Helicone request logged notifications on your phone.

Get a push notification for LLM requests logged by Helicone, with model, cost, tokens and latency, and a High-priority push when the provider returns an error.

Install in Studio See the setup

requestAI Monitoring

Tuesday 29 September

9:41

JustPushnow

Request logged

Get a push notification for LLM requests logged by Helicone, with model, cost, tokens and latency, and a High-priority push when the provider returns an error.

How it works

Three steps to your first push

STEP 01

Install the recipe

One click in Studio gives you a personal webhook URL.

STEP 02

Paste your URL into Helicone

Add it as a webhook for Request.

STEP 03

Get a push on your phone

With the text, sound and buttons from the recipe.

Setup

How to set up the Helicone request webhook

Click Install in Studio and sign in. The recipe is added to your account and you get a personal webhook URL.
In Helicone, open Webhooks and add a webhook with your webhook URL as the destination URL.https://••••••••/•••••••• your personal URL, shown after install
Set a Sample Rate and Property Filters so you only get the requests you care about (see the setup notes).
Save and send a test event.
Helicone sends this webhook for **every** logged request that passes its filters, so narrow it down:
**Property filters:** only send requests with a specific `Helicone-Property-*` header, e.g. tag the calls you care about with `Helicone-Property-Notify: true` and filter on it.
**Sample rate:** send only a percentage of requests.
Helicone's alerts (cost, error rate, latency) can only go to email or Slack, not to a webhook, so this recipe works per request.
Helicone signs requests with a `Helicone-Signature` header, but JustPush doesn't check it yet, so keep your endpoint URL private.

The code

What runs when the webhook arrives

Studio calls handleRequest(request) with the incoming webhook and sends the message it returns. It's yours after install; change anything.

transform.js
 1// Helicone → request logged
 2// Fires after Helicone logs an LLM request that passes your webhook's sample rate and
 3// property filters. Every matching request pushes, so filter hard (see INSTALL.md).
 4// Docs: https://docs.helicone.ai/features/webhooks
 5
 6// LLM calls often cost fractions of a cent: $0.0015, $0.12, $3.40.
 7function usd(value) {
 8    const n = Number(value)
 9    if (value == null || !Number.isFinite(n)) return null
10    return "$" + (n > 0 && n < 0.01 ? n.toPrecision(2) : n.toFixed(2))
11}
12
13// request_body / response_body arrive as JSON strings (or a "too large" notice).
14function parse(value) {
15    if (typeof value !== "string") return value && typeof value === "object" ? value : null
16    try {
17        return JSON.parse(value)
18    } catch {
19        return null
20    }
21}
22
23function handleRequest(request) {
24    const body = request.body || {}
25
26    // Guard: only Helicone request webhooks.
27    if (!body.request_id) {
28        console.log("Not a Helicone webhook — skipping")
29        return null
30    }
31
32    const meta = body.metadata ?? {}
33    const response = parse(body.response_body)
34    // Providers report failures as an "error" object in the response body.
35    const error = response?.error
36    const reason = typeof error === "string" ? error : error?.message
37
38    const model = body.model ?? "LLM"
39    const provider = body.provider ? ` (${String(body.provider).toLowerCase()})` : ""
40    const cost = usd(meta.cost)
41    const tokens = Number.isFinite(Number(meta.totalTokens)) ? `${meta.totalTokens} tokens` : null
42    const latency = Number.isFinite(Number(meta.latencyMs)) ? `${(Number(meta.latencyMs) / 1000).toFixed(1)}s` : null
43    const stats = [cost, tokens, latency, body.user_id ? `user ${body.user_id}` : null].filter(Boolean)
44
45    return {
46        title: error ? `❌ ${model} request failed` : `🤖 ${model} request`,
47        message: (reason ? `${reason} · ` : "") + (stats.join(" · ") || `Request ${body.request_id}`) + provider,
48        topic: "Helicone",
49        priority: error ? 1 : -1, // High for failures, Low for routine requests
50    }
51}

Payload

The Helicone request webhook

This is what Helicone sends to your URL for request. It is a sample, trimmed to the fields recipes use.

request · sample.json
 1{
"model": "gpt-4o-mini",
"user_id": "user-123",
"metadata": {
  "cost": 0.0015,
  "latencyMs": 1200,
  "totalTokens": 25,
  "promptTokens": 10,
  "completionTokens": 15
},
"provider": "openai",
"request_id": "550e8400-e29b-41d4-a716-446655440000",
"request_body": "{\"model\":\"gpt-4o-mini\",\"messages\":[{\"role\":\"user\",\"content\":\"Hi\"}]}",
"response_body": "{\"id\":\"chatcmpl-1\",\"object\":\"chat.completion\",\"choices\":[{\"message\":{\"role\":\"assistant\",\"content\":\"Hello!\"}}]}",
"request_response_url": "https://s3.example.com/helicone/550e8400?signature=abc"
16}

FAQ

Helicone request logged notifications: questions

Does this work on iPhone and Android?

Yes. Install the JustPush app from the App Store or Google Play and sign in. Every phone signed in to your account gets the notification.

Do I need to write code?

No. Install the recipe in Studio, paste your webhook URL into Helicone and you are done. The code is there if you want to change the text, the sound or the buttons.

Can I change what the notification says?

Yes. After install the recipe's code is yours. Edit it in Studio and test it against the sample payload before you save.

What does it cost?

JustPush is free for 30 days. After that it's $19.99 a year, or $39.99 once. There is no extra charge for recipes.

Ready when you are

Helicone on your phone in two minutes.

Install the recipe, paste one URL, done. Free for 30 days, no credit card required.

Install in Studio Download for iOS Download for Android