● query_cost: $0.0042

↑ synced 847 queries

✓ prompt_content: NEVER_STORED

trial_days: 14 remaining

Privacy-first · No prompts stored · 5-line SDK

Know the true cost
of your AI stack.
Every layer. Without the risk.

Per-query cost visibility across LLMs and infra — Anthropic, OpenAI, Azure, Redis, Qdrant, Appwrite. Your prompts are never captured, stored, or transmitted.

Start free trial — 14 days How it works

No credit card

Zero prompt capture

Works offline

costpilot — zsh

Whitelist sanitizer active — prompt_content, response: NEVER_CAPTURED

$284

LLM cost / mo

$41

Infra cost / mo

2,847

queries tracked

prompts stored

Privacy by design

Your prompts never
leave your machine.

CostPilot captures only what's needed for cost math: token counts, model names, latency. Prompt text, response content, and documents are dropped at the interceptor layer before any storage or transmission occurs.

Whitelist-based sanitizer

Only 14 explicitly safe fields are stored. Every other field is silently dropped — no override possible.

One-way hashed identifiers

User and session IDs are SHA-256 hashed with a per-account salt that never leaves your environment.

GDPR + SOC2-ready controls

Audit logging, rate limiting, AES-256 at rest. Full data export and deletion endpoints built-in.

Python

Node.js

# 5-line integration
import anthropic
from costpilot import CostPilotClient

cp = CostPilotClient(project="my-app")
client = cp.wrap(anthropic.Anthropic())

# Your code is unchanged ↓
response = client.messages.create(
    model="claude-sonnet-4-5",
    messages=[{"role": "user", "content": prompt}]
)

# What CostPilot stores:
# ✓  input_tokens: 4280
# ✓  output_tokens: 892
# ✓  cost_usd: 0.00426
# ✗  prompt → NEVER CAPTURED
# ✗  response → NEVER CAPTURED

How it works

From install
to insights
in 5 minutes.

No infra changes. No code rewrites. Drop it in, wrap your client, done.

# Python — 5 lines

from costpilot import CostPilotClient

cp = CostPilotClient(project="my-app")

client = cp.wrap(anthropic.Anthropic())

# Your existing code unchanged ↓

Step 01

Install & init

Run pip install costpilot then costpilot init. Your 14-day trial starts automatically — no card needed.

30 seconds to set up

Step 02

Wrap your client

One line: cp.wrap(anthropic.Anthropic()). Your existing call signatures are completely unchanged.

Zero code changes required

Step 03

Data flows — silently

Every LLM call is intercepted after the response. Token counts, cost, latency stored in local SQLite. Prompt text dropped. Nothing blocks your app.

0ms added latency

Step 04

Open your dashboard

Run costpilot serve to open Overview, Queries, Scenario, Services, and Migration pages — all fed from your local SQLite.

5 dashboard pages included

Features

Everything you need to
understand AI spend

Privacy guarantee

Your prompts never
leave your machine.

Whitelist sanitizer enforced at the interceptor layer — not a config option.

✗ NEVER prompt_content

✗ NEVER response_text

✗ NEVER rag_documents

✓ STORED input_tokens, cost_usd, latency_ms

Per-query tracking

$0.004

cost per query · exact, not estimated

Input tokens

$0.0026

Output tokens

$0.0009

Cache read

$0.0004

Multi-provider

Anthropic, OpenAI, Azure OpenAI — one SDK, unified view.

Anthropic

OpenAI

Azure

Scenario projections

Dial in user count and queries/day. Real measured averages power the math.

1,000 users × 5 queries/day

$1,840/mo

💡 Cache system prompt → save 34%

Pre-migration reports

Local Docker → Azure/AWS/GCP projection from 14 days of real usage. No guesswork.

Azure OpenAI$220/mo

Redis Cache$41/mo

Qdrant Cloud$38/mo

Total$323/mo

Infra tracking

Redis ops, Qdrant queries, Appwrite calls — alongside LLM costs.

Redis

Qdrant

Appwrite

Live pricing registry

3-layer accuracy: Azure Retail API → hardcoded rates → Claude-powered change detection.

last_verified: 2026-05-02

claude-sonnet-4-5: $3/$15 per M

gpt-4o: $2.50/$10 per M

Local-first, cloud-optional

All data is written to your local SQLite database first. Cloud sync is non-blocking and optional — a network issue never touches your observability stack.

SQLite encrypted AES-256-GCM

Why CostPilot

Built for cost governance,
not general observability

Other tools trace everything. CostPilot answers the question teams actually ask: what is my AI stack costing me, and why?

Exclusive

Privacy by design

Prompts never stored — not a setting, a product principle. Whitelist-only field capture.

