Developer library

Deploy Agent DNS in your stack

Guides per agent runtime and environment, the management API, and webhooks.

No SDK in the agent. Set the resolver, handle a held destination in code if you choose to, and drive scope through the API.

Deployment standard Hermes Agent
Start here

Guides for the agents your teams run

Hermes Agent, OpenClaw, OpenHands, and the coding agents Claude Code, Codex CLI and Gemini CLI. Each guide follows the deployment standard: requirements, numbered steps with one command each, verification, and the limits stated plainly.

Read the deployment standard

Hermes Agent

One policy for the gateway, one for the terminal backend. Closes the DNS gap in the vendor egress guide.

OpenClaw

Gateway on the host, sandboxes in Docker. Opening the sandbox network opens a policy, not the internet.

OpenHands

App container and runtime containers on separate policies.

Claude Code, Codex CLI and Gemini CLI

Apply the policy to the environment the agent runs in: devcontainer, CI runner, managed workstation.

Integration levels

Four levels of integration

LEVEL 0

Assign the resolver

A VPC DHCP option, a CoreDNS forward, a dnsConfig on the pod, or a DoH URL in the runtime. Configuration only; no code changes.

Platform, DevOps

LEVEL 1

Handle a held destination in code

Held and blocked names resolve to the policy's block page address. A short guard lets tool code distinguish a held destination from an unavailable service and escalate instead of retrying.

Agent developers

LEVEL 2

Manage scope through the API

Create policies and sites, approve held names, push IOCs, publish and roll back. Scoped API keys (13 scopes) for automation; a JWT for the endpoints that have no scope.

Platform, MLOps

LEVEL 3

Consume events

Six signed webhook events. Verify the signature, then open a case, page on-call staff, or approve from a chat integration.

Security engineering

Held destinations as seen by application code

A destination the policy has not previously resolved, or one it blocks, answers with the policy's block page address. Application code receives a successful lookup to a Securd address rather than a timeout. Each Level 1 sample resolves the hostname, compares the answer, and raises a typed exception so the agent escalates instead of retrying.

Every configuration in this library assumes one policy per agent role and one identity per agent instance. A compartment (a company in Securd) holds policies; each policy maintains its own first-seen baseline and lists; identities and static sites bind to a policy and share its baseline.

The block page address is displayed in the console on the site. Set it in SECURD_BLOCK_ADDRS for the samples below.

pod.yaml 
# Level 0 for a Kubernetes pod
spec:
  dnsPolicy: "None"
  dnsConfig:
    nameservers: ["<your compartment resolver IPs>"]
Agent runtimes (Levels 0 and 1)

Deployment guides for the agents you install.

Hermes Agent

One policy for the gateway, one for the terminal backend. Closes the DNS gap in the vendor egress guide.

OpenClaw

Gateway on the host, sandboxes in Docker. Opening the sandbox network opens a policy, not the internet.

OpenHands

App container and runtime containers on separate policies.

Claude Code, Codex CLI and Gemini CLI

Apply the policy to the environment the agent runs in: devcontainer, CI runner, managed workstation.

Environments (Level 0)

Assign the resolver. No code.

Kubernetes

A policy per namespace with restricted DNS egress.

Docker and Compose

Per-container resolver configuration.

AWS

VPC DHCP options and a security group that restricts DNS egress.

Google Cloud

Cloud DNS forwarding zone and a firewall rule.

Azure

VNet custom DNS and an NSG rule.

Linux VMs and bare metal

systemd-resolved, NetworkManager, or resolv.conf.

DoH virtual sites

Policy assignment without a static IP, for workstations, serverless and roaming runtimes.

CI runners

Resolver assignment for coding agents in pipelines.

MCP server hosts

A separate policy for the host that serves tools.

Agent frameworks (Level 1)

Catch a held destination in code. Ten to thirty lines.

Python: LangChain and LangGraph

Guard tool calls and handle held destinations without retry loops.

Python: CrewAI

One guard shared by every crew tool.

Python: OpenAI Agents SDK

Function tools that respect the policy.

Python: Anthropic SDK and tool use

Tool results that report a hold instead of a fake failure.

TypeScript: LangChain.js and LangGraph.js

A guarded DynamicTool for Node runtimes.

TypeScript: Vercel AI SDK

Guarded tools with the tool() helper.

Go

A Securd-aware dialer.

Java: Spring AI

A guard before any RestClient call.

.NET: Semantic Kernel

Kernel functions that check the policy first.

Rust

Resolve with the system resolver, refuse block page answers.

API and events (Levels 2 and 3)

Drive scope from code. React to events.

Management API

Create policies and sites, approve held names, push IOCs, publish and roll back.

Webhooks

Six signed events with receiver examples.

Scope templates

Allow list templates for common model providers, tool hosts and MCP servers, applied before enforcement begins.

Event reference

Query log fields, the CEF mapping, and a sample payload for each webhook event.

Integrations

Splunk, Sentinel, Elastic, Datadog, Google SecOps, QRadar, Sumo, Cribl, OTel, SOAR, Slack, PagerDuty.

Evaluate Agent DNS with your own agent traffic

Deploy on a single policy in learning mode and review the recorded destinations with your team.

See It Live View Pricing