Learn AI /Data and Infrastructure /Endpoint
Endpoint
- Intermediate
- Data and Infrastructure
- Technical readers
Also called: API endpoint.
Back to the milestone you came from
In five seconds
What it is
A network address where software sends requests to a service and receives responses according to an API's rules.
In plain words
Think of an endpoint as the exact service doorway your software uses for one job. You send a request in the required format, and you get back a reply in the required format.
An everyday example
A railway station may have a ticket machine, an information desk and a lost-property counter. They belong to the same station, but each accepts a different kind of request and follows its own procedure. To buy a ticket, you use the ticket machine, enter the required journey details, provide an accepted form of payment and receive a ticket in a standard format. Taking the same request to the lost-property counter will not work, even though you are in the right building. The endpoint is the specific counter and procedure; the API is the wider set of counters and rules.
A real endpoint is not a person and does not manually inspect each request. Many endpoint checks and routing steps happen automatically in milliseconds. Also, one endpoint can sometimes trigger several internal services, while the weather-office window suggests a simpler one-step process.
Remember thisAn endpoint is the specific address and operation used to request one service function. It is not the whole API, the deployment or the hosted model.
Understand it
Understand it
Why it matters
If you use or buy AI services, the requirements and behaviour of the service accessed at an endpoint are where cost, speed, security, reliability and governance often become visible. They affect what data format you must send, what authentication is required, what errors you must handle, which limits apply and sometimes where processing happens. For non-specialists, this matters when reviewing vendor claims, planning integrations, assessing privacy and data sovereignty risks, or debugging why a system works in testing but fails in production. Useful questions include: Which endpoint does this function use? What data leaves our system? What logs are kept? What happens if the service is unavailable or rate-limited?
How an endpoint turns a request into a structured response
After viewing this, the reader should be able to explain that an endpoint is the specific callable entry point for one service function, not the whole API or the model itself.
01
1. Caller sends request
An application sends input data, settings and credentials to a specific network path for one task, such as text generation or embeddings.
Draw a box labelled 'Client application' with a packet icon. Arrow to the next box. Under the arrow, add small labels: 'input', 'settings', 'credentials'.
02
2. Endpoint checks the request
The service verifies format, permissions and limits. If something is wrong, it returns an error instead of continuing.
Draw a box labelled 'Endpoint'. Inside, stack three mini-check labels: 'validate', 'authenticate', 'rate limit'. Add a downward branch to a small red box labelled 'Error response'.
03
3. Backend does the work
If the request passes checks, the endpoint forwards it to the hosted model or another backend component that performs the requested task.
Draw a box labelled 'Backend service or hosted model'. Use a gear or chip icon. Main arrow continues from the endpoint to this box.
04
4. Structured response returns
The system sends back a defined response format, often including the output, status and usage metadata.
Draw a box labelled 'Response'. Inside, show three lines: 'output', 'status', 'usage metadata'. Arrow from backend to this box.
A left-to-right flow diagram with four boxes and arrows. Read from the caller’s request on the left to the structured response on the right. Each stage shows one causal step: send, check, route, return. A side note under the second box shows that invalid or unauthorised requests can stop there with an error.
Hovering or tapping a box reveals a short tooltip with an example, such as a prompt in step 1 or a 429 rate-limit error in step 2.
Describe this without the diagram
The diagram explains an endpoint as a four-stage flow. First, a client application sends a request containing input, settings and credentials to one specific service location. Second, the endpoint checks whether the request is valid, authorised and within usage limits. If not, the flow can stop and return an error. Third, if the checks pass, the endpoint forwards the request to the backend service or hosted model that performs the task. Fourth, the system returns a structured response that may include the result, a status code and usage details. The main point is that the endpoint is the controlled entry point, not the whole API and not the model itself.
With reduced motion: Show the same four boxes and arrows as a static horizontal flow, with a fixed side branch from the endpoint box to an error box. No animation; emphasis is shown with colour and line weight only.
An endpoint is a specific network location to which a program sends a service request. It often supports a particular operation, but one endpoint can support several operations. In practice, the location is usually a URL, while the API contract defines the request and response formats. For example, a service may expose locations for creating text, listing available models or checking usage.
What goes in depends on the service. A request often includes input data, settings and credentials, which the service uses to authenticate the caller. The service receiving the request may validate it, check permissions, apply limits, route it to a hosted model or another backend service, and record usage for monitoring or billing.
What comes out is a response in a defined structure, often JSON, a common machine-readable text format. The response may contain model output, an error message, timing information or metadata such as token counts. A well-designed service gives predictable structure even when the answer content varies.
People often confuse an endpoint with an API. An API is the broader interface, including the rules and available operations. An endpoint is a network location used to access that interface. People also confuse an endpoint with the model itself, but the endpoint is only an access point. The actual model may run elsewhere on different hardware or in a different region.
How “The right counter at a railway station” maps onto the real thing
| In the story | In the AI system |
|---|---|
| Forecast service window | Endpoint for one specific function |
| Standard form with location and dates | Structured request with input fields and settings |
| Clerk checking the form | Validation, authentication and rate checks |
| Forecast printout in a standard layout | Structured response such as JSON |
A worked example
A programme officer tests the drafting integration before circulating the side-event concept note under silence procedure.
