Permission Control
for AI Agents
Control what your AI agents can and can't do. Policy-driven permissions, real-time audit logging, and human-in-the-loop approval — in 3 lines of code.
pip install vaultagentAI agents have too much power
Today's frameworks give agents unlimited tool access with no audit trail. VaultAgent changes that.
Unchecked Tool Access
Agents call any tool with any arguments — no boundaries.
Zero Visibility
No record of what tools were called, when, or with what data.
Compliance Blind Spots
No way to prove agents follow organizational policies.
Policy-Driven Permissions
YAML rules define exactly which tools each agent can use.
Full Observability
Every tool call logged with parameters, latency, and decision context.
Compliance Ready
Audit logs and approval workflows for SOC2, HIPAA, GDPR.
Secure in 3 lines of code
Wrap your existing AI client. No architecture changes.
Define your policy
Write a simple YAML file that says which tools each agent can use, with constraints like table whitelists or domain restrictions.
Wrap your AI client
One function call wraps OpenAI, Anthropic, or LangChain. Every tool call now goes through VaultAgent first.
Monitor & control
Watch tool calls in real time. Review approvals. Export audit reports. All from the Dashboard.
[0mfrom openai import OpenAI
from vaultagent import VaultAgent
from vaultagent.middleware import wrap_openai
# 1. Load your permission policy
vault = VaultAgent(policy="policy.yaml")
# 2. Wrap your client — one line
client = wrap_openai(OpenAI(), vault)
# 3. Use as normal — all tool calls
# are now policy-controlled ✓
response = client.chat.completions.create(
model="gpt-4o",
messages=messages,
tools=tools,
)Everything you need to secure AI agents
Enterprise-grade controls, open-source simplicity.
Policy Engine
YAML-based rules with glob patterns, constraints, and per-agent scoping. Human-readable, version-controllable.
Real-time Dashboard
Live monitoring of agent activity, tool calls, approval queues, and policy violations.
Audit Logging
JSONL audit trail with full context. Export for SOC2, HIPAA, or GDPR compliance.
Multi-Framework
Drop-in wrappers for OpenAI, Anthropic, LangChain. Custom adapters via Python decorator.
MCP Proxy
Protect any MCP server with policy enforcement. Works with Claude Desktop, Cursor, OpenClaw.
Approval Workflows
Sensitive operations require human approval. Slack, email, and dashboard notifications.
Simple, transparent pricing
Open-source SDK is always free. Pay only for the cloud dashboard.
Free
For individual developers and open-source projects.
- Unlimited SDK usage
- Local audit logging
- Community support
- Self-hosted dashboard
Pro
For teams building production AI applications.
- 10 agents
- 100K events/day
- Cloud dashboard
- 90-day log retention
- Approval workflows
- Email support
Team
For organizations with advanced security needs.
- 50 agents
- 1M events/day
- SSO / SAML
- 1-year log retention
- Compliance reports
- Priority support
- Custom integrations
Enterprise
On-premise deployment with dedicated support.
- Unlimited agents
- Unlimited events
- On-premise option
- Custom retention
- 99.9% SLA
- Dedicated CSM
- SOC2 certification
- Professional services
Ready to secure your AI agents?
Get started in under 5 minutes. Open source. No credit card required.