Anthropic

What each plan can show

Anthropic publishes billing data for the Claude API and for Claude Enterprise only. For every other plan there is no bill to read, so LevelFour shows the usage that Claude Code and Cowork report themselves: who used which model, how much, what that usage is worth at API list price, and how developers work with Claude Code.

Connect any mix of the three sources, in any order. Billing data holds no activity, so add Claude Code and Cowork usage beside a Claude API or Claude Enterprise connection to see sessions, tools and output next to what you pay.

A source can hold more than one account: a second Claude API organization, a second Claude Enterprise organization, or a settings block per team. Add another at any time from the same page. Name is optional free text, and LevelFour uses it only to tell accounts apart in the connected accounts list.

On the Claude API, the account is the Claude Console organization. A second Admin key of an organization you already connected adds no account: it ends as "already connected".

LevelFour does not ask you to type what a subscription costs. A subscription is the same every month whatever is used, so it has nothing to show day by day. Anthropic bills usage credits beyond a subscription at API rates, and publishes no report LevelFour can read for them.

History and what is left out

Prerequisites

Connect

Open Connect Providers and select Anthropic. The first time an admin opens it, Connect Claude starts by itself; after that, click Add account. Pick each plan your company pays for and click Continue:

Two plans with the same source share one step. You can skip a step or hand it to a colleague, and a review at the end lists what is connected and what is still open. Run it again at any time to add a second organization or another team: it only adds accounts. Only admins connect, rotate and disconnect; a member or viewer sees each account and its status.

{
  "env": {
    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
    "OTEL_LOGS_EXPORTER": "otlp",
    "OTEL_METRICS_EXPORTER": "otlp",
    "OTEL_EXPORTER_OTLP_PROTOCOL": "http/json",
    "OTEL_EXPORTER_OTLP_ENDPOINT": "https://api.levelfour.ai/api/v1/otlp",
    "OTEL_EXPORTER_OTLP_HEADERS": "Authorization=Bearer l4i_...",
    "OTEL_LOGS_EXPORT_INTERVAL": "60000",
    "OTEL_METRIC_EXPORT_INTERVAL": "60000"
  }
}

What LevelFour keeps from Claude Code and Cowork

Claude Code and Cowork report each thing a developer does as an event or a counter. LevelFour adds it to a daily total and stores the total, never one row per event.

LevelFour keeps usage, sessions, active time, prompt counts, API errors, lines of code, commits and pull requests per person. Every other total covers all developers together and carries no name.

Each person has one profile: the Claude Code version, terminal and operating system they last reported, their Claude organization ID, whether managed settings applied, and the first and last day they reported.

What LevelFour never reads

LevelFour never reads or stores:

• the text of a prompt or a response
• a tool's input or parameters
• a command
• file content
• an error message

Claude Code redacts prompt and response text by default. If you turn that text on in Claude Code, LevelFour does not read it.

On a Team or Enterprise plan, Cowork sends prompt text, response text and tool input with its events by default, along with each person's email. LevelFour keeps the email to tell people apart and never stores the text. Anthropic's Cowork monitoring guide says how to stop Cowork sending the text.

Names Claude Code redacts by default

LevelFour shows them as they arrive. An official plugin, and a skill it ships, arrives under its real name.

Repository names and team tags

Claude Code sends no repository name and no team until you turn them on. The block LevelFour generates holds neither, so add them yourself to the env object that holds the block.

Each card appears once data arrives with a repository or a team on it. LevelFour stores a repository as owner/name.

Sending through an OpenTelemetry Collector

You can put an OpenTelemetry Collector between Claude Code and LevelFour. Have it export OTLP over HTTP as JSON (http/json), to the endpoint and with the header from the block. LevelFour refuses protobuf, a batch over 2 MB, and a batch that holds more than 1,024 usage events or more than 1,024 other events or data points.

What you see

Anthropic spend appears under Costs and Usage and on its own Anthropic board under Boards, which appears once a source is connected.

Billed and usage value

LevelFour keeps two figures apart, and a switch on the chart and the table chooses between them.

• Billed is what Anthropic charges you through the Claude API and Claude Enterprise.
• Usage value is Claude Code and Cowork usage priced at API list price. It estimates what that usage would cost per token, and it shows who uses Claude and how much. LevelFour never adds it to a billed total.

On a Team, Pro or Max plan there is no billed spend in LevelFour, and the usage value shows how the subscription is being used.

The Anthropic board

The board sorts its cards into groups, each under a heading. A card appears once it has data, and a heading appears once its group has a card.

Bill reads the Claude API and Claude Enterprise. Every other group reads what Claude Code and Cowork report, as it arrives and with the current day included. Under Costs and Usage, the usage value is as of the last daily sync.

Some cards need two or more people who have reported usage: with one person they have nothing to compare. The Needs column names them. The top-three lists of people inside other cards follow the same rule.

Bill

Usage

Activity

How Claude is used

Trust and control

Speed and reliability

Output

Fleet

Limits of the figures

• One profile per person. Someone who runs two installs, such as a terminal and an editor extension on different versions, shows under whichever reported last.
• Inactive developers ignores the period. The card counts from today, among people seen in the last 400 days.
• Average prompt length can read low. Claude Code 2.1.288 and older send no prompt length, and LevelFour counts those prompts as empty.
• Response times are estimates. LevelFour sorts each request's time into a range, then works out the median and the 90th percentile from the count in each range.

Confirming it worked

Each account you connect moves to Connected on the Anthropic page. Claude API and Claude Enterprise data appears in about an hour. Claude Code and Cowork usage turns Connected after the first prompt and Verify connection, and its figures appear on the board within minutes.

Rotating or removing access

Rotating a credential

Create a new key in Claude Console or claude.ai. On the Anthropic page under Connect Providers, click Rotate credential on the account, paste the new key, then revoke the old key at its source.

You cannot rotate an ingest key. Disconnect the Claude Code and Cowork usage source, keeping its data, generate the settings again, and replace the block wherever you pasted it.

Removing access

Click Disconnect on the account under Connect Providers, then revoke the key. An Admin API key is revoked from Claude Console's Settings > Admin keys; an Analytics key from claude.ai's Organization settings > API. Disconnecting the Claude Code and Cowork usage source stops its ingest key working at once. Remove the block from your settings so Claude Code stops sending.

Troubleshooting

Next

• Connect Providers lists every provider you can connect
• OpenAI connects with a single Admin key