Map / Runtime & Guardrails / CanaryVaults
CanaryVaults
canaryvaults.com ↗ · LinkedIn ↗ · Open on the map
Runtime & GuardrailsObservability & Governance
Capabilities
Deception-based security for AI apps and agents, split into six modules: CanaryVaults (decoy email identities seeded where stolen data surfaces), CanaryRAG (three invisible, plausible trap facts woven into each document so a leak into someone else's assistant or knowledge base can be proven), CanaryShield (a three-layer prompt-injection firewall: rules, semantic analyzer, sanitizer), CanaryHoneypot (decoy AI endpoints that answer with synthetic data and classify attacker intent as jailbreak, extraction, probing or normal), CanaryAudit (a SHA-256 hash-chained, tamper-evident log of actions) and CanaryAgent (heartbeats, pre-install skill scanning and a live action timeline for OpenClaw agents, plus decoy MCP tool servers added October 2026).
Sensors / integration
SaaS API with an API key: POST any untrusted string (user message, retrieved chunk, tool output) to the Shield inspect endpoint, or swap the base URL for its OpenAI-compatible proxy in front of chat completions. CanaryRAG returns the trap-laced document to store in the RAG corpus and a monitoring API queries external assistants for the planted facts. CanaryAgent is a drop-in middleware file for OpenClaw plus env vars; decoy MCP servers are added beside an agent's real tools in Claude Code, Cursor, VS Code or the OpenAI Agents SDK. No self-hosted option is advertised.
Protections
Prompt-injection detection and sanitization before model execution (block or flag-only), pre-install blocking of poisoned agent skills (injection payloads, backdoors, undeclared network calls, credential harvesting), and high-signal tripwire alerts by email and Telegram when a decoy identity, honeypot endpoint, trap fact or decoy MCP tool is touched. Evidence is hash-sealed and exportable for incident and legal review.
Differentiator
Applies classic canary-token deception to the AI stack: leak proof for RAG content via trap facts that survive paraphrase and re-embedding, honeypot assistants and decoy MCP tools that only a steered or malicious agent would call. Very early-stage, self-serve product with public pricing (Free, Shield $12/mo, Pro $24/mo, Enterprise $120/mo, billed via PayPal); built and run by a single India-based founder, Nagababu Pothuraju, with no formal legal entity published yet. Positions itself against Thinkst Canary and CanaryTokens. No disclosed funding.
Team & funding
- Founders
- Nagababu Pothuraju
Other Runtime & Guardrails companies
AariaSec · ActaClad · ActiveFence · Acuvity · AGAT Software · AgentFox · AI Aramid · AI Prompt Shield · Aiceberg · Aim Security · AIR · Aira Security · See all →
Last updated · Get the “Listed on” badge