Agent report card
Can an agent use OpenRouter API through Model access lifecycle?
Five-stage assessment
Where agent autonomy holds and where it breaks
- EvaluateReady
- Sign upLimited
- PayLimited
- ProvisionReady
- OperateLimited
The exact service and stable evaluation or access entrypoints are publicly discoverable.
OpenRouter documents its unified API service and public chat completions endpoint at https://openrouter.ai/api/v1/chat/completions. The entrypoints and integration methods are discoverable in llms.txt.
Price or no-charge status, variables, currency, and material conditions are explicit.
Verified structured artifacts accelerate discovery of the evaluation surface.
Web-agent access policy is explicit and verified for the assessed resources.
Evidence and actions4 signals
Can an agent estimate cost or confirm no-charge status before commitment?
supportedPrice or no-charge status, variables, currency, and material conditions are explicit.
OpenRouter explicitly details model pricing across prompt, completion, request, and cache tokens in USD, and defines a value of 0 as free. This allows an agent to calculate costs or verify no-charge status before making requests.
Can an agent find the exact service and its stable entrypoints?
supportedThe exact service and stable evaluation or access entrypoints are publicly discoverable.
OpenRouter documents its unified API service and public chat completions endpoint at https://openrouter.ai/api/v1/chat/completions. The entrypoints and integration methods are discoverable in llms.txt.
Are verified structured discovery artifacts available?
supportedVerified structured artifacts accelerate discovery of the evaluation surface.
OpenRouter provides verified structured discovery artifacts including llms.txt and an OpenAPI specification document at openrouter.ai/openapi.json.
Does the vendor deliberately describe access for web agents?
supportedWeb-agent access policy is explicit and verified for the assessed resources.
OpenRouter's robots.txt declares crawler policy, explicitly allowing all user agents access to root while disallowing /seo/ and providing sitemap locations.
Only part of the access flow is operable or a bounded human step remains.
OpenRouter documents an OAuth PKCE flow directing users to /auth, but it requires a user login and authorization step before exchanging the authorization code for an API key.
A stable canonical route begins the required access bootstrap.
Scoped revocable credentials or delegated authorization have explicit consent and resumption boundaries.
Evidence and actions5 signals
Can an agent operate the access controls and safe handoffs deterministically?
supportedOnly part of the access flow is operable or a bounded human step remains.
OpenRouter documents an OAuth PKCE flow directing users to /auth, but it requires a user login and authorization step before exchanging the authorization code for an API key.
Is there a stable route to begin obtaining service access?
supportedA stable canonical route begins the required access bootstrap.
OpenRouter provides a stable canonical sign-up landing route at /sign-up to begin account creation.
Can an agent obtain access without an unsupported CAPTCHA boundary?
supportedCurrent admissible evidence does not resolve this finding.
The captured sign-up page and OAuth PKCE documentation do not declare or rule out CAPTCHA boundaries across the flow.
Can an agent obtain scoped, revocable authority for this service?
supportedScoped revocable credentials or delegated authorization have explicit consent and resumption boundaries.
OpenRouter supports delegated authorization via OAuth PKCE returning user-controlled API keys and Workload Identity token exchanges scoped to inference.
Can required phone verification be completed through a supported boundary?
supportedCurrent admissible evidence does not resolve this finding.
The retained evidence contains no explicit declaration regarding phone verification requirements or compatible handoffs.
What would improve this stage
- Expose deterministic controls, validation, and resumable authorization handoffs.
Self-service is tier- or condition-limited, or a bounded vendor review exposes status.
OpenRouter enables direct self-service pay-as-you-go onboarding and credit top-ups for Standard and Business tiers. Self-service is tier-limited because Enterprise access requires contacting sales.
Charge or no-charge status, currency, recurrence, and material conditions are disclosed.
Evidence and actions4 signals
Can an agent construct checkout, hand off approval safely, and resume?
supportedCurrent admissible evidence does not resolve this finding.
Documentation notes manual credit additions and auto top-ups. It does not document deterministic agent checkout construction, safe approval handoff, or programmatic resumption.
Is the exact commercial commitment disclosed before authorization?
supportedCharge or no-charge status, currency, recurrence, and material conditions are disclosed.
OpenRouter discloses US dollar pricing, credit replenishment rules, and a one-year credit expiration window. It also discloses pay-as-you-go platform fees of 5.5% for Standard and 8% for Business.
Can payment be authorized within scoped agent or explicit human authority?
supportedCurrent admissible evidence does not resolve this finding.
First-party sources list accepted payment methods including credit cards, AliPay, and cryptocurrency. They do not document a scoped delegation rail or human-confirmed authorization workflow.
Is a direct self-service path to paid access documented?
supportedSelf-service is tier- or condition-limited, or a bounded vendor review exposes status.
OpenRouter enables direct self-service pay-as-you-go onboarding and credit top-ups for Standard and Business tiers. Self-service is tier-limited because Enterprise access requires contacting sales.
What would improve this stage
- Document a direct purchase path or bounded vendor review with machine-readable status.
Provisioning is machine-triggerable or follows deterministically from an allowed handoff.
API key provisioning is machine-triggerable via a POST request to /api/v1/keys using Bearer token authorization.
Credentials, grants, or configuration are delivered through a documented agent-usable flow.
A synchronous success response returns a usable created resource or access material; otherwise asynchronous work exposes progress, terminal state, reconciliation, and a documented bound.