Exclusive

Infra → cost causality

Redis pressure, Qdrant latency, Appwrite health correlated directly to token spend.

Exclusive

Local-first, zero friction

5-line install. Value in seconds — no cloud account, no proxy, no infra changes needed.

Exclusive

Migration planning

Project current usage cost on AWS, Azure, or GCP — before you migrate, not after.

Capability	CostPilot	Langfuse	Helicone	LangSmith	Phoenix
Prompt-free by default	Yes	Opt-in	Opt-in	Opt-in	Opt-in
Infra health → cost correlation	Yes	No	No	No	No
Local SQLite — no cloud needed	Yes	Self-host	No	No	Self-host
Cloud migration cost projection	Yes	No	No	No	No
Per-query cost breakdown	Yes	Yes	Yes	Yes	Yes
5-line SDK integration	Yes	More setup	Yes	LC-first	More setup
Full prompt tracing & evals	Not the goal	Yes	Partial	Yes	Yes

Langfuse tells you what happened.
CostPilot tells you what it cost — and what to do next.

Need deep tracing and prompt management? Langfuse is excellent. Need to govern AI spend, catch infra-driven cost spikes, and plan migrations? CostPilot is the sharper tool.

Developers

SDKs and documentation,
without the dead ends.

Open the in-dashboard docs for setup and reference material, or jump straight to the published SDK packages for Python and Node.js.

Documentation

Docs live inside the CostPilot dashboard.

Setup guides, product walkthroughs, and reference docs are available at the dashboard docs route, so the latest product experience and instructions stay in one place.

https://app.thecostpilot.dev/docs

Open documentation Start free trial

Dashboard docs include onboarding, usage guidance, and the latest product reference.

Python SDK

Install from PyPI

Use the published Python package for local-first cost tracking in your existing app stack.

pip install costpilot

View Python SDK

Package: costpilot

Node.js SDK

Install from npm

Use the published Node.js package for JavaScript and TypeScript integrations with the same CostPilot flow.

npm install @costpilot/sdk

View Node.js SDK

Package: @costpilot/sdk

Pricing

Free while we're in early access

CostPilot is currently free for everyone — 1 project per account. Paid plans are coming soon. Get notified first.

$0 · 1 project · no limits on tracking

Paid plans launching soon — Starter, Pro, and Enterprise. Enter your email to get notified.

Learn about upcoming plans →

FAQ

Common questions

The sanitizer uses a strict whitelist approach — only 14 explicitly safe fields (token counts, model name, cost, latency, hashed IDs) are ever stored. Everything else is silently dropped before any write to SQLite or transmission to the cloud. The whitelist is fixed and non-configurable: costpilot/privacy/sanitizer.py.

No. The interceptor writes to SQLite asynchronously after the LLM response is returned to your application. Your call latency is completely unaffected. Cloud sync is also non-blocking — a network issue will never surface to your application.

The SDK keeps capturing every call — your data never stops flowing. Only the dashboard windows change: history beyond 24 hours, scenario projections, migration reports, and CSV export lock behind a paid plan. Your data is already in your account from day 0, so upgrading instantly unlocks everything with no data movement needed.

Yes. The Enterprise plan includes a self-hosted license key. No data ever leaves your infrastructure. The license validates monthly with a 30-day offline grace period — a connectivity issue won't take down your observability stack.

Phase 1 supports Anthropic (all Claude models), OpenAI (GPT-4o, GPT-4o mini, embeddings), and Azure OpenAI. LangChain, LangGraph, Mistral, Gemini, and others are on the Phase 2 roadmap. Because CostPilot wraps the provider SDK directly, it works with any LangChain setup that calls the underlying anthropic or openai client.

Only aggregated, non-sensitive metrics (token counts, model name, cost, latency, hashed user IDs). The cloud dashboard is optional — the SDK writes to local SQLite first and syncs asynchronously. You can run entirely local with no cloud account if needed (self-hosted Enterprise license).

Know the true costof your AI stack. Every layer. Without the risk.

Your prompts neverleave your machine.

Whitelist-based sanitizer

One-way hashed identifiers

GDPR + SOC2-ready controls

From installto insightsin 5 minutes.

Everything you need tounderstand AI spend

Built for cost governance,not general observability

SDKs and documentation,without the dead ends.

Docs live inside the CostPilot dashboard.

Install from PyPI

Install from npm

Free while we're in early access

Common questions

Know what you're spending.Ship with confidence.

Know the true cost
of your AI stack.
Every layer. Without the risk.

Your prompts never
leave your machine.

From install
to insights
in 5 minutes.

Everything you need to
understand AI spend

Built for cost governance,
not general observability

SDKs and documentation,
without the dead ends.

Know what you're spending.
Ship with confidence.