Offensive-grade security AI that doesn't refuse your authorized work.

01Capabilities

Each mode invokes a dedicated hybrid retrieval pipeline (dense + sparse over a curated security corpus) before generation, on a domain-tuned security model served on a B200-class dense-inference cluster with up to 1M-token context. Native tool use and token-by-token streaming are live.

Beyond the four security modes, generic and code routes handle general reasoning and code work in the same session. The gateway auto-selects the mode from query content, or you force one with a suffix.

02Agentic-native — built for Claude Code

The overwhelming majority of our traffic is agentic: security teams point Claude Code (or any OpenAI/Anthropic client) at AI0Day and run full tool-loop workflows. Your slash commands, MCP servers, and hooks keep working — only the model and endpoint change. This is the difference from a plain chat API: AI0Day is tuned to drive real agent loops on authorized security work.

# In ~/.zshrc / ~/.bashrc
export ANTHROPIC_BASE_URL="https://api.ai0day.com"
export ANTHROPIC_AUTH_TOKEN="sk-ai0day-…"

# Restart your shell, then verify:
claude --version
claude -p "Detect APT41 lateral movement signatures in a Linux env."

export OPENAI_BASE_URL="https://api.ai0day.com/v1"
export OPENAI_API_KEY="sk-ai0day-…"

from openai import OpenAI
cli = OpenAI()
r = cli.chat.completions.create(
    model="ai0day",   # any non-empty string; gateway picks the model
    messages=[{"role": "user", "content": "Triage CVE-2024-3400."}],
    stream=True,
)

curl https://api.ai0day.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ai0day-…" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ai0day-reverse_analysis",
    "messages": [
      {"role": "user", "content": "Reverse engineer 0x55 0x48 0x89 0xe5 in x86_64 context."}
    ],
    "max_tokens": 1024
  }'

03Trust & Governance

Offensive capability and enterprise governance are not opposites. AI0Day is unrestricted on authorized security work and fully auditable at the same time — so security, legal, and procurement can say yes.

04Why a specialized model

General-purpose assistants over-refuse authorized offensive-security work — malware RE, exploit conditions, red-team tradecraft. AI0Day is domain-tuned to help on authorized tasks while still refusing genuine abuse.

The point is not "no guardrails" — it is the right guardrails for security research: help authorized practitioners do work that mainstream models decline, while declining prompts that are genuinely abusive.

05Getting Access

Access is invite-only and vetted. Both trial and paid keys are issued manually after a short conversation about your identity and use case.

1. Email [email protected] with: your name, organization, intended use case (red team / blue team / research / audit firm / Web3), and an estimate of your monthly request volume.
2. We respond within 24 hours with either a trial key or a paid-plan arrangement.
3. You receive a key of the form sk-ai0day-… by email. Store it securely — keys are SHA-256 hashed at rest and cannot be recovered if lost.
4. Test with the snippet below. If anything is wrong, we revoke, fix, and reissue.

06Pricing

We meter access duration, not tokens — deploy long-context queries (up to 1M) without watching a per-token counter. All access is vetted; billing is arranged by email.

Dedicated cluster — for higher concurrency, custom corpus, or private deployment, we provision a dedicated GPU pool starting at $50,000 / month. Billing is per key; renewal is opt-in each month, no auto-charge. Contact [email protected].

07Use Cases

Concrete workflows our users run in production, all through Claude Code with ANTHROPIC_BASE_URL pointing at AI0Day.

Smart-contract audit before deploy

$ claude
> Audit contracts/Vault.sol and contracts/Strategy.sol.
> Flag every reentrancy path, oracle dependency, and access-control
> bypass. Output a severity-sorted finding list with line refs.

Reverse engineering a malware sample

$ claude
> Disassemble the entry point of ./samples/loader.bin and identify
> the unpacking routine. If you find a custom XOR loop, decode the
> embedded config block and explain the C2 generation algorithm.

CVE triage in a CI pipeline

curl -s https://api.ai0day.com/v1/chat/completions \
  -H "Authorization: Bearer $AI0DAY_KEY" \
  -d "{\"model\":\"ai0day-vuln_triage\",
       \"messages\":[{\"role\":\"user\",\"content\":\"$CVE_DESC\"}],
       \"max_tokens\":1500}" | jq -r '.choices[0].message.content'

08FAQ

What model is behind AI0Day?

A domain-tuned security model built on an unrestricted base, served on our own B200-class dense-inference cluster with FP8 and speculative decoding, up to 1M-token context. We keep the exact base and weights model-agnostic on the public site. The base's standard refusal layer is disabled; domain training keeps it refusing genuine abuse (96.7% adversarial refusal) while answering authorized security work.

How is my data handled?

Inference runs in-memory on our own infrastructure with no third-party model round-trip. For auditability, request and response content is captured and retained for 90 days under access control, then purged. The RAG corpus is read-only public security data. Enterprise plans can negotiate custom retention.

Do you support tool use and streaming?

Yes. Native tool use (function calling) and token-by-token SSE streaming are live on both the Anthropic /v1/messages and OpenAI /v1/chat/completions endpoints. This is what makes agentic Claude Code workflows work end-to-end.

What's the latency?

Agentic tool-loop turns are fast (typically a few seconds). Deep single-shot analysis with large context and long generations takes longer (tens of seconds), dominated by generation time on the GPU, not network.

Can I get higher concurrency?

Per-key concurrency is governed to keep the shared platform responsive. For higher concurrency, custom corpus, or private deployment we provision a dedicated GPU pool starting at $50K/month. Email us.

Is this legal? What are the limits?

AI0Day is for lawful, authorized security research, red/blue-team, and audit work. Access requires accepting an authorized-use agreement, and all activity is auditable and revocable. It is a dual-use tool; unlawful or unauthorized use is prohibited.

What if a key gets compromised?

Email us. We disable it within minutes and reissue. Device fingerprints are tracked per key; anomalous multi-device use is automatically flagged.