SECURE AGENTICS LTD

What SECURE AGENTICS LTD does

Secure Agentics Ltd is a London-based private limited company founded in 2024 by Max Corbridge (CEO) and Steve Street (COO) to address security risks arising from autonomous AI agents. Its product, Adrian, is an open-source runtime security monitoring and control toolkit that intercepts both an agent's observable activity (tool calls, inputs/outputs) and its internal chain-of-thought reasoning traces to detect prompt injection, jailbreaks, tool poisoning, unsafe tool calls, data exfiltration, credential leakage, privilege escalation, and out-of-remit actions before execution. Adrian comprises an Apache 2.0 Python SDK that auto-instruments LangChain/LangGraph via standard callback interfaces, a Go backend server with a WebSocket ingestion API and a Gemma 4–based reasoning classifier served via Llama.cpp, and a Next.js dashboard. Three operating modes — Audit (alert-only), Human Review (pause for human approval), and Block (auto-block) — let customers choose the intervention level.

The business runs a pure product-led growth motion. Adrian is offered free at launch via a freemium hosted backend on AWS (with token-based fair-use limits of 100,000/hour, 500,000/day, and 2,500,000/month per account), alongside a fully self-hosted Docker Compose deployment for data-sovereign and air-gapped environments. Distribution is self-serve: customers sign up at app.adrian.secureagentics.ai, generate an API key, and install the Python SDK. There is no sales team; acquisition is driven by the open-source GitHub repository (373 stars, 77 forks), Discord, YouTube, LinkedIn, Product Hunt, and a thought-leadership blog authored by the CEO. The Terms of Service reserve the right to introduce paid tiers in the future. There are no disclosed funding rounds, paying customers, or revenue figures.

Secure Agentics serves teams building and deploying autonomous AI agents, multi-step agent swarms, and production AI systems with real-world impact, with secondary focus on regulated sectors (financial services, healthcare, legal), edge AI, and Critical National Infrastructure. The hosted backend operates in three AWS regions (London, Dublin, US East/West) with UK–EU adequacy and EU–US Data Privacy Framework transfer mechanisms, and the company is registered with the UK ICO (data controller ZC118237) and has appointed an EU GDPR Article 27 representative via Prighter Group. Adrian's design aligns with the AARM (Autonomous Action Runtime Management) open specification. The company is at an early commercial stage, with the product publicly launched in May 2026 and active developer community engagement but no monetisation in market yet.

Name

SECURE AGENTICS LTD

Legal name

Secure Agentics Ltd

Website

https://secureagentics.ai

Company type

Private

Founded year

2024

Operating status

Operating

Short description

Secure Agentics Ltd, founded in London in 2024, builds Adrian, an open-source runtime security toolkit that monitors and blocks malicious behavior in autonomous AI agents by analyzing their chain-of-thought and tool calls. The product targets teams deploying production AI agents.

Ownership category

akta.pro rank

Product category

AI Agent Runtime Security

NAICS

Software Publishers (5132), Computer Systems Design and Related Services (54151)

SIC

Services-Prepackaged Software (7372), Services-Computer Programming Services (7371)

akta.pro primary industry

Secure Model Deployment & Runtime Protection (sandboxing, isolation) (HDAAAKAH)

akta.pro secondary industries

Responsible AI, Security & Privacy Platforms (Safety, Guardrails, PII) (HDAEANAG), Identity, Access & Secrets Management for AI Systems (IAM for agents/models) (HDAAAKAJ)

Keywords

• AI agent security
• Runtime monitoring
• Prompt injection detection
• Agentic AI governance
• Open source security toolkit

Headquarters

HQ city

London

HQ country

United Kingdom

HQ region

Europe

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Infrastructure, Personnel, Marketing or Sales, Operations

Revenue model

1. Freemium hosted backend: At launch, Adrian's hosted backend is provided free of charge with no fees, subscriptions, or charges. The service is offered with token-based usage limits: 100,000 tokens per hour, 500,000 per day, and 2,500,000 per calendar month per account, measured against output tokens of backend models. The company may introduce paid tiers in the future with reasonable notice.

