AgentBridgegovernance for AI agents
The problem
Your AI agents are acting.
Can you prove what they did?
They call tools, move data, spend money — with no record, and no way to stop a bad call.
AgentBridge
One neutral mesh every agent speaks through.
translate · route · verify · govern
MCP
A2A
ACP
OpenAI
Gemini
AGNTCY
Governance in the call path
Policy · agent-007
DENYcapabilitieswire_transfer · delete_database
MAXcost per call3.0
APPROVALrequired above2.0
Enforced before the tool is ever touched
ALLOWEDsafe call→ 5
BLOCKEDwire_transferdenied by policy
BLOCKEDcost 4.0exceeds cap 3.0
BLOCKEDcost 2.5needs human approval
The proof
#0allowaddd196e0c5…
#1denywire_transfer500586d1…
#2denyadd51c8326e…
#3denyadd0a15540a…
✓ Audit integrity verified
tamper-evident · hash-chained · ~0.4 ms overhead
Govern your agents.
Prove what they did.
EU AI Act · Article 12 · automatic event logging for high-risk AI · from Aug 2026
github.com/shadowhunter-92/agentbridge
Open source · Apache-2.0 · any protocol in, any protocol out