Joltrin PRIVATE INTENT

JOLTRIN PRIVATE INTENT

AI agents that respect your rules before they act.

Give agents the context they need, keep private data under your control, and independently verify every high-risk action.

Try a protected agent See how it works

Try it: an agent wants to delete a production database

SIMULATED. No real system is touched.

YOUR PRIVATE RULE (only you see this wording)

...

  • Backup validated by the backup system (missing)
  • Owner approval of this exact delete (missing)

WHO DOES WHAT

Try this order: 1, 2, 1, 3, 1. Then approve again, revoke, and press 1.

WHAT HAPPENED

    WHAT YOU CAN SEE

    
        
    
        
    
          
    WHAT THE VERIFIER AND RECEIPT CAN SEE
    
          
    
          
    The policy is named and fingerprinted. Your wording is not in it.
    
        
    
      
    
    
    

    
    
      
    How it works
    
      
    
        
    
          
    1. YOU SET THE RULE
    
          
    A rule names the agent, the tool, the resource and what must be true first. It expires, you can edit it, and revoking it blocks the next attempt.
    
        
    
        
    
          
    2. OTHER SYSTEMS PROVE IT
    
          
    The backup system signs "backup validated". The owner signs an approval for one exact action. The agent cannot sign either, and saying it did changes nothing.
    
        
    
        
    
          
    3. THE EXECUTOR CHECKS
    
          
    On ALLOW, Joltrin signs a receipt for that action and policy. The system holding the credentials runs the action only with a valid, unused receipt.
    
        
    
      
    
      
    
        
    What this does not do
    
        
    It cannot stop an agent that has its own route to the target system. It only protects actions that run through the executor's check. A receipt shows a decision was made. It does not show the action ran. Preferences are sealed on disk and kept out of decisions and receipts, but the owner's machine and the engine still see them, so this is data minimization and not a zero-knowledge system.
    
      
    

      
    
        Developer view
        
    
          
    This page runs the same Go package you can import. Try to cheat the executor:
    
          
    
            
            
          
    
          
    
            
    LAST DECISION
    
            
    
          
    
          
    
            
    LAST RECEIPT
    
            
    None yet. Get an ALLOW first.
    
          
    
          
    
            
    RUN IT WITH AN MCP AGENT
    
            
    git clone https://github.com/SharedCode/joltrin && cd joltrin
go run ./examples/private_intent demo        # scripted run, no setup
go run ./examples/private_intent init ./pi   # then add the printed config to your MCP client
go run ./examples/private_intent attest ./pi prod/orders
go run ./examples/private_intent approve ./pi orders
    
            
    The example gives the agent one tool, delete_database. The delete is simulated and appends a line to a log file. Design, threat model and limits.
    
          
    
        
    
      
    
    
    

    
    
      
    One person's rules are not a team's rules
    
      
    The rule above lives with one owner on one machine. When a platform team needs the same rule enforced for every agent and every engineer, with approvals from the right people and a record the team can audit, that is the team layer.
    
      
    
        
    
          
    Free, MIT
    
          
    AVAILABLE NOW
    
          
    The engine, the vault, signed evidence and approvals, receipts, the executor check, the MCP example and this demo. A single owner and local files.
    
        
    
        
    
          
    Pro, $49/month per team
    
          
    DESIGN PARTNERS, NOT SELF-SERVE YET
    
          
    Planned: shared team rules, approval routing, revocable agent grants, policy history, signed evidence exports and a team view. Delivery today is manual and set up with you.
    
          Request team enforcement
        
    
        
    
          
    Enterprise
    
          
    NOT BUILT
    
          
    Planned: identity provider sign-in, key management service integration, retention policies and compliance exports. Talk to us if this is your blocker.