and Sifu 3.5 are built for authorized offensive security: red team and pentest work, security research and code. Made for hackers, coders and security researchers, and just as good for everyday work.
Our strongest model, built for authorized offensive security: red team and penetration testing work other models refuse, plus deep code review and research.
Quick and low-cost, for engagement triage, security research and code review at volume, and for everyday work.
For authorized testing and security research, as our Terms of Service require.
The same models and the same balance in the chat and your own tools.
Open Nef 4.0 in the web chat. Paste code, logs or a crash trace, attach files, and keep every conversation in your account.
Open the chatRun npx fullsendai connect: it signs you in and sets up Claude Code, Codex, Cursor, Cline and more with Nef, Sifu and every other model we serve. Or call the OpenAI- and Anthropic-compatible API from your own scripts.
Connect your agentSet four environment variables and run Claude Code from your Kali box or lab VM with Nef 4.0. Any OpenAI-compatible tool works the same way.
export ANTHROPIC_BASE_URL=https://fullsendai.com
export ANTHROPIC_AUTH_TOKEN=fs_live_...
export ANTHROPIC_MODEL=nef-4.0
export ANTHROPIC_DEFAULT_HAIKU_MODEL=nef-4.0
claudeWhat security teams ask before the first engagement.
It is built not to. Nef 4.0 and Sifu 3.5 are tuned for authorized offensive security, so in-scope red team and pentest requests get done instead of declined. Mainstream models are fine for defensive questions; this is where they stop.
Yes, within your authorization. Use FullsendAI for systems you own or have written permission to test, as our Terms of Service require. Keep client data in projects with body storage off, which is the default.
Nef 4.0 for hard analysis and multi-file code review; Sifu 3.5 when you want quick, low-cost answers at volume. The same key opens every other model we serve too, so you can switch from your tool's model picker. Everything bills per token from one balance.
API request and response bodies aren’t stored unless you turn on body storage for a project, and then only for 30 days. Web chat conversations are saved to your account until you delete them. Nothing is used for training.
Every model on FullsendAI, Nef 4.0 and Sifu 3.5 included, is hosted in the United States.
Get a key in a minute and pay only for the tokens you use. Every model hosted in the USA.
Get API Key