What is a function in Fetchply?

A function is one HTTPS endpoint you register with a chat agent. The model reads a description you write and calls the endpoint during a chat when a question would benefit from a live answer. The response feeds the reply the shopper sees.

How does the agent know when to call?

You write a plain-language description and a "call when" hint, and the model decides per turn. If a question can be answered from the crawled knowledge base, the model skips the call. If the question needs live data, the model fills the parameters you defined and triggers the endpoint.

Which HTTP methods can a function use?

GET and POST. With GET, parameters are sent as query string values. With POST, parameters are sent as a JSON body. Other methods are not supported, by design, to keep the surface small.

How are auth secrets stored?

Auth secrets (bearer tokens and API key header values) are encrypted at rest with AES-256-GCM using a server-side key. They are decrypted only at request time, never returned in API responses, never sent to the model, and never logged.

Can my function reach internal services?

No. Every outbound call goes through an SSRF guard. URLs must be HTTPS on port 443, and DNS resolution rejects loopback, link-local, cloud-metadata, and other private addresses. Redirects are not followed.

How is the response turned into a reply?

You choose between two response modes. Raw mode sends the entire response to the model and lets it write the reply. Template mode lets you write a sentence once and fill fields by JSON path, with array limits and prototype-pollution protection.

How many functions can I register?

Plan limits: one function on Starter, five on Pro, and twenty on Business. The free plan does not include Functions. Upgrade prorates automatically and you keep your existing configuration through the change.

Functions

Your chatbot,
backed by live APIs.

Register any HTTPS endpoint once, and your agent calls it mid-conversation for real-time answers. Order status from your own backend, live stock counts, account details: the agent pulls the data, then replies in plain language.

Get started free View pricing

Included on paid plans · 1 endpoint on Starter, 5 on Pro, 20 on Business

HTTPS only. Secrets encrypted at rest, never shown to the model.

Store Assistant

Fetchply AI

Ask about orders or stock

Brands answering faster, selling more, and supporting better with Fetchply

Fetchply is an AI sales and support agent for online brands. It answers pre-sale and post-sale questions across your website, WhatsApp, Instagram, Slack, and email, using only your own products, policies, and documents, then hands the conversation to your team with full context when a person is needed. See what that costs on pricing.

How it works

Three steps from your API to a

No middleware, no webhook relay, no SDK. Describe the endpoint once and the model handles every call from there.

Register your endpoint

Pick GET or POST, paste your HTTPS URL, name the parameters, and add auth if needed.

Tell the model when to call

Write a description and a call-when hint. The model decides per turn.

Publish and test

Fire a test call, preview the response, then publish. Live on your next turn.

Typical setup time:about 2 minutes

Get started free

Capabilities

Six things Functions handles automatically

Not a webhook relay. The model reads your description, decides per turn, calls your endpoint safely, and turns the response into an answer a customer understands.

Live answers from your backend

The agent fills the parameters you defined, calls your endpoint mid-conversation, and folds the response into a plain-language reply. No copy-pasted data to keep fresh.

Store Assistant

Fetchply AI

Where is order #6102?

custom_order_status200 OK

It shipped on May 20 via USPS and is out for delivery today.

The answer is as current as your API.

The model decides when to call

You write a description and a call-when hint. Questions your knowledge base already answers skip the call entirely, so your API only fires when live data is actually needed.

Store Assistant

Fetchply AI

What is your return policy?

Answered from the knowledge base, no call made

You can return eligible items within 30 days of delivery.

No wasted calls on policy questions.

Replies shaped your way

Raw mode hands the whole response to the model and lets it write the reply. Template mode lets you write the sentence once and fill fields by JSON path, for fully predictable wording.

Store Assistant

Fetchply AI

How many loyalty points do I have?

custom_loyalty_balance200 OK

You have 1,240 points, enough for a $12 reward.

Template mode never improvises.

HTTPS onlySSRF guardNo redirects

Locked-down requests

Every call is HTTPS on port 443, checked by an SSRF guard that rejects loopback, link-local, cloud-metadata, and private addresses. Redirects are never followed.

