Rules for your applications and agents.
Your applications and agents call AI models and tools through Palveron. Every request is checked before it reaches the model, and every decision is stored as a record.
How your rules apply
Connect
You connect applications and agents through an SDK for TypeScript, Python, Go, and Java, through the OpenAI-compatible proxy or through MCP.
Check
Every request is checked against baseline protection and your own rules before it reaches the model. Your rules apply per project.
Approve
For tool calls through MCP you decide for each tool whether it is allowed, blocked or usable only after approval by a person. An approval applies to exactly that call.
Prove
Every decision is stored as a record with an integrity hash and can be searched in the Trace Explorer.
What is available today
- SDKs for TypeScript, Python, Go, and Java
- A proxy for OpenAI-compatible calls
- An MCP gateway with approval per server and rules per tool
- An agent registry with risk levels
- An emergency stop per project
What is planned
- Human oversight of agents at three levels: monitor and intervene, approve defined actions, approve every action
- Sign-in to MCP target servers through OAuth, after the launch
Put your agents under rules
In a demo we show you, using your own applications, how rules, approvals and records work together.
To the platform page: AI Gateway