Debug LLMs. Ship the right model.
Observability tells you what you're spending. Promptive tells you if you're spending it on the right model — with logs, evals, and real cost per provider.
Overview
7-day calls
2,847
↑12%
Avg latency
1.4s
P95: 4.1s
Avg tokens/call
1,240
820 in · 420 out
Top Issues
Slowest Prompts
Most Expensive
Recent Logs
Last 5 calls — click a row to inspect
Log Everything, Automatically
Every LLM call is captured — latency, token count, cost, model, and the full input/output. Zero extra code after the one-time setup.
- Latency & p95 tracking
- Full prompt & completion capture
- Token counts per call
Logs
2,847 calls in view · filter, search, and inspect
| Prompt | Version | Model | Latency | In | Out | Cost | Time | |
|---|---|---|---|---|---|---|---|---|
| summarizer | v1.0.1 | haiku-4.5 | 423ms | 820 | 312 | $0.0008 | 2m ago | |
| classifier | v2.0.0 | sonnet-4.6 | 1.2s | 1240 | 88 | $0.0041 | 5m ago | |
| summarizer | v1.0.1 | haiku-4.5 | 5.8s | 840 | 298 | $0.0009 | 18m ago | |
| doc-extractor | v1.0.0 | opus-4.6 | 8.1s | 3100 | 720 | $0.082 | 1h ago | |
| classifier | v2.0.0 | sonnet-4.6 | 980ms | 1180 | 92 | $0.0039 | 1h ago |
Compare Prompt Versions
Tag each call with a prompt ID and version number. Metrics update in real-time so you can see exactly what changed between iterations.
- Side-by-side metric comparison
- Real-time version diff
- Regression detection
Prompts
4 prompts tracked
| Prompt | Source App | Versions | Calls | Avg Latency | Total Cost | Last Called | |
|---|---|---|---|---|---|---|---|
| summarizer | prod-api | v1.0.1v1.0.0 | 1,420 | 4.2s | $1.14 | 2m ago | |
Performance Avg latency4.2s Avg cost/call$0.0008 Total calls1,420 Total cost$1.14 Versions v1.0.1activelatest v1.0.0 Actions ↗ Open in Playground ⊙ Version Analysis → $ Cost Breakdown | |||||||
| classifier | prod-api | v2.0.0 | 890 | 1.1s | $0.94 | 14m ago | |
| doc-extractor | — | v1.0.0 | 312 | 8.1s | $1.62 | 1h ago | |
| chat-agent | staging | v3.0.0 | 225 | 2.3s | $0.22 | 3h ago | |
Hard Caps, Zero Surprises
Track spend per prompt and set hard monthly caps. When the limit is hit, calls stop gracefully — no overages, no bill shock.
- Cost per prompt breakdown
- Monthly hard caps
- Projected spend forecast
Cost Analytics
30-day spend
$3.92
2,847 calls
Avg per call
$0.0014
3 models
Daily burn rate
$0.13/day
This month avg
Proj. monthly
$3.92
18 days left
Daily Spend — Last 30 days
Cost by prompt version
| Prompt | Version | Calls | Total cost | Avg / call | % of spend |
|---|---|---|---|---|---|
| doc-extractor | v1.0.0 | 312 | $1.62 | $0.0052 | 42% |
| summarizer | v1.0.1 | 1,420 | $1.14 | $0.0008 | 29% |
| classifier | v2.0.0 | 890 | $0.94 | $0.0011 | 24% |
| chat-agent | v3.0.0 | 225 | $0.22 | $0.0010 | 6% |
Three Lines of Code
Add Promptive to any existing project in under a minute.
// 1. Install
npm install @promptive/sdk// 2. Wrap any LLM call
import { Promptive } from "@promptive/sdk";
const promptive = new Promptive({
apiKey: "YOUR_API_KEY",
});
const response = await promptive.wrap({
promptId: "my-prompt",
version: "v1",
call: () => openai.chat.completions.create({
model: "gpt-4o",
messages,
}),
});Common Questions
Everything you need to know before getting started.
Promptive is an LLM observability tool. Add three lines to your app and every prompt call gets logged — latency, token counts, cost, input, output — all in a searchable dashboard.
Install the SDK with npm install @promptive/sdk, create a project to get an API key, then wrap your LLM calls with promptive.wrap(). That's it — no proxy required.
Yes. Promptive works with any provider — OpenAI, Anthropic, Google, Mistral, local models. You pass the call result directly, so there's no SDK lock-in.
Console logs vanish. Promptive persists every call with structured metadata, lets you diff prompt versions, see cost trends, replay requests with copy-as-curl, and alert you when costs spike.
Your app keeps running — only logging stops for the rest of the month. No surprise shutdowns, no broken production calls. Upgrade to Builder for 25,000 calls/month.