Pipeline

Let Your Agents Run.
We Verify Every Step.

The more autonomous your agents, the more you need zero trust. AgentSign verifies identity, code, execution, and output -- cryptographically, on every interaction. Your agents stay free. Your org stays safe.

0
Agents
0
Active (Prod)
0
Verified Executions
100%
Verification Rate
Signing: --
Keys: --
License: --
Agents: --
Phones Home: NO

Onboarding Pipeline

🛡

Prompt Injection

Agents are autonomous -- they interpret instructions from data. Runtime code attestation catches when agent behavior deviates. Every execution is hashed and compared pre/post. Tampered? Blocked.

🏠

Data Leaks

Autonomous agents touch sensitive data. AgentSign runs 100% on-prem -- your data never leaves your network. Every agent-to-tool interaction is logged and verified. Full audit trail.

🚧

Supply Chain

Agents come from vendors, open source, partners. The onboarding pipeline vets every agent before it touches production. Security scan, code signing, sandbox testing. No shortcuts.

🔒

Agent Misalignment

Agents decide what to do next. Execution chain verification creates a signed DAG of every decision. If an agent goes off-script, the chain breaks. Instantly detectable.

⛔

API & Tool Misuse

Agents call tools autonomously. The MCP Trust Gate verifies identity and trust score before every tool access. Low trust? Blocked. Revoked? Every tool knows immediately.

🔌

Zero Visibility

Agents run in the background -- you can't see what they're doing. AgentSign gives you a live monitor: who's calling what, trust scores, health status, anomaly detection. Full observability.

How It Works

Don't constrain agents. Let them run. Verify everything.

Your Agents (Autonomous)

Agents Act Freely

Your agents call tools, make decisions, chain actions. They're autonomous -- that's the point. AgentSign doesn't restrict them. It watches and verifies.

AUTONOMOUS

Carry Their Passport

Every agent carries a self-contained cryptographic passport. Identity, code hash, permissions, trust score -- all signed. Any system can verify, even offline.

SELF-VERIFYING

Build Trust Over Time

Every successful execution increases trust. Every verified output raises the score. Agents earn their reputation through proof, not promises.

EARNED TRUST
→
→
←
↓
←
AgentSign (Governance Layer)

Verify Every Step

Every function call hashed. Every agent-to-tool interaction signed. Every output checked for tampering. Cryptographic proof at every step -- not sampling, not logging. Proof.

VERIFIED

Gate Every Tool

Agent wants to call your database? MCP Trust Gate checks: passport valid? Trust score high enough? Pipeline stage allowed? ALLOW or DENY. Every interaction.

MCP GATE

Revoke Instantly

Agent goes rogue? One click -- trust drops to zero. Swarm compromised? Revoke them all. Every system that checks the passport knows immediately.

INSTANT

Activity Log

All Agents

Agent Stage Trust Executions Source Actions

MCP Trust Layer

Agent-to-MCP-server identity verification. No network tap needed -- all traffic goes through passport checks.

0
MCP Servers
0
Verifications
0
Allowed
0
Denied
--%
Allow Rate

Network Graph -- Agent-to-MCP Connections

Agent Monitor

Darktrace for AI agents. Query, observe, and audit every agent-to-MCP interaction in real time.

Ask About Your Agents

What is OpenHands doing? Which agents are blocked? Health status Busiest agent Database MCP connections Trusted agents Agent counts 

      

    


    

      

        
 Activity Stream

        

          
Load demo data to see activity.

        

      

      

        
Agent Health

        

          
No agents yet.

        

      

    


    

      
Traffic Flow -- Top Agent-MCP Pairs

      

        
No traffic data yet.

      

    

  







  

    
Agent Catalog

    
Deploy pre-built enterprise agents. Each goes through the full onboarding pipeline.

    

      
Loading catalog...

    


    

      
Recent Runs

      

        
No agent runs yet. Deploy an agent and run a task.

      

    

  










  

    

      
AgentSign Pay

      SANDBOX
    

    
Financial infrastructure for AI agents. Developers sign up, agents get wallets, payments are cryptographically signed.


    
    

      
0
Developers

      
0
Wallets

      
£0
Total Funded

      
£0
Total Spent

      
0
Transactions

      
0
Denied

    


    
    

      
      
      
    


    
    

      

        
Developers

        

      

      

        
Agent Wallets

        

      

    


    
    
Transaction Feed

    

      
No transactions yet. Register a developer and fund an agent to start.

    

  










  

    
Integrate AgentSign

    
Add zero trust verification to any agent in 3 lines of code


    
1. Install the SDK

    
pip install agentsign


    
2. Wrap Your Agent

    
from agentsign import AgentSign

# Initialise -- creates local signing keys automatically
agent = AgentSign(name="my-agent", description="What I do")

# Decorator: every call to this function is auto-signed
@agent.signed
def process(data):
    result = do_work(data)
    return result

# That's it. Every execution is now:
# - Hashed (SHA-256 of input + output)
# - Signed (HMAC-SHA256, local key)
# - Logged (audit trail)
# - Verifiable (by any system, even offline)


    
3. The Agent Carries Its Passport

    
# The passport is self-contained and tamper-proof
passport = agent.passport()

# Any system can verify it -- NO server needed
is_valid = AgentSign.verify_passport(passport)
# True = agent is who it says it is, code hasn't changed
# False = tampered, revoked, or invalid

# Check what the agent is allowed to do
print(passport["permissions"])   # ["code_read", "file_write"]
print(passport["pipeline_stage"]) # "ACTIVE"
print(passport["revoked"])        # False


    
4. Connect to Your Org's Engine (Optional)

    
# Point the SDK at your org's zero trust engine
agent = AgentSign(
    name="my-agent",
    engine_url="https://agentsign.internal.yourorg.com",
    api_key="ask_..."
)

# Now the agent:
# - Registers with the engine on startup
# - Syncs execution hashes (NEVER actual data)
# - Gets trust score from the engine
# - Can be revoked remotely
# - Still works if engine is down (passport)


    

      
What NEVER Leaves Your Org

      

        

          
STAYS LOCAL (always)

          
    
            
  • Actual input/output data
    • 
            
  • Agent source code
    • 
            
  • Signing keys
    • 
            
  • Execution context/prompts
    • 
            
  • User/customer data
    • 
          

        

        

          
SHARED (hashes only, if connected)

          
    
            
  • SHA-256 input hash
    • 
            
  • SHA-256 output hash
    • 
            
  • Execution timestamp
    • 
            
  • Agent ID + pipeline stage
    • 
            
  • Trust score
    • 
          

        

      

    

  







  










  
AgentSign by CyberSecAI Ltd -- Patent Pending GB2604808.2

  
Enterprise Zero Trust Engine for AI Agents