# AI0Day

> AI0Day is Innora's agentic-native, offensive-grade security LLM for authorized security research. It answers the authorized offensive-security work that mainstream models decline — APT analysis, reverse engineering, vulnerability triage, and Web3 audits — and runs inside existing Claude Code / OpenAI / Anthropic Messages API workflows. All access is vetted and fully auditable.

## What it is
- A domain-tuned security model on an unrestricted base, served on a four-cluster B300 dense-inference fleet, up to 1M-token context, with native tool use and token-by-token streaming.
- The base's standard refusal layer is disabled; domain training keeps it refusing genuine abuse (high adversarial-refusal rate) while answering authorized security tasks.
- Endpoints: `https://api.ai0day.com` (Anthropic Messages API compatible) and `https://api.ai0day.com/v1` (OpenAI Chat Completions compatible).

## Security modes (LLM, self-serve)
- `web3_audit` — Solidity/smart-contract audits: reentrancy, access control, oracle manipulation, flash-loan vectors, proxy upgrade safety, MEV surface.
- `vuln_triage` — CVE analysis, CVSS breakdown, exploit conditions, PoC outlines, memory-corruption review, cross-file dataflow.
- `apt_detection` — threat-actor TTPs, MITRE ATT&CK mapping, kill-chain reconstruction, C2 patterns, detection-rule drafting.
- `reverse_analysis` — disassembly walkthroughs, packing identification, anti-debug bypasses, malware attribution.

## Reverse-engineering tools via MCP (self-serve)
- Model Context Protocol server at `https://mcp.ai0day.com/mcp` (Streamable HTTP transport).
- Add to Claude Code: `claude mcp add --transport http ai0day https://mcp.ai0day.com/mcp --header "Authorization: Bearer sk-ai0day-…"`
- Access is key-gated: every tool call requires a valid AI0Day key; unauthenticated or invalid-key calls are rejected. The MCP drives AI0Day's own model and analyzers only — it is not a proxy to any third-party LLM API.
- Offensive RE modes (model-driven): ai0day_reverse_binary, ai0day_reverse_apk, ai0day_audit_code, ai0day_triage_vuln, ai0day_triage_crash, ai0day_apt, ai0day_web3.
- Deterministic static analyzers (same input, same output, auditable): ai0day_apk_certificate (signature schemes v1–v4), ai0day_apk_secrets (hardcoded keys/tokens/URLs), ai0day_binary_protocol (protocol reverse-engineering), ai0day_account_security (auth/credential weaknesses), ai0day_api_tracer (endpoints and call chains).
- Server-side exploit verification: ai0day_exploit_verify_server — angr symbolic execution on a physically isolated worker; the target binary is analyzed server-side and never leaves it.
- Usage: ai0day_get_usage — the calling key's own quota and usage.

## Full Innora-Sentinel platform (enterprise, contact sales)
- The self-serve MCP exposes AI0Day's core deterministic analyzers.
- The complete Innora-Sentinel engine — dynamic instrumentation (Frida), fuzzing, malware-family pipelines, protocol reverse-engineering, and multi-engine (Ghidra / MobSF) fusion — is available under enterprise engagement on the Innora platform.
- Enterprise / full-capability inquiries: https://innora.ai

## Access & pricing
- Invite-only, vetted (KYC). Email access@ai0day.com with identity and use case; response within 24 hours.
- Trial: free, time- and volume-limited evaluation.
- Monthly: $10,000 per key per month; no monthly request cap (per-key concurrency governed); full 1M context; 90-day audit trail.
- Dedicated cluster: from $50,000 per month for higher concurrency, custom corpus, or private deployment.

## Trust & governance
- Inference on own infrastructure, no third-party model round-trip.
- Full audit trail retained 90 days under access control, then purged.
- Dual-use tool; bound by an authorized-use agreement; unlawful or unauthorized use is prohibited and grounds for revocation.

## Links
- Home: https://www.ai0day.com
- Health: https://api.ai0day.com/health
- Parent platform (Innora-Sentinel, enterprise): https://innora.ai
- Access: access@ai0day.com