// sdk & install

Drop one file in.
Optimize every AI call.

The PromptForge SDK is two TypeScript files. No build step, no lock-in. Auto-routes to the cheapest model, tracks every dollar, and stops runaway spend.

Auto model routing

Every call automatically picks the cheapest model that meets your quality floor. No code changes needed.

Per-call cost tracking

Exact cost logged for every request. Aggregated by feature, model, and provider.

Budget alerts

Set session, daily, or monthly limits. Get notified — or stop execution — before you overspend.

Dashboard sync

All spend automatically appears in your PromptForge dashboard. One view across all projects.

// one-line install

Or install in any project with one command

Run this from your project root. Downloads the SDK, config, CLAUDE.md, and installs peer deps.

curl -fsSL https://promptforge-ebon.vercel.app/sdk/install.sh | bash

// quick start

Or step-by-step in 4 steps

01Install peer dependencies
npm install ai @ai-sdk/anthropic @ai-sdk/openai @ai-sdk/google
02Download the two SDK files
# Download from your dashboard or copy from below
# promptforge-sdk.ts
# promptforge.config.ts
03Set your env vars
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
GOOGLE_GENERATIVE_AI_API_KEY=AI...
PROMPTFORGE_API_KEY=your-key  # optional, enables dashboard sync
04Replace your AI calls
import pf from "./promptforge-sdk"

// Before:
const { text } = await generateText({ model: anthropic("claude-3-5-haiku"), prompt })

// After:
const result = await pf.text(prompt, { feature: "my-feature" })
console.log(result.text, result.costUsd)

// usage

Full example with budget monitoring

agent.ts
import pf from "./promptforge-sdk"

// Start background cost monitor — logs every 30s
const stopMonitor = pf.startMonitor({ intervalMs: 30_000, reportTo: "console" })

async function runAgent(task: string) {
  // Cheap model for planning
  const plan = await pf.text(`Break this into 5 steps: ${task}`, {
    feature: "planning",
    prefer: "cheapest",
  })

  // Better model only where it matters
  const result = await pf.text(`Execute: ${plan.text}`, {
    feature: "execution",
    prefer: "balanced",
    minQuality: 85,
  })

  console.log(`Done — spent $${result.sessionSpend.toFixed(6)} total`)
  return result.text
}

await runAgent("Analyze Q1 sales data and write a summary report")

stopMonitor()
pf.printReport()
// ╔══════════════════════════════════════╗
// ║      PromptForge Spend Report        ║
// ╚══════════════════════════════════════╝
//   Total spend : $0.000312
//   API calls   : 2
//   By model:
//     gemini-2.0-flash          $0.000041  (1 calls)
//     claude-3-5-haiku-20241022 $0.000271  (1 calls)
//   By feature:
//     planning                  $0.000041  (1 calls)
//     execution                 $0.000271  (1 calls)

// downloads

Download the files

Sign in for a personalized config with your API key and project ID pre-filled.

// claude code

Works with Claude Code

Drop CLAUDE.md into your project root. Claude Code will automatically follow cost-optimization rules on every AI call it writes.

CLAUDE.md — add to your project root
# PromptForge Cost Rules

- ALWAYS use pf.text() instead of calling provider SDKs directly
- ALWAYS pass a feature label: pf.text(prompt, { feature: "name" })
- Use prefer: "cheapest" for drafts, summaries, and classification
- Use prefer: "best" only for final user-facing outputs
- Call pf.printReport() at the end of every agent session
Download CLAUDE.md

Ready to stop overpaying?

Sign up free — get your API key, project ID, and a pre-filled config in 60 seconds.

Get started free →