Agent identities
Every agent authenticates with a Credential — same identity plane as API consumers. OIDC / JWT policies apply; revoke the credential and the next call is denied.
AI teams · A2A
Agents that call agents need their own passport.
Apinizer's A2A gateway gives every agent an identity, a discoverable capability card, and a scoped permission set. Conversations flow through the gateway with the same auth and audit as any other request.
The problem
Agent-to-agent traffic invented itself faster than any governance layer.
An ops agent calls a security agent. A research agent calls a planning agent. None of them have a real identity; none of the calls have an audit trail. The first time a regulator asks 'which agent did this', the org has nothing. Apinizer treats A2A as a first-class plane: identities, scopes, contracts, and a single audit ledger.
Capabilities
What Apinizer does here
Discoverable agent cards
Publish a proxy as an agent and it serves a standard agent card at the well-known path. Front someone else's agent and the gateway rewrites the card — your hostname, your security schemes, only the skills you chose to expose.
Real task lifecycle
Send a message, poll a task, cancel a task — with streaming relay, resubscribe after a dropped connection, and push notification. A cancel actually interrupts the work on the worker; it is not a status flag.
Outbound skills on the traffic record
Which agents a flow reached and which skills it invoked are written onto the ordinary traffic record, alongside a call count — no trace to switch on first. Fronting an external agent behaves the same way: task status and cancellation are relayed to the agent that owns the task, so a client can follow a task it started through a passthrough proxy.
Conversation audit and replay
Every message between agents captured with both sides' identities and the result. Chained calls form one trace tree, and a captured request can be replayed to reconstruct exactly what happened.
Scope enforcement
An agent that's allowed to call 'search' can't call 'modify'. Permission applied at the runtime; nothing depends on prompt obedience.
Same plane as MCP and APIs
A2A lives on the same gateway as API and MCP traffic. One identity surface, one audit ledger, one set of policies.
Use cases
In production, this looks like…
- Banking
Istanbul bank registers 27 agents across customer-care and risk teams
Each agent has an owner, a capability profile, and a permission set. Cross-team calls go through the A2A surface, not direct webhooks.
- Manufacturing
Stuttgart OEM coordinates 9 plant agents through a central planner
Planner agent discovers each plant's local agent via the registry; A2A messaging carries contracts. Plant operators see the entire chain in audit.
- Telecom
Madrid carrier limits a customer-service agent to read-only billing peers
Scope enforced at the gateway. The agent can answer 'what's my balance' through a billing peer but cannot trigger a refund — different scope, different agent.
- Insurance
Paris insurer audits a triage chain end-to-end
Triage agent → underwriting agent → fraud agent. The audit query joins all three legs into one timeline; the DPO traces a complaint in minutes.
- Retail
Amsterdam marketplace registers partner agents alongside internal ones
Partner agents authenticate with JWTs. Their agent cards are published through the gateway; partners never see internal agents they don't have access to.
- Media
Milan publisher revokes a runaway agent in seconds
Agent credential revoked; all conversations drop on next call. No need to chase webhooks across services.
- Public sector
Prague ministry routes citizen-services agents through one A2A bus
Two ministries, one citizen-services chain, one audit trail. Each ministry's agent stays in scope; cross-ministry calls go through the A2A surface explicitly.
- Energy
Baku utility separates ops agents from analytics agents on the registry
Ops agents can call analytics agents; analytics cannot call ops. The boundary is enforced by the gateway, not by hopeful design.
Recommended products
What we'd reach for first
AI Gateway
A2A proxy modes, identity issuance, task lifecycle, conversation audit.
Open the AI Gateway pageIdentity Manager
Agent identities issued and revoked alongside human identities.
Open the Identity pageAnalytics Engine
End-to-end conversation timelines across agent chains.
Open the Analytics pageMonitoring
Severity-aware alarms when an agent chain drifts or stalls.
Open the Monitoring pageResources
Keep going
A2A gateway overview
Identity, discovery, scopes, contracts — the building blocks of governed agent-to-agent traffic.
AI Gateway
Where A2A traffic lives — alongside LLM, MCP, and API calls.
Identity Manager
Issuing and revoking agent identities the same way you handle users.
Analytics Engine
Conversation telemetry — who called whom, with what contract, at what cost.
MCP server governance
When agents call tools, not just other agents.
Architecture overview
Where the A2A plane sits in the topology.
A2A, governed
Give every agent a passport. Audit every conversation.
A 30-minute walkthrough — registry, identity, contracts, audit — on a Kubernetes of your choice.