MAQPNA

AI companies and agent builders

Ship agents your customers' security team will approve.

Your agent works in the demo. Then the enterprise security review asks where it runs, what it can touch, who approves risky actions and where the logs are. MAQPNA answers those questions in your customer's own cloud, without rewriting your agent.

Why enterprise deals stall in security review

Buyers want your agent, on their terms.

  • It must run in their cloudRegulated buyers will not send data to your SaaS. They want the agent in their VPC or data centre.
  • Least privilege per callSecurity teams ask which tools your agent can call, with which arguments, and who can stop it.
  • Prompt injectionAn agent that reads emails, tickets or web pages can be steered by what it reads.
  • Evidence on demandAuditors want a record of every action, not screenshots or mutable application logs.

How MAQPNA solves it

Governance as a runtime, not a rewrite

  • SDKs and adapters

    Python, TypeScript and Go SDKs with adapters for LangGraph, LangChain, the OpenAI Agents SDK, the Claude Agent SDK and CrewAI.

  • Same code, laptop to cluster

    maqpna dev up runs the gateway locally; the same agent runs in a sandbox in your customer's Kubernetes.

  • Taint tracking

    Content from untrusted sources marks the session, and later sensitive tools then need approval or are denied.

  • Token vault

    Users' OAuth tokens stay in the vault, outside the agent. The gateway injects credentials, so the agent never sees upstream keys.

  • Policy your customer controls

    Per tool and per argument, with policy test, replay and eval before rollout.

  • Evidence bundles

    Export verifiable audit records per session or period for your customer's auditors.

Architecture

Your agent, in your customer's cloud

Your agent image runs in a sandbox in the customer's cluster. Their gateway, their policy, their keys. You ship the agent; they keep control.

Customer VPC or data centre · their policy · their keys

Sandboxes

tier-0 (gVisor) or tier-1 (microVM)

  • your-agent (Python SDK)
  • your-agent-worker
  • browser profile

MAQPNA gateway

identity · policy · DLP · budgets · approval · audit

  • Allowedgithub.get_issueread-only-tools
  • Held for approvalgithub.merge_pr after web readtainted session needs approval
  • Deniedvault.read_secretno-secret-exfiltration

Your systems

  • MCPcustomer's tools
  • modelcustomer's model route
  • A2Aother governed agents
Customer-owned ledger · OpenTelemetry traces linked to records
Illustrative names. You write your own policies.

Evidence

Answers for the security questionnaire

Controls and evidence, mapped. Not a certification.

  • OWASP Top 10 for Agentic Applications

    a control mapping for each of ASI01–ASI10, with honest coverage.

  • EU AI Act Art. 12 and 14

    record-keeping and human oversight your customers configure. How the EU AI Act applies to agents.

  • EU AI Act Art. 26

    your customers, as deployers of high-risk systems, must assign competent people to oversight and keep the logs under their control for at least six months, from 2 December 2027 for Annex III uses (Regulation (EU) 2026/1744). Role-based approver groups and a ledger kept in the customer's own storage support both.

  • No vendor dependency

    no phone-home, standard protocols (MCP, A2A, OIDC, OpenTelemetry) and any model.

  • Supply chain

    signed releases and SBOMs your customers can verify.

FAQ

Common questions

What does MAQPNA do for aI companies?

Your agent works in the demo. Then the enterprise security review asks where it runs, what it can touch, who approves risky actions and where the logs are. MAQPNA answers those questions in your customer's own cloud, without rewriting your agent.

What stops aI companies from putting AI agents into production?

It must run in their cloud: Regulated buyers will not send data to your SaaS. They want the agent in their VPC or data centre. Least privilege per call: Security teams ask which tools your agent can call, with which arguments, and who can stop it. Prompt injection: An agent that reads emails, tickets or web pages can be steered by what it reads. Evidence on demand: Auditors want a record of every action, not screenshots or mutable application logs.

Which controls does MAQPNA enforce?

SDKs and adapters: Python, TypeScript and Go SDKs with adapters for LangGraph, LangChain, the OpenAI Agents SDK, the Claude Agent SDK and CrewAI. Same code, laptop to cluster: maqpna dev up runs the gateway locally; the same agent runs in a sandbox in your customer's Kubernetes. Taint tracking: Content from untrusted sources marks the session, and later sensitive tools then need approval or are denied. Token vault: Users' OAuth tokens stay in the vault, outside the agent. The gateway injects credentials, so the agent never sees upstream keys. Policy your customer controls: Per tool and per argument, with policy test, replay and eval before rollout. Evidence bundles: Export verifiable audit records per session or period for your customer's auditors.

Which regulations does it supply evidence for?

OWASP Top 10 for Agentic Applications: a control mapping for each of ASI01–ASI10, with honest coverage. EU AI Act Art. 12 and 14: record-keeping and human oversight your customers configure. How the EU AI Act applies to agents. EU AI Act Art. 26: your customers, as deployers of high-risk systems, must assign competent people to oversight and keep the logs under their control for at least six months, from 2 December 2027 for Annex III uses (Regulation (EU) 2026/1744). Role-based approver groups and a ledger kept in the customer's own storage support both. No vendor dependency: no phone-home, standard protocols (MCP, A2A, OIDC, OpenTelemetry) and any model. Supply chain: signed releases and SBOMs your customers can verify. MAQPNA supplies controls and evidence; it does not certify compliance.

Where does it run?

Your agent image runs in a sandbox in the customer's cluster. Their gateway, their policy, their keys. You ship the agent; they keep control. The same runtime runs on a laptop, in your cloud account, on Kubernetes or air-gapped.

What does EU AI Act Article 26 ask of customers who deploy our agents?

Your customers, as deployers of high-risk systems, must assign competent people to oversight and keep the logs under their control for at least six months, from 2 December 2027 for Annex III uses (Regulation (EU) 2026/1744). Role-based approver groups and a ledger kept in the customer's own storage support both.

Make your agent enterprise-ready.

Policy, approval flow and evidence, walked through with you.

More install options
curl -fsSL https://maqpna.com/install.sh | sh