Aurelian
Secure the code. Guard the agent. Prove the chain.
Aurelian finds the vulnerabilities in the AI you are shipping, attacks your own agent before somebody else does, and signs a cryptographic record of everything it did. The support-intelligence desk it grew out of ships alongside as a second module.
Support Intelligence, included
The helpdesk Aurelian started as, kept as a full module. This is the shipped Command Center: live KPIs, SLA breach alerts, and pattern detection across every connected app.
Built for teams shipping AI
Not a scanner bolted onto a chatbot. Aurelian fingerprints every LLM call site in your code, attacks your agent the way an adversary would, and signs a record you can hand to an auditor.
LLM Vulnerability Scanner
Static analysis built for AI code rather than retrofitted onto it. Every LLM call site in your repository is checked against the OWASP LLM Top 10, and results export as SARIF for one-click upload into GitHub Advanced Security.
128-bit Smart Tags
Each LLM call site carries a fingerprint that survives refactoring, so Aurelian can tell you the moment a function you already reviewed and cleared has quietly become vulnerable.
Adversarial Pen Tester
Prompt injection, jailbreak fuzzing, system-prompt extraction and agent hijacking, run against your own system. The attacks are generated for your specific target, and a second model judges which ones actually landed.
Agent Runtime Monitor
Hooks into your running agents through TypeScript and Python SDKs, including LangChain, CrewAI and AutoGen, and flags behaviour that should not be happening: unexpected tool calls, runaway loops, data heading somewhere new.
Signed Execution Chain
Every action an agent takes is recorded and signed with Ed25519 using rotating keys, producing a tamper-evident chain of what your AI did and when - the kind of evidence EU AI Act Article 12 record-keeping asks for.
Support Intelligence
The helpdesk Aurelian grew out of, kept as a full module: tickets triaged the moment they arrive, spike and repeat-issue detection across every connected app, SLA countdowns, and a knowledge base that writes itself as issues get resolved.
Scan. Attack. Watch. Prove.
How a release goes from unreviewed to defensible.
Scan
Point Aurelian at your repository. Every LLM call site is fingerprinted and checked against the OWASP LLM Top 10, with findings exported as SARIF.
Attack
Adversarial payloads are generated for your specific target - injection, jailbreaks, extraction, hijacking - and a second model grades which ones landed.
Watch
Runtime hooks follow your agents in production and flag unexpected tool calls, runaway loops, and data heading somewhere it has never gone before.
Prove
Every step is signed into a tamper-evident chain, so what your AI did is a matter of record rather than a matter of trust.
Get early access
If you are shipping LLM features or autonomous agents, we want you in the first group that uses this.
Phase 1, the code scanner, is in development. Tell us what you are building and we will bring you in as each module lands.