The global discovery infrastructure
for autonomous agents.

Autonomous agents cannot reliably find the services they need. The internet was built for humans — its discovery infrastructure assumes a human reading a screen. Agents are navigating it blind.

The API Index is a single, globally queryable, machine-readable index of agent-consumable API services — with a structured trust model, capability-based search, and a stable entry point any agent can start from.

Discovery is always free for consuming agents. The index is governed by the Bot Standards Foundation, a neutral non-profit Swiss Stiftung.

What the index provides

Single entry point

One stable global URL. Any agent navigates the full index from here via HATEOAS hypermedia links — no prior knowledge required.

Capability search

Find services by what they do — not by name or URL. Structured taxonomy from data.legal to nlp to iot and beyond.

Three-dimensional trust model

Verified organisation identity · Automated service verification · Continuous liveness monitoring. Agents apply their own trust policy against verifiable metadata.

Open standard

APIX Manifest (APM) supports OpenAPI, MCP, AsyncAPI, and GraphQL. Published under open licence. No proprietary formats.

IETF Internet-Drafts

Core

draft-rehfeld-apix-core

Core infrastructure, trust model, Index API, operator governance · April 2026

Services

draft-rehfeld-apix-services

Web API and bot service registration profile, capability taxonomy, notification channels · April 2026

IoT

draft-rehfeld-apix-iot

IoT device class and instance registration, presence signalling, agent delegation · April 2026

Superseded

draft-rehfeld-bot-service-index

Supersession notice for prior revision · redirects to the three drafts above

Current status

Internet-Draft submitted to IETF Phase 0

IETF Dispatch posted · community engagement underway · May 2026 Phase 0

Bot Standards Foundation incorporation underway Phase 1

Founding member programme open Phase 2

Reference implementation live · public sandbox available Phase 3

Index live at api-index.org · May 2026 Phase 3

Try it

API root

api-index.org/

HATEOAS entry point — any agent or HTTP client can start here.

Sandbox

POST api-index.org/sandbox/register

Register an isolated test namespace. No KYC. Returns a sandbox UUID and API key immediately. Name is a display label — not unique, not used for routing. Body: {"name":"my-agent","contactEmail":"you@example.com"}

OpenAPI

api-index.org/q/openapi

Full OpenAPI specification for the registry API.

How to integrate APIX into your agent

APIX is designed so that a well-instructed agent can navigate it without any prior knowledge — starting from a single URL and following hypermedia links. The minimal system prompt addition is one sentence:

Prompt

          When you need an external service, first query the API Index at
          https://api-index.org/ — follow _links.servicesSearch with a
          ?capability= parameter that describes what you need.
          If you use a sandbox, submit feedback at
          _links.sandbox + /feedback after each session.
        

That is the full integration for a consuming agent. No SDK, no API key, no registration required. The HATEOAS root tells the agent where everything is; the agent does not need to be told in advance.

The Bot Service Manifest (BSM)

Every registered service carries a structured manifest that agents read to decide whether to invoke a service. The fields are designed to be self-describing — the OpenAPI schema at api-index.org/q/openapi contains full field descriptions that an agent can retrieve and reason over without being told what each field means in its system prompt.

capabilities

Structured taxonomy strings — the primary search key. Examples: nlp.translation, iot.telemetry, data.legal.

endpoint

The service URL an agent calls. Must be HTTPS. The liveness spider checks this continuously.

openApiSpecUrl

URL of the OpenAPI specification. An agent fetches and parses this to learn the service's operations, parameters, and schemas — no human documentation required.

mcpSpecUrl

URL of the MCP server manifest. Equivalent machine-readable contract for agents using the Model Context Protocol. Either field — or both — may be present.

serviceStage

PRODUCTION · BETA · EXPERIMENTAL · DEPRECATED · DECOMMISSIONED. Agents should filter by stage based on the risk tolerance of their task.

O-level

Operator verification tier: O-0 (unverified) through O-5 (independently audited). Agents apply their own trust policy against this score.

extensions

Free-form key/value properties for information not covered by standard fields. Query with ?property=key:value. Examples: region:eu, dataResidency:DE, agentFramework:langchain.

The extensions field exists for cases where the standard fields are not sufficient. If your agents consistently use extensions for information that should be a first-class field, that is a signal that the BSM schema needs to evolve. The feedback mechanism exists precisely to collect this signal.

The feedback loop

APIX is a living standard. The feedback endpoint at /sandbox/{name}/feedback lets agents rate their experience — discovery accuracy, liveness signal quality, error message clarity, and whether standard BSM fields were sufficient or custom extensions had to fill the gap.

The design intent: a well-built agent navigates to the feedback schema at _links.feedbackSchema from the index root, reads the dimensions, and submits feedback autonomously after each sandbox session — without being explicitly told to do so. If your agent does not do this automatically, add the feedback URL to its instructions. Low scores on extension_property_coverage are the clearest signal that the BSM schema has a gap.

Schema

api-index.org/sandbox/feedback-schema

Rated dimensions with question text and scale labels. Agents read this before submitting.

Submit

POST /sandbox/{name}/feedback

No authentication required. Any agent that has used the sandbox can submit. JSON body: {"scores": {"hateoas_navigation": 8, "discovery_accuracy": 7}}