Pricing tiers

Go-to-market motion1 record

Distribution channels4 records

Marketing channels7 records

Core offering

Secure Agentics sells Adrian, an open-source runtime security monitoring and control toolkit for AI agents. Adrian installs inside agents (LangChain/LangGraph), captures paired LLM and tool-call events via a Python SDK over WebSocket, and uses a reasoning-based world-model classifier (Gemma 4) to evaluate both the agent's chain of thought and its activity against a configured remit. It can operate in Audit, Human Review, or Block mode, allowing pre-execution intervention on prompt injection, tool misuse, data exfiltration and out-of-remit actions, delivered via a hosted AWS backend or a self-hosted Docker Compose stack.

Product overview

Secure Agentics Ltd offers Adrian, a unified open-source runtime security monitoring and control toolkit for AI agents. The product consists of three components: the Adrian Python SDK (auto-instrumenting LangChain/LangGraph, Apache 2.0 open-source), the Adrian Hosted Backend (managed GPU-accelerated service on AWS, free with fair-use limits), and the Adrian Dashboard/Control Plane (web-based for configuration, event feeds, and human-in-the-loop approvals). A self-hosted deployment option runs the full stack (Go backend, Next.js dashboard, Llama.cpp with Gemma 4) locally under Docker Compose for data sovereignty. Three operating modes are available: Audit (alert-only), Human Review (human approval required), and Block (automatic pre-execution blocking). The system detects threats by analysing both agent activity logs and chain-of-thought reasoning traces, aligning with the AARM open specification for agentic runtime security.

Differentiator

Problem solved

Functional benefit

Brands

• Adrian: Adrian is the open-source runtime security monitoring and control toolkit for AI agents, providing real-time detection and blocking of malicious behaviour, prompt injection, and policy drift.

Products and services

• Adrian (Core Product) Open-source runtime security monitoring and control engine for AI agents. Analyses agent activity logs and reasoning traces to detect malicious, misaligned, or out-of-remit behaviour and optionally intervene in-flight. Aligns with the AARM open specification. Three operating modes: Audit (alert-only), Human Review (pause for human approval), and Block (auto-block without human oversight).
• Adrian Python SDK Python package (pip install adrian-sdk) that auto-instruments LangChain and LangGraph agent runtimes via standard callback interfaces, captures PairedEvents (LLM call pairs and tool execution pairs), and ships them to the Adrian backend over WebSocket. Apache 2.0 licensed. Requires Python 3.12+. Supports custom callbacks (on_event, on_verdict, on_audit, on_block, on_mcp_server, on_disconnect, on_reconnect).
• Adrian Hosted Backend Managed backend service operated by Secure Agentics on AWS infrastructure with UK (London), EU (Dublin), and US (Virginia/Oregon) regions. Ingests telemetry from the SDK, classifies events via GPU-accelerated world-model classifier, and returns verdicts. WebSocket endpoint at wss://adrian.secureagentics.ai/ws. Latency ~100-600 ms per event on L4/L40S/H100 GPUs. Provided free with token-based fair-use limits; metadata may be retained for model improvement unless opted out.
• Adrian Self-Hosted Backend Fully offline, data-sovereign deployment via Docker Compose. Runs the Go backend server, Next.js dashboard, and Llama.cpp container serving a local Gemma 4 (E2B or E4B) classifier model on the customer's own infrastructure. No managed cloud; no telemetry leaves the host. Includes BYOK encryption for enterprise customers. Requires Docker, Docker Compose v2, and an NVIDIA GPU with CUDA.
• Adrian Dashboard (Control Plane) Web-based control plane for account management, agent configuration, policy settings, event feed, verdicts, and alert review. Human Review mode allows human operators to approve or reject paused agent actions. Alert thresholds and notification preferences are configurable per deployment. Available as a hosted instance or as a Next.js bundle in the self-hosted deployment.

