“Summarize the ACME Manufacturing pump design notes for my presentation.”
Open-source AI governance
AI SAFETY GATEWAY
Take AI safety into
your own hands.
Don’t wait for someone else to tell you how to control what goes in and out of your corporate network. Put a policy and audit layer between your team and frontier AI.
PROTECT IP BEFORE IT LEAVES
Turn company policy into an active boundary.
SafetyRouter evaluates requests from both human prompters and automated agents before they reach an external LLM. Organizations can identify confidential project names, customers, products, or internal initiatives and decide that those references must never leave the approved network boundary.
EXAMPLE COMPANY POLICY
Treat “ACME Manufacturing” and “Pump Manufacturer LLC” as confidential company identifiers. If either name appears in a human prompt or agent request, classify the request as RED, do not send it to an external LLM, and explain that company policy prevents processing.Your last line of defense at the AI gateway. SafetyRouter can help keep protected information from being submitted for external model processing, retention, possible training, or reuse by others. It complements provider agreements, access controls, and employee training with enforcement at the moment a request is made.
“Compare Pump Manufacturer LLC specifications with our internal prototype.”
BLOCKED BEFORE THE LLM
Policy match recorded in the tenant audit trailSafetyRouter blocked this request because it contains information protected by company policy. The request was not sent to the configured external model. Contact your administrator if you believe this decision should be reviewed.
INSIDE YOUR TENANT
See what your managers get.
Every tenant includes a focused control center for monitoring traffic, investigating decisions, managing identities, and tuning company policy.
01
Overview
Monitor live traffic, tokens, spend, and API-key activity.
Live trafficUpdated now
CLIENT2 active→ROUTER → LLM1 active→RELEASED4 recent
REQUESTS148TOKENS82.4KCOST$0.42
02
Requests
Search every capture by theme, identity, model, or decision.
Search theme, ID, key, or prompt
THEMEKEYSTATUS
Pump design summaryEngineeringBLOCKED
OAuth implementationPlatformGREEN
Quarterly analysisFinanceREVIEW
03
API keys
Create identities with tags, model access, budgets, and limits.
Engineering-ClaudeCodeEnabled
sr_live_8kQp…60requests/min$100monthly budget12.8Ktokens
04
Signals
Surface potential duplicate work across shared project tags.
DUPLICATE GUARD · NEW
OAuth implementation overlap
Two identities tagged FinanceProjectOne requested similar authentication work.
Engineer-AOAuth callback78% matchEngineer-BOAuth login flow
05
Ask SafetyRouter
Ask management questions without exporting employee prompts.
Who had the most blocked content today?
SREngineering-ClaudeCode had 4 blocked requests. Three matched the confidential-company policy.
Ask about activity, warnings, or spend… Ask
06
Policy
Turn organization-specific restrictions into enforced decisions.
Block requests that mention ACME Manufacturing or Pump Manufacturer LLC. Do not send matching content to an external LLM.
FROM ZERO TO GOVERNED AI
One. Two. Three.
It really is that easy.
Choose where it runs
Use a server in your office, your own cloud account, or a managed SafetyRouter tenant.
Create identities
Issue a separate hashed API key for each employee, team, device, or application. Add project tags and limits.
Point your AI tools at it
Change the base URL and API key. SafetyRouter evaluates, routes, and records approved traffic.
BUILT FOR THE TOOLS PEOPLE USE
One policy layer.
Every AI workflow.
Connect compatible coding agents and applications to the same controlled gateway. Owners define models, limits, project tags, and stronger company policy.
View integration guidesO2
OpenCode v2Tested · new provider schema
→O1OpenCode v1Tested · legacy configuration
→CXCodexTested · Responses API
→CCClaude CodeTested · Messages API
→GHGitHub Copilot CLICommunity tested · BYOK provider
→WUOpen WebUIDocker browser chat · compatible now
→CLOpenClawOpenAI-compatible · available now
→HAHermes AgentOpenAI-compatible · available now
→CRCursorChat compatible · Agent adapter roadmap
→Before data leaves
Combine a semantic safety evaluator with deterministic controls for secrets, model access, size, rate, and budget.
Before output returns
Buffer model responses and proposed tool calls, evaluate them, then release only approved content.
Without hunting through logs
Ask the local management assistant about blocked content, warning trends, cost, and Duplicate Guard signals.
One tenant, one container
Keep each customer’s SQLite audit data and configuration inside a dedicated Docker boundary.
SECURITY WITHOUT A SECURITY TEAM
Need the protection—but not another system to support?
Talk to us. We can install SafetyRouter at your office or operate your tenant on our DigitalOcean droplet network.
Talk to SafetyRouterNo oversized platform project. Start with one team and grow.