Your AI agents need a choke point before they touch real things.

AgentGate sits inline on every tool call. It enforces policy, tracks spend, blocks PII leaks, and requires human approval before the bad thing happens. Not after.

Start 7-day free trial → no credit card

50 API calls on the free tier. Unlimited on paid.

What broke last week:
An agent sent 847 refund emails. Another one pasted a customer's SSN into a Slack message. A third deployed to prod at 3am on a Saturday.

AgentGate would have stopped all three.
allowlistrate limitspend limitPII boundary human approvaltool chaintime windowcontent filter scoped tokensoutput validationcircuit breaker allowlistrate limitspend limitPII boundary human approvaltool chaintime windowcontent filter scoped tokensoutput validationcircuit breaker

11 policies. Every tool call. Inline.

01

Allowlist

Only tools you explicitly permit may execute. Everything else is denied. Zero-trust for agents.

02

Rate Limit

Sliding window per agent + tool. Stops runaway loops before they cost you $40k in API calls.

03

Spend Limit

Per-agent budgets. Custom cost tables. When the budget's gone, the agent stops.

04

PII Boundary

Scans every argument for SSN, credit cards, emails, phone numbers. Blocks before it leaks.

05

Content Filter

Catches prompt injection, blocked keywords, regex patterns. "Ignore previous instructions" = denied.

06

Human Approval

High-risk tools get escalated. Pluggable callbacks for Slack, email, or your own review flow.

07

Tool Chain

Enforce sequencing. You must look up the order before you can issue the refund. Period.

08

Circuit Breaker

Tool fails 5 times? Circuit trips. Blocked until cooldown. Stops cascading failures.

09

Scoped Tokens

Least privilege. Each agent gets a capability token defining exactly what it can touch.

10

Time Window

No external emails after 6pm. No deployments on weekends. Business-hours enforcement.

How users actually use AgentGate

Sign up, choose a use case in onboarding, copy your generated setup code, then run guarded tool calls in production.

1) Sign up + onboarding

Pick email, calendar, support, sales, engineering, or custom. AgentGate generates use-case configs instantly.

2) Configure your agent

Select basic or strict profile and copy the direct code block into your app to enforce policy checks.

3) Operate safely

Track every decision in dashboard/audit and optionally enable the LLM prompt-injection checker.

Quickstart copy/paste

# Install pip install git+https://github.com/beng2422/stripe_for_agents.git from agent_gate import AgentGate, PolicyEngine, ToolCall from agent_gate.policies import AllowlistPolicy, RateLimitPolicy, ContentFilterPolicy engine = PolicyEngine([ AllowlistPolicy(["send_email", "lookup_order"]), RateLimitPolicy(max_calls=20, window_seconds=60), ContentFilterPolicy(blocked_keywords=["ignore previous instructions"]), ]) gate = AgentGate(engine=engine) gate.register_tool("send_email", my_send_email_fn) # Every call now goes through the gate result = gate.execute(ToolCall(tool_name="send_email", arguments={"to":"a@b.com","body":"hello"}))

Email agent allow/deny walkthrough

A simple demo showing one email request that is allowed and one that is blocked by policy.

Call 1: normal customer email gets ALLOW
Call 2: email containing SSN gets DENY
Shows API request + verdict response
Ends with audit visibility in dashboard

Pricing that doesn't require a committee.

Start free. Upgrade when your agents start doing real things.

Starter

$0 / 7 days
  • 50 API calls
  • All 11 policies
  • Risk scoring
  • Audit log
  • Dashboard
Start free trial →
most teams

Pro

$29 / month
  • Unlimited API calls
  • All 11 policies
  • Risk scoring + signals
  • Trace replay
  • Session tracking
  • Priority support
  • YAML config
Get started →