---
title: "Caveman Cloud"
description: "Run the agents you ship for less. Caveman Cloud shows what every agent costs, finds the waste, and tests each fix on your recorded traffic before it goes live."
canonical: https://caveman.so/cloud
last-updated: 2026-11-21
---

# Caveman Cloud

Run the agents you ship for less.

Caveman Cloud shows what every agent costs, finds the waste, and tests each fix
on your recorded traffic before it goes live.

[Start for free](https://app.caveman.so/signup?next=https%3A%2F%2Fplatform.caveman.so%2F),
or [see demo](https://caveman.so/demo). Free for one person, no card needed.

## Caveman Cloud

Point your SDK at Caveman Cloud and keep your provider keys. Every call becomes
a trace, priced and checked for waste. [Try the demo](https://caveman.so/demo).

- **See what each agent costs.** Spend split by agent, model and day, across
  every provider.
- **Traces every request.** Look up any call your agents made, and what it
  cost.
- **Finds the fix.** It spots the waste and proposes a change, with what it
  would save. It tries the fix on your old requests first.
- **Tests every change.** Your recorded requests become tests, and every change
  is graded against what runs today.
- **Works with what you run.** Keep your stack and your provider keys. Every
  call becomes a trace.
  - Agents you ship: point your SDK at Caveman. OpenAI SDK, Anthropic SDK,
    Vercel AI SDK, LangChain.
  - The gateway you run: keep LiteLLM. Its OpenTelemetry exporter sends the
    traces.
  - Coding agents: every session, with spend per person and per agent. Claude
    Code, Codex, Gemini CLI, OpenCode.

## Governance

Set what your agents may spend and send, and keep a record of who changed what.
Your coding agents talk straight to your provider on your own key, and
Enterprise can run in your own cloud.

- **Budgets and guardrails.** Caps per project, limits per key and workflow,
  and rules for what should never reach a model.
- **An audit log for every change.** Budgets, guardrails, keys and roles: who
  changed what, and when.
- **Your data stays yours.** Your coding agents talk straight to your provider.
  Caveman gets the routing ask, never the request.

## Enterprise

Run Caveman Cloud in your own VPC, with your keys and your storage, or let us
host it.

- **Runs in your own cloud.** One Helm chart on your Kubernetes, inside your
  own cloud account, on AWS or GCP. The same images as hosted Cloud, signed, so
  you can verify them before you install.
- **Your data stays with you.** Keys, traces and artifacts stay in your
  account. Enterprise never shares your content.
- **Priced on spend, not seats.** One annual contract, for Cloud hosted or in
  your own cloud. Priced on the AI spend Caveman manages, never per seat; 100%
  of what Caveman saves you is yours; SSO and SCIM included.

Questions about security, deployment or the contract?
[Security and deployment](https://caveman.so/enterprise) ·
[Book a call](https://cal.com/caveman/chat).

## Pricing

Caveman Cloud is for the agents you ship to your customers. Free is for one
person and needs no card. Team covers your whole team, with no seats.
[See full pricing](https://caveman.so/pricing?product=cloud).

- **Free**, for one developer trying everything Caveman does. $0 a month, no
  card needed. Observability and evals for one person, 100,000 events a month,
  600 agent minutes a month, $10 of Caveman AI a month.
  [Start free](https://app.caveman.so/signup?next=https%3A%2F%2Fplatform.caveman.so%2F).
- **Team**, for teams that want agent costs cut, with proof. $300 a month,
  including $300 of usage every month. Your whole team with no seats, standard
  rates past the included usage, $100 of Caveman AI a month, semantic cache
  plus guardrails, budgets and an audit log.
  [Try Team free for 7 days](https://app.caveman.so/signup?next=https%3A%2F%2Fplatform.caveman.so%2F).
- **Enterprise**, for orgs that keep data in their own cloud. Your VPC on AWS
  or GCP, or hosted; one Helm chart on your Kubernetes; your KMS and storage;
  signed releases you can verify before install; never per seat, priced on the
  AI spend Caveman manages. [Book a call](https://cal.com/caveman/chat).

## Questions

### What does Caveman Cloud do, and who is it for?

Caveman Cloud is for companies that run agents for their customers: it shows
what every agent and workflow costs and records every request as a trace. It
reads your traffic for repeated work, bloated context and the wrong model for
the job, and prices each finding with a proposed change and a way to test it.
Your team decides what ships.

### How do I connect my agents?

Point your SDK at Caveman and keep your own provider keys; the OpenAI,
Anthropic and Vercel AI SDKs and LangChain all work this way. If you run
LiteLLM, keep it: its OpenTelemetry exporter sends the traces. Cloud also
imports the logs and traces you already have, and coding agents your team runs
can be recorded too.

### Do you mark up model tokens, and what do you bill for?

No. Token markup is 0%: model calls run on your own provider keys, and your
provider bills you directly. Caveman bills for events, agent minutes and
routing decisions, plus the Team fee and any package. One event is one recorded
LLM call, span or eval score; a call through the Caveman gateway counts one
event per started 64 KB it sends.

### What happens when I reach my allowance or billing limit?

Caveman stops recording that product until the month resets and never charges
past the limit. Gateway calls keep going to your provider unchanged; only their
traces are not written. On Free without a card, the allowances are the limit.
Add a card to go past them at standard rates, up to a $50 limit per product
unless you set your own.

### How do I know a saving is real?

A saving counts as verified only once your production traffic proves it, and
every figure says whether it is measured, estimated or verified. Spend is
measured at catalog prices, failures and retries included. A finding is
estimated, with a daily range, a sample size and a confidence, and a proposed
change runs against your recorded work and your scorers before it ships. No fee
is ever calculated from a savings figure.

### What does Caveman Cloud store, and can I turn it off?

It stores usage metadata (model, provider, token counts, cost, latency and
status) for every request within your limits, and request and response bodies
by default, redacted and encrypted, though redaction will not catch everything.
On Team and Enterprise an admin can turn body storage off, and on every plan a
request marked zero-retention keeps only its usage metadata. Data is kept until
you delete it or an admin sets a window.

### Will you train on my data?

Only on Free, unless a Team admin turns it on: on Free, Caveman may use your
prompt text and recorded requests to improve its models. That is a condition of
the Free plan, agreed at sign-up. Team has it off by default, and an admin can
turn it on. Enterprise never does, and it cannot be turned on.

### Can Caveman Cloud run in our own cloud?

Yes, on Enterprise. It installs one Helm chart and the same signed images as
hosted Cloud on your Kubernetes, on AWS or GCP, and your keys, KMS and storage
stay in your account. It still keeps data to run the platform, so it is not a
zero-data-retention service. If you prefer, we host it.

### Do I need Caveman Code to use Caveman Cloud?

No. One account holds both, and each has its own plan: a Cloud plan for the
agents you ship, and a Caveman Code plan for the coding agent. Neither needs
the other, and Cloud usage and Code credits are billed separately.

## Start with Caveman Cloud

[Start for free](https://app.caveman.so/signup?next=https%3A%2F%2Fplatform.caveman.so%2F),
or [book a call](https://cal.com/caveman/chat). Free for one person, no card
needed.
