Uncensored AI for cybersecurity
4 Red Team and security-tuned builds (Red Team (Pro), Red Team (Flash), GLM-5.3 Flash Cybersecurity and CyberHeabsy) and 5 uncensored models that answer the security question instead of refusing it, on one OpenAI-compatible key. The same key runs 460 more chat models, Claude and GPT included. Prepaid from $10, card or crypto.
Security models
USD per million tokens, read live. A triage turn is 6,000 input tokens (a log excerpt or a finding) and 800 output tokens.
| Model | Type | Context | Input | Cached input | Output | 1,000 triage turns |
|---|---|---|---|---|---|---|
| Red Team (Pro)uplink-red-team-pro | Red Team | — | $1.50 | $0.425 | $6.00 | $13.80 |
| Red Team (Flash)uplink-red-team-flash | Red Team | — | $0.90 | $0.1543 | $3.00 | $7.7999 |
| GLM-5.3 Flash Cybersecurityglm-5.3-flash-cybersecurity | Security tune | 1M | $0.525 | $0.2625 | $1.75 | $4.55 |
| CyberHeabsycyberheabsy | Security tune | 1M | $5.25 | $0.9625 | $15.75 | $44.10 |
| GLM-5.3 Uncensoredglm-5.3-uncensored | Uncensored | — | $0.282 | $0.0799 | $0.846 | $2.3688 |
| DeepSeek V4.1 Flash Abliterateddeepseek-v4.1-flash-abliterated | Uncensored | — | $0.8488 | $0.1455 | $1.6975 | $6.4505 |
| GLM-5.3 Abliteratedglm-5.3-abliterated | Uncensored | — | $2.25 | $0.66 | $4.50 | $17.10 |
| DeepSeek V4 Pro Uncensoreddeepseek-v4-pro-uncensored | Uncensored | — | $1.00 | $0.20 | $2.00 | $7.60 |
| Kimi K3 Uncensoredkimi-k3-uncensored | Uncensored | 262K | $5.745 | $1.437 | $28.68 | $57.414 |
Cached input applies to a repeated prefix, such as a system prompt or a document sent on every turn.
Connect
The chat URL is https://api.relayswitch.io/v1/chat/completions. Any OpenAI SDK takes the base URL and the key.
curl https://api.relayswitch.io/v1/chat/completions \ -H "Authorization: Bearer $RELAYSWITCH_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "uplink-red-team-pro", "messages": [{"role": "user", "content": "Write a Sigma rule that detects CVE-2024-3094 exploitation attempts and explain the detection logic."}] }'from openai import OpenAI client = OpenAI(base_url="https://api.relayswitch.io/v1", api_key="$RELAYSWITCH_KEY")reply = client.chat.completions.create( model="uplink-red-team-pro", messages=[{"role": "user", "content": "Write a Sigma rule that detects CVE-2024-3094 exploitation attempts and explain the detection logic."}],)print(reply.choices[0].message.content)Use it from Strix, HackingBuddyGPT, CAI and PentestGPT
Each agent takes an OpenAI-compatible endpoint. The variable names are the ones each project documents; the model id is any row of the table.
# .env for Strix (README: STRIX_LLM, LLM_API_KEY, LLM_API_BASE)STRIX_LLM="openai/uplink-red-team-pro"LLM_API_KEY="$RELAYSWITCH_KEY"LLM_API_BASE="https://api.relayswitch.io/v1"# HackingBuddyGPT (README: --llm.api_base, --llm.model in litellm format, --llm.api_key)python src/hackingBuddyGPT/cli/wintermute.py LinuxPrivesc \ --llm.api_base="https://api.relayswitch.io/v1" \ --llm.api_key="$RELAYSWITCH_KEY" \ --llm.model="openai/uplink-red-team-pro" \ --llm.context_size=100000# CAI (README: CAI_MODEL, OPENAI_API_KEY; docs/models.md: OPENAI_BASE_URL)CAI_MODEL="openai/uplink-red-team-pro"OPENAI_API_KEY="$RELAYSWITCH_KEY"OPENAI_BASE_URL="https://api.relayswitch.io/v1"# PentestGPT's autonomous mode runs on Claude Code. Point Claude Code here and it bills Claude on this key.export ANTHROPIC_BASE_URL="https://api.relayswitch.io/v1"export ANTHROPIC_AUTH_TOKEN="$RELAYSWITCH_KEY"- Strix and HackingBuddyGPT route through litellm, so the model id carries the
openai/prefix. - CAI documents
CAI_MODELandOPENAI_API_KEYin its README; the base URL variable is in its docs/models.md. - Claude Code speaks Anthropic's format, so under it use a Claude id; the Claude Code guide has the full setup.
Questions
Models, tools and what the setup does.
One key for the security builds and everything else
Prepaid from $10. Card or crypto, no KYC.