Starting from: A POST request to the configured drafting endpoint. Its headers carry the content type and an access token, while its JSON body contains cleared, non-confidential background text and an instruction to produce a short summary.
- The form sends the request to the drafting endpoint, rather than to the service's usage or model-information endpoints.
- The endpoint checks the access token, validates the request body against its schema and confirms that the caller is within the applicable rate limit.
- If the request is authorised and valid, the service routes it to the hosted model and applies the configured logging and output controls.
- The endpoint returns a structured response containing the draft, usage metadata and a success status. Invalid credentials or fields produce the defined error response instead.
Result: The form places the returned text in the concept note template. The programme officer reviews it for accuracy and tone before circulating the file. If the endpoint returns an authentication or schema error, the interface displays that error rather than treating it as an empty draft.
In a United Nations workplace
At close of business, a chief of section supporting a UN Country Team working group is clearing the concept note for a side event in the margins of the session. The programme officer has connected the drafting form to an approved internal AI service, but the form is returning an authentication error.
The error concerns the endpoint, the precise network address and request method used for the drafting operation. The service may expose separate endpoints for drafting, usage information and other operations. Checking the drafting endpoint's required headers, credentials, request fields and error codes is more useful than saying vaguely that the AI is down.
Before you try thisOnly the approved endpoint may be used. The concept note must not be sent to a public substitute when clearance is due, and confidential positions, personal contact details or access credentials must not be included in the request. Endpoint logs can retain request content, so data minimisation and the office's access and retention controls still apply.
Where you may hear it
- An ICT governance board reviewing an integration architecture
- A procurement unit's technical evaluation during an AI vendor demo
- A service desk call after a drafting tool returns authentication errors
- A design session before an application is uploaded to the portal
Avoid confusion
Avoid confusion
Often confused with
API
An application programming interface defines rules that software uses to request data or actions from another system.
Model hosting
The systems and services that keep an AI model available so it can receive requests, run on computing hardware, and return outputs reliably and securely.
Deployment
Deployment means putting an AI system into real use so people or software can send it live inputs and receive live outputs.
Common misconceptions
Common misconception: An endpoint and an API are the same thing.
Correction: An API is the broader contract: the collection of operations, rules and formats a service exposes. An endpoint is one specific callable location or operation within that contract. A single API can have many endpoints for different tasks.
Common misconception: The endpoint is the AI model.
Correction: The endpoint is the access point, not the model itself. It may forward the request to a model hosted elsewhere, add security checks, enforce limits, transform formats, or combine several backend services before sending a response.
Common misconception: If I know the endpoint address, I can always use it.
Correction: Most endpoints also require valid credentials, correct request structure, allowed network access and compliance with rate or quota limits. Knowing the address alone does not grant permission or guarantee success.
Quick check
Which statement best describes an API endpoint in an AI service?
An endpoint is one entry point for one operation, such as generating text or listing models. The full set of operations is the API. The endpoint may connect to infrastructure elsewhere, and using it does not guarantee factual or approved output.
Go deeper
Go deeper
The simple version
Think of an endpoint as the exact service doorway your software uses for one job. You send a request in the required format, and you get back a reply in the required format.
The technical version
In web services, an endpoint is a concrete network-reachable interface for a specific operation, usually identified by a URL path and HTTP method such as POST /v1/responses. It is part of a larger API contract and defines how a client invokes that operation: expected headers, authentication scheme, request schema, response schema, status codes and error behaviour. In AI systems, an endpoint often sits in front of model hosting infrastructure and performs request validation, authorisation, rate limiting, routing, observability and billing before or after calling the underlying model service. The endpoint may expose synchronous or asynchronous behaviour, stream partial outputs, or proxy requests to region-specific backends for governance or performance reasons.
Reviewed external resources
Primary sourceAI Risk Management Framework (opens in a new tab)NIST
This framework helps readers think about risk, governance and controls around service interfaces and deployments, so it suits technical managers and assurance-focused practitioners.
Intermediatelink checked 2026-07-31
Primary sourceOECD AI Principles (opens in a new tab)OECD
These principles give high-level guidance on trustworthy AI governance, which helps readers place endpoint choices in a broader organisational context and suits policy and programme staff.
Intermediatelink checked 2026-07-31
EncyclopaediaWikipedia search: Endpoint (opens in a new tab)Wikipedia
This search-results page helps readers see that 'endpoint' has multiple meanings across computing, so it suits readers who need broad orientation before focusing on API usage.
search results, not one pagelink checked 2026-07-31
VideoVideo explanations: Endpoint (opens in a new tab)YouTube search
This search-results page offers a range of video explanations with different technical depths, so it suits readers who learn better through demonstrations and walkthroughs.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Endpoint (opens in a new tab)Reddit search
This search-results page shows community questions and practical troubleshooting discussions, so it suits readers who want informal examples and common real-world confusions.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
An endpoint is the specific address and operation used to request one service function. It is not the whole API, the deployment or the hosted model.