The authority layer for the A2A protocol
The universal authority layer
for AI agents
Identity, badges, and policy enforcement for the A2A protocol — in two lines of code.
Think Let's Encrypt, but for AI.
guard = CapiscIO.connect()pip install capiscio-sdkBuilt for the ecosystems you already use
Agents talk to each other.
But who's listening?
OAuth tokens prove a user authorized an agent. They don't prove which agent is calling, what it's allowed to do, or who delegated that authority.
Tokens aren't identity
OAuth proves a user authorized an agent. API keys identify apps. Neither cryptographically proves which agent is calling at hop 2, 3, or 10 — or that the payload wasn't modified in between.
No payload integrity
Bearer tokens don't bind to payloads. A valid request captured in transit can be replayed, modified, or reordered. Cryptographic signatures on every message close this gap.
No delegation chain
Agent A calls Agent B calls Agent C — across teams, vendors, even organizations. Your IAM proves who logged in. It can't prove who delegated what to whom, or that scope wasn't escalated along the way. No existing protocol produces this artifact. Until now.
CapiscIO addresses 6 of the OWASP Top 10 for Agentic Applications
OAuth secures the first hop.
CapiscIO secures every hop after.
Think Let's Encrypt, but for AI agents: open, automated, and infrastructure-level.
Proves a user said go. After that first hop, it's bearer tokens — no proof of which agent is calling, what it changed, or who delegated what to whom.
Discovers service accounts and shadow identities. Essential for visibility — but no inline enforcement on every call, no per-message signatures.
Agent identity and every request cryptographically signed at every hop. Cross-org delegation chains verified locally — no callbacks to the issuing organization. Authority provably narrows at every hop.
Open protocol (8 published RFCs) · Open source Go core · Sub-millisecond overhead · Works alongside your existing identity stack
Three layers of trust
From development to production. From single agents to fleet-wide enforcement.
Runtime Identity Verification
Add trust enforcement to A2A and MCP endpoints in two lines of code. Ed25519 signatures, 60-second replay windows, verified DIDs.
Learn more about Guard →from fastapi import FastAPI
from capiscio_sdk.integrations.fastapi import CapiscioMiddleware
app = FastAPI()
app.add_middleware(CapiscioMiddleware)Built on open standards
Real engineering depth. Open source from day one.
Start in 60 seconds
From zero to trust-enforced. No dashboard signup required.
$ pip install capiscio-sdkfrom capiscio_sdk import secure, SecurityConfig, CapiscIO
# Connect (reads CAPISCIO_API_KEY from env)
agent = CapiscIO.connect()
# Wrap your agent with trust enforcement
secured = secure(agent, SecurityConfig.production())
Building with AI Agents? Let's Talk.
I'm looking for 5 design partners building production AI agent systems. You get hands-on deployment support. We get real-world validation. No sales pitch — just problem-solving together.
— Beon de Nood, Founder
Become a Design PartnerLatest Insights
Learn about A2A Protocol, agent validation, and authority infrastructure
Your agents are already talking.
Give them an identity they can prove.
Start with the CLI. Validate your agent cards. Add Guard when you're ready for runtime enforcement.
