// Works with Agentics
Run any agent. Prove every one.
An Agentics Certified agent is one you can let into your tenant on day one — because every model call, every tool call, and every outcome it produces is governable, scoped, and verifiable on the record. The mark isn't a logo. It's a passed standard: the Agent Compatibility Standard.
Certification asserts compatibility with the record — that an agent is governable and verifiable — not that it's accurate or safe in absolute terms. Require it before a partner agent runs in production.
The directory is open and live, and the standard is final and self-certifiable today. We don't list agents we haven't certified, so you'll see the directory in its certification-ready state: the real ACS criteria, the live category map, and open slots — not a wall of borrowed logos. Certified agents appear here the day they pass.
Agentics Certified
Sample format
acs_versionacs-1.0
statuscertified
scopedeclared · enforced
identityagent_id @ version
R1 · Tools routed through MCP
R2 · Models on a gateway base URL
R3 · Outcomes resolve to the record
R4 · Stable identity on every call
R5 · Scope declared and enforced
Illustrative layout of a certification record. No partner is named because none is certified yet. Certified records are scoped per partner and per tenant.
// What the mark means
Certification is a passed standard, not a handshake
The Agent Compatibility Standard (ACS) is the small, stable contract that makes an agent compatible with the record. An agent earns the mark by passing all five checks — each independently verifiable, each machine-checked. The builder keeps their prompts, framework, and model choices; they change roughly one line and declare what the agent may touch.
R1
Tools via MCP
Every tool and service call goes through MCP, so each one is governable, scopeable, and on the record — never an invisible direct SDK call.
R2
Gateway-routable models
Model calls point at a gateway-routable base URL. One environment variable; the SDK, model names, and streaming all stay.
R3
Outcome resolution
A webhook or polling endpoint resolves each action to approved, reversed, or incident — sealed as a chained outcome receipt.
R4
Stable identity
A stable agent_id and agent_version ride on every call, so a receipt can be traced to the exact certified version that produced it.
R5
Scope declaration
The agent declares its scope at registration; the gateway enforces it and receipts every violation as evidence, not a silent drop.
Self-certifiable in 1–2 days. The standard is additive and stable — conforming never changes how receipts hash, chain, or anchor. Read the full standard ↗
// Why an enterprise should require it
Don't let an unproven agent into your tenant
A third-party agent is a third-party actor inside your boundary. Requiring the mark is how you make that actor accountable before it does anything — not after an incident.
Governable on arrivalA certified agent routes every model and tool call through paths your gateway can see, scope, and stop. Nothing it does is invisible to you.
Bounded by a declared scopeIt told the platform exactly what it may touch — tools, data classes, spend ceiling, tenants. A call outside that scope is denied and receipted, not silently dropped.
Verifiable after the factEvery move it made is on a record your auditor can check without us — traced to the exact certified version. The cert can be revoked without touching the sealed receipts underneath.
// Lifecycle
A certification is per agent version. A new version is a new certification — passing one never silently re-certifies another. A mark can be revoked on scope drift or repeated violations.
pending
certified
revoked / expired
// The directory
Browse certified agent platforms
Search and filter by category. Each card shows certification status, the categories the agent serves, and a link. The directory lists only agents that have passed the ACS — the categories below are open and waiting for their first certified entry.
No categories match that search. Clear it to see the full map.
// Read the mark honestly
What "Agentics Certified" does — and doesn't — assert
A mark that overclaims is worthless. ACS certifies one specific thing precisely, and is deliberately silent on the rest.
It asserts
- This exact agent_id @ agent_version passed every ACS check.
- Its tool and model calls are governable through the gateway.
- Its outcomes resolve onto a sealed, chained record.
- Its scope is declared and enforced — violations are receipted.
- An enterprise can verify its moves without trusting us.
It does not assert
- That the agent is accurate, "good", or safe in absolute terms.
- That Agentics endorses or vouches for the vendor.
- That the agent fits your specific use case or risk appetite.
- Anything about a different version than the one certified.
- That a cert, once granted, can never be revoked.
ACS certifies compatibility with the record — that an agent is governable and verifiable — not quality. That's exactly what makes the mark worth requiring.
// Two ways in
Require the mark. Or earn it.
Enterprises: make ACS certification a procurement requirement, and never run an unproven agent in your tenant again. Builders: self-certify in 1–2 days and get listed the moment you pass.