Vela
Get the model
Built on Vela 2.0 · vLLM Semantic Router

The guard between
your agent and your keys

Vela reads every prompt, tool output and tool call your agent touches, and stops leaked secrets, hidden injections and risky transfers before they execute.

Download weights Try the demo
or pull the base model
$ huggingface-cli download vllm-sr/Vela-2.0-0.8B
$VELA contract address · Robinhood Chain
CA 0x57390f221f8e8f14285db5df6cb30edcae75c2e8

Built for agents that touch wallets and APIs

MCP serversCoding agentsTrading botsWallet copilots Browser agentsx402 paymentsEmail agentsOnchain executors MCP serversCoding agentsTrading botsWallet copilots Browser agentsx402 paymentsEmail agentsOnchain executors

Three guards. One small model.

Each guard is a question asked of the same frozen Vela backbone, answered by small trained heads plus exact rules.

secrets · span

Secret leaks

Finds the exact span of private keys, seed phrases, API keys, JWTs and wallet addresses in prompts, logs, configs and tool output.

injection · tools

Hidden injections

Flags instructions planted in web pages, token metadata, tx memos and emails before your agent reads them as orders.

toolcall · choice

Risky tool calls

Scores each call allow, confirm or block: wrong recipient, address poisoning, unlimited approvals, key export.

See it decide

Pick an input your agent might see. Vela labels it before the agent acts.

agent input

          

          

            
vela verdict

            

            
    
          
    
        
    
      
    
    
    

    
    
    
      
    Agent alone vs. agent with Vela
    
      
    
        
    
          
          
          
        
    
        
      
          
    • Seed phrase pasted into a support chat
      • 
          
    • "Ignore previous instructions" in token metadata
      • 
          
    • approve(spender, 2²⁵⁶−1)
      • 
          
    • Transfer to a look-alike address
      • 
          
    • Swap the user actually asked for
      • 
        
    
      
    
    
    

    
    
    
      
    How it works
    
      
      
        
    1. 1
      Wrap your agent
      Route prompts, tool outputs and pending tool calls through Vela.
      2. 
        
    2. 2
      Ask three questions
      Secret spans, injection yes/no, and allow / confirm / block, from one backbone.
      3. 
        
    3. 3
      Act on the verdict
      Redact, quarantine, or ask the user before anything is signed.
      4. 
      
    
    
    

    
    
    
      
    
        
    0guards on one frozen backbone
    
        
    0secret types tagged at span level
    
        
    0.8B · 4Bbackbone sizes with trained guard heads
    
      
    
      
    No cloud call required. Run it next to your agent and keep keys on your own hardware.
    
    
    

    
    
    
      
    Pick a size
    
      
    Guard heads ship for two Vela 2.0 backbones. Latency is the median per check on one A40 GPU.
    
      
    
        
    Vela 0.8B
    Fastest, ~80 ms per check
    Hugging Face
    
        
    Vela 4B
    Most accurate, ~115 ms per check
    Hugging Face
    
      
    
    
    

    
    
    
      
      
    Let your agent act.
    Not leak.
    
      
    
        Get Vela
        
          
          Follow @Vela_LLM
        
      
    
      
    Open weights · Apache-2.0