Map / Runtime & Guardrails / PromptGuard

PromptGuard

UK · Founded 2025

Quick answer: PromptGuard is an AI security company based in the UK and founded in 2025, working in Runtime & Guardrails, MCP & LLM Gateways, AI Browser Security and 1 more category.

Runtime & GuardrailsMCP & LLM GatewaysAI Browser SecurityAI Red Teaming

Capabilities

LLM firewall and guardrail API that scans prompts and responses for prompt injection (with normalization of homoglyphs, zero-width characters and encoded payloads), jailbreaks (LLM classifier, including multi-turn escalation), system-prompt extraction and data exfiltration, 46 PII entity types, secrets (40+ provider prefixes plus entropy), malicious URLs/IPs, phishing, toxicity and code injection, plus custom plain-English rules judged by an LLM and OCR of images. Outputs get the same scans, with mid-stream cutoff and RAG hallucination checks against retrieved context. Agent endpoints validate tool calls, ingest agent traces, scan agent memory and scan MCP manifests for tool poisoning, with MCP server allow/block lists and argument schema validation. A security-testing API runs built-in and custom attack tests. A separate Shadow AI line covers employee use of external AI tools.

Sensors / integration

Managed proxy (swap the provider base URL), standalone Guard API, and Python/Node SDKs that auto-instrument provider clients; integrations for LiteLLM, LangChain, LangGraph, LlamaIndex, CrewAI, FastAPI and the Vercel AI SDK across OpenAI, Anthropic, Gemini, Azure, Bedrock, Mistral, Cohere, Groq, Hugging Face, Ollama and vLLM. Self-hosted via Helm or docker-compose, including air-gapped. Shadow AI uses a browser extension and a macOS/Windows desktop agent enrolled via MDM.

Protections

Blocks or flags injected and jailbreak prompts, redacts, masks or blocks PII and secrets before they reach the model, terminates unsafe streaming responses, rejects disallowed or malformed agent tool calls and untrusted MCP servers, and stops employees pasting secrets or PII into outside AI tools.

Differentiator

Developer-first, self-serve guardrail API with published usage pricing (free tier; Developer $20/month; Team $24/seat; Enterprise from $50/seat with SSO and self-hosting), combining app-side guardrails, agent/MCP checks and shadow-AI endpoint controls. Very early-stage: launched late 2025 and built largely by founder Abhijoy Sarkar; UK entity PG Tech Ltd incorporated in London in 2026. No outside funding disclosed. Not to be confused with Meta's Prompt Guard model.

Team & funding

Founders
Abhijoy Sarkar

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