Quantifiable outcome

• ~35% increase in detection accuracy vs behaviour-only monitoring, based on OpenAI and DeepMind research cited by the company
• +3 more outcomes

Segments6 records

Ideal customer profiles2 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration11 records

AI capability4 records

Feature7 records

Partnerships

Three partnerships are on record, tiered core.

• Amazon Web Services EMEA SARLcoreTechnology or IntegrationAWS EMEA provides cloud compute, storage and networking for UK and EU customers of the Adrian hosted backend. Data resides in London (eu-west-2) and Dublin (eu-west-1). Transfer mechanism: UK–EU adequacy.
• Amazon Web Services, Inc.coreTechnology or IntegrationAWS US provides cloud compute, storage and networking for US-region customers of the Adrian hosted backend. Regions: us-east-1 / us-west-2. Transfer mechanism: EU-US Data Privacy Framework + UK Extension; UK IDTA + EU SCCs Module 2/3 in fallback.
• Prighter GroupcoreImplementation/ SI/ Consulting PartnerPrighter Group serves as Secure Agentics' Article 27 EU Representative under EU GDPR, acting as the contact point for EU data subjects and EU supervisory authorities. EU Representative: Andreas Maetzler, c/o iuro Rechtsanwälte GmbH, Vienna, Austria.

Scale indicators2 records

Recent moves6 records

Expansion highlights3 records

Direct peers

• Lakera: Lakera builds prompt-injection and AI-application security tooling (Gandalf, Guard). It is a direct peer addressing the same problem of runtime protection for LLM-powered applications and AI agents, with overlapping customer targets in enterprise AI deployments.
• Robust Intelligence: Acquired by Cisco, Robust Intelligence built an AI firewall and continuous validation platform for AI applications. Direct competitor in the AI-runtime-protection category, now backed by Cisco's enterprise distribution.
• Protect AI: Protect AI offers a platform for AI/ML security including model scanning, vulnerability management, and runtime guardrails for LLM applications. Direct overlap with Adrian on AI-system runtime protection and governance tooling.
• Mindgard: Mindgard is an AI security and red-teaming platform founded by Secure Agentics' co-founder Steve Street. It directly addresses AI model and application vulnerability assessment, sharing customer overlap and category framing with Adrian.
• HiddenLayer: HiddenLayer provides AI security posture management and runtime threat detection for AI models and applications. Targets the same enterprise buyer concerned with adversarial attacks, model misuse, and AI supply chain security as Adrian.

Emerging players

• Galileo: Galileo provides LLM observability and evaluation including guardrails and hallucination/misuse detection for production AI applications. Adjacent peer that overlaps on monitoring production agent behaviour, though more focused on quality than security.
• Rebuff: Rebuff is an open-source prompt-injection detection library with multiple detection heuristics. It overlaps Adrian on prompt-injection defence but focuses narrowly on detection heuristics rather than full agent-runtime reasoning analysis.

Broad incumbents

• Microsoft Azure AI Content Safety: Microsoft Azure AI Content Safety provides prompt-injection and jailbreak detection as a managed cloud service. As a broad incumbent, it competes on integration convenience for Azure customers rather than on reasoning-based agent monitoring depth.
• Palo Alto Networks (AI Runtime Security): Palo Alto Networks has expanded into AI security via acquisitions and product extensions targeting LLM and agent runtime protection. As a broad incumbent, it bundles AI security into larger enterprise security platforms rather than specialising in agent-native reasoning.
• Cloudflare AI Gateway / Workers AI: Cloudflare's AI Gateway includes prompt-injection detection and LLM traffic inspection at the network edge. Competes with Adrian's edge-deployment thesis from an infrastructure-platform position with global distribution.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat5 records

Key risks2 records

Key highlights6 records

Customer concentration

• GitHubgithub.com/secureagentics
• YouTubeyoutube.com/@secureagenticsai
• LinkedInlinkedin.com/company/secure-agentics

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles2 records

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about SECURE AGENTICS LTD