ThirdKey builds the trust layer for autonomous AI agents — policy enforcement, cryptographic identity, tool governance, and network visibility.
Symbiont — Policy-Governed Agent Runtime
AI agents are easy to demo and hard to trust. Symbiont is the Rust-native execution layer that separates agent intent from execution authority.
- Cedar-based fine-grained policy authorization
- Typestate-enforced ORGA reasoning loop (Observe → Reason → Gate → Act)
- MCP tool integration with SchemaPin cryptographic verification
- Docker sandboxing with resource limits and approval gates
- Tamper-evident cryptographic audit trails
- Secrets management via Vault/OpenBao, persistent memory, and RAG
👉 symbiont.dev | Source
ToolClad — Declarative Tool Interface Contracts
Stop writing repetitive custom code for every tool. ToolClad defines typed, validated, policy-aware tool contracts in
.clad.tomlmanifests.
- Three execution modes: oneshot (CLI), session (interactive PTY with Cedar gating), browser (governed headless via CDP/Playwright)
- Shell injection prevention, direct
execvedispatch, process group isolation - 14 built-in type validators, conditional evaluation, evidence envelope generation
- Reference implementations in Rust, Python, JavaScript, and Go
AgentSniff — AI Agent Network Scanner
Detect AI agents operating on your network through passive monitoring, active probing, protocol detection, and behavioral analysis.
- Seven detection techniques: passive DNS analysis (40+ LLM API domains), TCP port scanning, AgentPin identity discovery, MCP server probing, HTTP endpoint signatures, JA3 TLS fingerprinting, behavioral traffic patterns
- Deploy standalone, via Docker, or Docker Compose with web dashboard
- Continuous scanning, webhook/SMTP alerting, SQLite history
| Project | Description |
|---|---|
| 🔐 SchemaPin | Cryptographic protocol for signing AI tool schemas and policies |
| 🪪 AgentPin | Domain-anchored cryptographic identity for AI agents |
| Project | Description |
|---|---|
| 🕶️ AgentNull | Reference implementation of a restricted LLM agent for security testing |
| 📦 VectorSmuggle | Covert data exfiltration via vector embeddings (research prototype) |
- 🌐 Website: thirdkey.ai
- 🔬 Research: research.thirdkey.ai
- 🧠 Symbiont: symbiont.dev
ThirdKey.ai — Infrastructure for AI you can trust.