Compare · Obelisk vs Keycard

Obelisk vs Keycard

Looking for a Keycard alternative? Here's an honest, side-by-side comparison of Obelisk and Keycard — what each does best, where they differ, and where Obelisk falls short. Keycard is runtime auth for agents: it gives each agent its own identity, authority to act, and access to resources under your policy.

Obelisk vs Keycard, honestly

Keycard is runtime auth for agents: it gives each agent its own identity, authority to act, and access to resources under your policy. (source, checked 2026-11-06) Keycard is a genuinely good product for teams that want policy-checked, short-lived credentials for agents on top of the identity provider they already run, and we say so plainly below.

The short version: if your agents need tokens of their own today, Keycard issues them and Obelisk does not (see the table). Where Obelisk differs is the record it keeps of each agent.

Side-by-side: Obelisk vs Keycard

Obelisk's column says what Obelisk does today, including what it does not do. Every Keycard cell is drawn from the page it links, checked on the date shown; vendor capabilities change, so verify the latest from their docs.

DimensionObeliskKeycard
Agent as an identityEach agent is its own record in Agent Studio: its own key, held by the agent's runtime (Obelisk never receives the private key), a passkey-holding owner, and an opt-in public Proof Link that is opaque and revocable.Each agent gets its own identity, and Keycard verifies the agent making the request and where it runs. (source, checked 2026-11-06)
Tokens issued to an agentNo OAuth grant issues an access token to an agent on its own: the token endpoint grants only authorization_code and refresh_token. There is no client-credentials, token-exchange or ID-JAG grant.Yes: after a policy check, a short-lived authorization token scoped to the task at hand. (source, checked 2026-11-06)
Access policy for agentsAPI keys bound to an agent carry a fixed capability and scope. No runtime policy engine evaluates an agent's requests.Cedar policy as code, managed with Terraform and evaluated on each request. (source, checked 2026-11-06)
Relationship to your identity providerObelisk is the identity provider: an OpenID Connect provider that your apps sign in with. Its SAML 2.0 identity provider is built, but SAML apps such as GitHub, Google Workspace, and AWS can't be connected on the hosted Gate yet.Sits on top of the identity provider you already run and adds policy and token issuance. (source, checked 2026-11-06)
RevocationOwners rename agents, rotate their keys, transfer them and revoke them. Revoking an agent also revokes the agents it delegated to and kills their bound keys.Revoke a credential instantly, and later events are denied wherever it applies. (source, checked 2026-11-06)
Third-party check of agent authorizationRegistering, approving, transferring or revoking an agent writes a receipt to Obelisk's hash-chained ledger. Anyone holding a receipt's hash can fetch its position in the chain and a Merkle inclusion proof at /api/proof/<hash>, with no account.Every authorization event, denials included, carries full attribution and exports to your SIEM or an S3 bucket. The cited page describes no proof a third party could check without Keycard. (source, checked 2026-11-06)

How the pieces in Obelisk's column work (receipts, the signed ledger head, agent records) is explained once on the comparison hub.

Where Keycard shines

Each request is evaluated against policy at runtime: an approved one gets a short-lived token scoped to the task, and a denied one creates no credentials but is logged. (source, checked 2026-11-06) Policy is written in Cedar and managed through Terraform, and Keycard adds policy and token issuance on top of the identity provider you already run instead of replacing it. (source, checked 2026-11-06) Its site lists a SOC 2 Type II attestation. (source, checked 2026-11-06)

What matters most against Keycard

Of everything Obelisk does, these are the pieces that matter most for teams that want policy-checked, short-lived credentials for agents on top of the identity provider they already run:

  • Agent identities with public Proof Links. Agents enroll with their own runtime-held key (Obelisk never receives it), a passkey-holding owner stays accountable, and anyone can inspect the bounded live evidence at an opaque, revocable proof link. Reading a Proof Link needs no Obelisk account.
  • Tamper-evident, hash-chained receipts. Sign-ins, tokens and grants are written as hash-chained receipts, anchored under an ES256-signed ledger head you can check against our published JWKS. Save the head and its anchor count: a later count that went down, or a different head at the same count, shows the record was cut short or replaced. Consistency proofs between heads aren't published yet.
  • Sender-bound tokens (DPoP, RFC 9449). Opt-in, per request: an access token can be bound to a client-held P-256 key, and each call then carries a fresh proof signed with that key. A copy of a bound token does not work without that proof wherever the proof is checked: Obelisk checks it where it accepts these tokens, and your own server has to check it too. A token issued without a proof is an ordinary bearer token.
  • CAEP / Shared Signals propagation. Revocations and credential changes reach relying parties as signed Security Event Tokens — push or poll — instead of waiting for the next token refresh to notice.
  • A live public MCP tool server. Obelisk answers trust questions to the software that increasingly does the checking: read-only MCP tools at POST /mcp, rate-limited and account-free — the same checks a human runs on the website, callable by any assistant.

Frequently asked questions

Is Obelisk a Keycard alternative?

Only in part. Keycard sits on top of the identity provider you already run and issues short-lived, policy-checked credentials to agents at runtime. Obelisk is itself the identity provider, and it issues no policy-checked runtime credential and no OAuth access token of an agent's own. An owner can mint an expiring API key bound to one agent, and an enrolled agent can get a short-lived run token that proves a single run. Where the two differ most is the record: Obelisk writes agent lifecycle events to a hash-chained ledger, and anyone holding a receipt's hash can fetch its inclusion proof with no account.

Does Obelisk evaluate agent requests against a policy?

No. API keys bound to an agent carry a fixed capability and scope. No runtime policy engine evaluates an agent's requests. If you need a policy decision on every agent request, that is the problem Keycard addresses.

See it for yourself

Still weighing options? Head back to the full comparison hub to see Obelisk against every provider at a glance.

Was this page helpful?

Give your users a login that keeps receipts.

Passkey sign-in, OAuth for agents, and a record anyone can check.

Create your account →Verify it yourself →

Already have an account? Sign in.