Identify & classify
The API key says who is calling and for which workload. The request is scanned for PHI, PII, card data, secrets, CUI and confidential material, using patterns, checksums and clinical context.
AI governance gateway
agentKeyRoute sits between your AI tools and the models they use. Every request is identified, checked against your policy and compliance frameworks, redacted, routed to an allowed model, metered and written to a tamper-evident audit log.
Works with Claude Code, Codex, Cursor, the OpenAI and Anthropic SDKs, LangChain and your own apps.
Your tools
Claude CodeCodexCursorSDKs & LangChainInternal appsagentKeyRoute gateway
Approved models
OpenAIAnthropicAzure OpenAIOpenRouterOllama · vLLM on-premPoint a tool at the gateway with a key from your console. The key carries the team, workload and governance profile, so the right rules apply without any change to the tool itself.
# Claude Code through agentKeyRoute
export ANTHROPIC_BASE_URL=https://gateway.agentkeyroute.com
export ANTHROPIC_AUTH_TOKEN=cvn_live_…
claude
How it works
The API key says who is calling and for which workload. The request is scanned for PHI, PII, card data, secrets, CUI and confidential material, using patterns, checksums and clinical context.
Sensitive values are replaced before they reach a model. Ordered rules with default deny, risk scoring and optional guardrail models then decide: allow, hold or block.
Compliance frameworks become routing constraints, so data only goes to models allowed to see it. The model that answered is checked, and every request is metered and hash-chain audited.
Product
Detection and redaction of regulated data, plus a prompt-injection guard and output scanning. Credentials and API keys are always blocked.
Task-aware routing with fallback across your providers. When frameworks are combined, the strictest rule wins, and a request fails closed if the wrong model answers.
Screen requests and responses with Llama Guard, ShieldGemma or Granite Guardian running on your own hardware. You choose whether to fail closed or fail open.
A full decision trace for every request, a hash-chained audit log with an integrity check, and evidence pack export for your auditors.
Every request metered by team, workload and person. Workload budgets, plan limits, chargeback, and import of Claude usage from outside the gateway.
A risk register with key risk indicators, automatic incidents for high-severity blocks, and kill switches to stop a workload at once.
For everyone, not only developers
A chat window for everyone in the organisation. Every message follows the chat profile your administrators choose and is recorded under the person's name.
A coding agent in a private, hardened sandbox for each person. It can only use your governed models, with a key issued to that person.
Checks what people type into ChatGPT, Claude, Gemini, Copilot and Perplexity before it is sent. Your profile decides whether to allow it, redact it or block it.
Roles, teams, two-step sign-in, two-administrator approval for sensitive changes, and SIEM or webhook delivery of events.
Compliance
19 frameworks come with automated control checks, attestation and evidence export. 43 ready-made governance profiles cover healthcare, financial services, government, education, privacy and general use, and you can edit any of them.
Each API key carries a profile, and an effective-policy view shows exactly what that key can reach, and why.
Deployment
Hosted by us. Connect your providers and start governing within minutes of approval.
An isolated instance for your organisation alone, on the Enterprise plan.
Run the whole platform inside your own network, next to local models on Ollama, vLLM, TGI or NVIDIA NIM.
Pricing
For a first governed AI use case.
$499 / month
For organisations governing AI across teams.
$2,499 / month
Regulated enterprises, any deployment model.
Custom
Prices in US dollars, excluding tax. Every new organisation is reviewed before it is activated.
Create your organisation, connect a provider and issue your first governed key.
Get started