Fits what you run.
Works with your identity provider. We don't replace your IAM or AI governance; we're the identity layer under them.
Keep the stack you already trust. Your identity provider still says who may sign in, your agent directory still lists your agents, and your runtime still does the work. Oathera sits quietly beneath all of it and answers the one question they can't: which agent is really asking. It slots in beside the tools your teams know, so agent identity is something you switch on, not another platform to adopt.
Your platforms stay yours. Oathera proves which agent is asking.
Your identity provider signs people in, and your agent directory lists your agents. Oathera proves which agent is asking: the real one, on its own machine, with a person who answers for it. Then it checks every call against your rules, watching first, then enforcing.
Registering an agent isn't verifying it
A directory entry is a name and a secret. Copy the secret and the copy is the agent. Nothing in single sign-on can tell them apart.
Oathera adds the missing layer
Your agent's DNA is recorded once, at enrollment: what it said about itself, what Oathera found on its host, and the person who approved it. Your agent's fingerprint is on every call it makes, and a copy on another machine leaves a different one.
Alongside what you run
Oathera links each agent to the identity your agent directory already gives it, and writes your agent's DNA there wherever the directory can hold it. Your identity provider can accept a token Oathera issues to the verified agent, through federation. At your MCP server, every call is checked against your rules: watching first, then enforcing.
Flagging an impostor after it acts, or checking before it can.
Behaviour profiling flags an impostor after it has acted. Oathera checks the agent's fingerprint before every call. These are two approaches to the same problem, and they sit well together.
- The agent presents its identity token and signs each call with its session key.
- Oathera Agent Enforcement, on your MCP server, verifies the token and the operation proof, and checks the agent's fingerprint.
- Your rules decide the call: allow, deny or approval required. Each rule watches first, logging what it would refuse, and enforces when you turn it on.
- The service, such as Slack or email, receives a short, single-use gateway assertion in place of the token.
Your identity provider and your agent directory sit beside this flow, not in it. Your identity provider signs in the person who approves the agent, and Oathera links the agent to the identity your agent directory gives it.
What each one does, and what Oathera adds.
Okta
Okta registers an agent as an identity in its directory, with owners, short-lived tokens, discovery of unmanaged agents and governance. The agent presents an ordinary bearer token. Oathera adds proof that the caller is the enrolled agent on its own host, and checks every call it makes.
Microsoft Entra
Entra Agent ID gives each agent an identity and requires a sponsor for it. The credential sits with the agent's blueprint, as a secret, a certificate or a federated credential. Oathera's principal is the same idea, the person who answers for the agent. Oathera adds the tie to the host and the check on every call.
NVIDIA OpenShell
OpenShell is an open-source sandbox runtime for agents. It enforces policy with Landlock, seccomp and a network proxy, and it contains what an agent can reach. Oathera proves which agent is inside.
Your agent directory
Your agent directory lists your agents. Oathera links each agent to the identity the directory already gives it, and writes your agent's DNA there wherever the directory can hold it, so you don't run a second identity system.
What Oathera works with.
Your identity provider, over OIDC
A principal approves each agent by signing in at your identity provider, with OpenID Connect and PKCE. Oathera relies on your identity provider for who a person is. Our hosted demo uses a demo identity provider.
Your agents, over MCP
Bring your own agent. Claude Code adds the MCP bundle as two servers, oathera-identity and oathera-gateway.
Your MCP servers, in Node
Oathera Agent Enforcement attaches inside an MCP server built on the Node MCP SDK, so every enforced tool call is checked and logged.
Your hosts and clouds
On the agent's own machine, Oathera reads the host on Linux, macOS and Windows, detects Docker, Kubernetes and Podman, and reads AWS, Google Cloud and Azure instance details.
Google Cloud attestation
On Google Cloud, an instance identity statement bound to the agent's key reaches verified, and a setting makes it mandatory.
W3C verifiable credentials
When set to issue them, the credential service records the agent and its operational boundary as verifiable credentials, with suspend, reinstate and revoke status.
OWASP Non-Human Identity Top 10
An agent can answer the OWASP NHI questions at enrollment. The Oathera Identity Service derives a risk tier and NHI1 to NHI10 findings.
Microsoft Entra federation
Exchange an agent's Oathera identity for a token Microsoft Entra accepts.
SPIFFE and SPIRE
Accept a SPIRE-issued workload identity as extra evidence at enrollment. The agent's identity stays its anchor key.
MCP servers in any language
Oathera Agent Enforcement proxy in front of an MCP server written in any language, over stdio or Streamable HTTP.
Your SIEM or write-once store
Export the signed audit log exactly once to your SIEM or a write-once store.
People sign in where they already do.
Oathera never verifies who a person is. It relies on your identity provider and your proofing sources for that, and proves things about the agent. The identity provider in our hosted demo is a demo, not a product.
- OIDC sign-in with PKCE at your identity provider
- The sign-in is bound to the agent being enrolled
- Fits beside your IAM, egress controls and AI governance
Run it beside what you have.
Design partners run Oathera in front of one real tool with real agents. Tell us what you run and we'll tell you what fits.