Encrypted at restUsed at request timeNever logged

Secrets stay secret

Bearer tokens and API key headers are encrypted at rest with AES-256-GCM and decrypted only at request time. They are never shown to the model, returned by the API, or written to logs.

Test callPreview responsePublish

Test before you publish

Fire a test call with sample parameters and preview exactly what the agent will see before shoppers do. Publish when it looks right; the function is live on the next turn.

Comparison

Trained content answers what was true. Functions answer what is true

A knowledge base is perfect for policies and product facts, but order status, stock, and balances change by the minute. Functions let the same agent reach your systems for those.

Question type	With FunctionsLiveAgent calls your API per turn	Static content onlyCrawled and imported data
Order status	Looked up live from your backend, per question	Stale the moment the order ships
Stock and pricing	Current counts and prices at answer time	Whatever was true at the last re-crawl
Account details	Fetched on demand with the parameters the model fills	Not answerable at all
Setup	Register the endpoint and describe it once	Copy-paste data, then keep re-pasting it
Keeping answers fresh	Your API is the source of truth	A manual chore that silently drifts
When it runs	The model decides per turn; known answers skip the call	Every answer limited to what was imported
Credentials	Encrypted at rest, never shown to the model or logged	No live systems, no credentials, no live answers

Stop answering “where is my order?” with a shrug.

One endpoint on Starter, five on Pro, twenty on Business.

Connect your first function

How do Functions work on a Fetchply agent?

A function is one HTTPS endpoint you register with your agent: a URL, a method, the parameters it takes, and a plain-language description of when to call it. From then on the model reads every customer question, decides per turn whether it needs live data, fills the parameters, and calls your endpoint. The response feeds the reply, either written freely by the model or through a template you control field by field.

The same agent keeps answering from your trained content everywhere it runs, and conversations land in the same shared inbox and analytics. Read the setup documentation, pair it with webhooks for outbound events, or check pricing.

Store Assistant

Fetchply AI

Can I still get Saturday delivery on order #4417?

custom_order_status200 OK

Yes. It leaves the warehouse tomorrow, so Saturday delivery is still on track.

And is the matching belt in stock?

custom_inventory_check200 OK

Three left in brown, size fits all. Want a link to it?

Ask about orders or stock

What makes it different

Built for the question only your systems can answer

The model decides, per turn.

There is no flow to build and no trigger phrase to guess. The model weighs every question against your description and call-when hint, skips the call when the knowledge base already has the answer, and fires it when live data would make the reply better.

Locked down by default.

HTTPS only, an SSRF guard that blocks private and cloud-metadata addresses, no redirects, and GET and POST only, by design. Auth secrets are encrypted at rest with AES-256-GCM and are never shown to the model, returned in responses, or written to logs.

Answers before you wire a call

Short, plain answers for store owners and developers weighing whether Functions fits their setup.

Plug your APIs into the
.

Register an endpoint, write a one-line description, and your agent starts pulling live answers from your own backend on its very next turn.

Get started free

Your chatbot,
backed by live APIs.

Included on paid plans · 1 endpoint on Starter, 5 on Pro, 20 on Business

HTTPS only. Secrets encrypted at rest, never shown to the model.

Brands answering faster, selling more, and supporting better with Fetchply

Question type

With FunctionsLiveAgent calls your API per turn

Static content onlyCrawled and imported data

Order status

Looked up live from your backend, per question

Stale the moment the order ships

Stock and pricing

Current counts and prices at answer time

Whatever was true at the last re-crawl

Account details

Fetched on demand with the parameters the model fills

Not answerable at all

Setup

Copy-paste data, then keep re-pasting it

Keeping answers fresh

Your API is the source of truth

A manual chore that silently drifts

When it runs

The model decides per turn; known answers skip the call

Every answer limited to what was imported

Credentials

Encrypted at rest, never shown to the model or logged

No live systems, no credentials, no live answers

How do Functions work on a Fetchply agent?

Your chatbot,backed by live APIs.

Brands answering faster, selling more, and supporting better with Fetchply