Evidence and actions3 signals
Can usable access material be delivered securely to an authorized agent?
supportedCredentials, grants, or configuration are delivered through a documented agent-usable flow.
OpenRouter delivers usable API keys via a programmatic creation endpoint. The plaintext key is returned directly in the synchronous JSON response under the key field.
Can an agent determine successful provisioning completion and reconcile asynchronous failure?
supportedA synchronous success response returns a usable created resource or access material; otherwise asynchronous work exposes progress, terminal state, reconciliation, and a documented bound.
The API key creation endpoint returns a synchronous 201 response containing the usable API key and key metadata.
Can provisioning be initiated within supported agent authority?
supportedProvisioning is machine-triggerable or follows deterministically from an allowed handoff.
API key provisioning is machine-triggerable via a POST request to /api/v1/keys using Bearer token authorization.
Safe failure handling is documented only partially.
OpenRouter defines standard error response formats, HTTP status codes, and normalized provider error types across streaming and non-streaming modes.Full reconciliation and safe idempotency contracts are only partially documented.
Only part of the credential lifecycle is agent-operable.
An agent-native interface is documented but no protocol exchange was verified.
Every essential target has a stable documented agent-usable interface alternative.
Request-time authentication is documented and usable under scoped agent authority.
Essential operations have stable readable request, response, and effect semantics.
Evidence and actions6 signals
Does a declared service interface respond through a verified agent-native protocol?
supportedAn agent-native interface is documented but no protocol exchange was verified.
OpenRouter documents an MCP server interface providing tools for live model discovery and chat inference.Admissible evidence does not verify a live protocol exchange.
- inference-api
- model-catalog-api
- openrouter-mcp
Can an agent manage credential expiry, rotation, revocation, compromise, and recovery?
supportedOnly part of the credential lifecycle is agent-operable.
OpenRouter provides API key creation with configurable expiration and spending limits, alongside key deletion endpoints.Delegated recovery and compromise remediation workflows remain only partially operable.
Can an agent handle applicable failure modes safely?
supportedSafe failure handling is documented only partially.
OpenRouter defines standard error response formats, HTTP status codes, and normalized provider error types across streaming and non-streaming modes.Full reconciliation and safe idempotency contracts are only partially documented.
Can an agent authenticate service operations with scoped authority?
supportedRequest-time authentication is documented and usable under scoped agent authority.
OpenRouter operations authenticate via Bearer tokens in the HTTP Authorization header.API keys can be provisioned with explicit spending limits and scoped workspace attribution.
Are essential operation inputs, outputs, and effects stable and decidable?
supportedEssential operations have stable readable request, response, and effect semantics.
Model catalog discovery returns structured JSON with model parameters, pricing, and context limits.Inference endpoints define explicit schemas for message histories and tool call payloads.
- inference-api
- model-catalog-api
Does every essential assessment target have a documented agent-usable interface?
supportedEvery essential target has a stable documented agent-usable interface alternative.
The service exposes machine-operable HTTP interfaces for both model catalog discovery and chat completion inference.Each essential assessment target has documented endpoints accessible to agents.
- inference-api
- model-catalog-api
What would improve this stage
- Document applicable structured errors, retry, cancellation, and reconciliation semantics.
- Provide scoped expiry, rotation, revocation, compromise, and recovery operations.
What works
- Evaluate: The exact service and stable evaluation or access entrypoints are publicly discoverable.
- Provision: Provisioning is machine-triggerable or follows deterministically from an allowed handoff.
What blocks agents
No stage is established as blocked.
Where autonomy is limited
- Sign up: Only part of the access flow is operable or a bounded human step remains.
- Pay: Self-service is tier- or condition-limited, or a bounded vendor review exposes status.
- Operate: Safe failure handling is documented only partially.
What this grade establishes
11 of 11 graded signals have current supporting evidence. 4 of 8 additional barrier checks are verified.Unknown checks are not a pass or a confirmed blocker. This report is not proof of an end-to-end purchase or service session.
How the grade is derived
A through C grades count Limited stages; D and F reflect Blocked stages by lifecycle severity.
Every graded signal needs fresh, supported evidence. Unknown or stale barriers remain visible and cap A or A+ at B+. Only fresh, supported barriers affect stage outcomes.
Each stage takes its worst graded signal or verified barrier. A fresh, supported mandatory barrier that passed its admission evidence rule can block that stage; incomplete or stale barrier evidence remains context only. The overall grade is derived from the five stage states.
Startup offers
OpenRouter on the startup-offers board
- Up to $5,000 in creditsOpenRouter for Startups
Distribution
The grade’s badge
One stable image resolving the record’s current live facts, linking to the canonical page. Embed it once; it keeps itself current. Placement is never evidence and never affects rank.
[](https://sourcey.com/c/openrouter/agent-readiness/openrouter-api/model-access-lifecycle)Improve the report
Correct it, rerun it, or improve the funnel.
- Profile
- arp_01m0j1a50vat933f76xwy3mec5
- Projection
- sha256:f50e6a0c9d5e9b406262033733e3d5c93ab3bb01a0d167c1580295379c4fb7b5
- Policy
- service-use-2026-10-05-protocol-proof · sha256:ce48785812cee754688f4b03fed7e3e2a6eaeaf3af35a57f85ed34fb93759db5
Every request carries these exact identifiers.How report cards are assessed →View the exact declaration