Claude Code

If you already pay for Claude Code (Pro or Max), this lets the chronicle pipeline use that allowance instead of an API key. No per-token billing, because there's no API key involved at all.

The realisation behind it was slightly annoying: I was paying a monthly subscription for Claude and paying per token through the API to write up my sessions. Two bills for the same company. This closes that gap — it shells out to the claude command already on your machine, running against the session you logged into.

It never touches your login, and it strips any API key out of the call before running, specifically so the work can't quietly land on a metered bill by accident.

Usage limits — chronicling burns through your plan fast

Claude subscriptions (Pro / Max) meter headless usage in rolling ~5-hour windows, not per token. The chronicle pipeline is token-heavy — it runs your whole transcript through several phases — so a single session can consume a large slice of that window.

Warning

In testing, chronicling one full session used up to ~60% of the usage allowance in a 5-hour window. Budget for roughly one to two full sessions per window before you hit the limit.

Ways to stretch it:

If a run stops with a usage / limit message, wait for the window to reset, or move the remaining phases to another connection in the routing rows.
Pair it with Chronicle Reforge: let Claude Code do the phases it's strongest at (grounding and the chronicle prose), then re-run the cheaper extras and condensed recap elsewhere. That keeps the heaviest, most refusal-prone phases off your Claude allowance entirely.
The lighter audit / extras phases default to haiku, which is cheaper against your plan than sonnet / opus.

What you need

Install Claude Code — follow the official guide at docs.claude.com/en/docs/claude-code. After install, the claude command must be on your PATH.
Log in with your plan — run:
```
claude login
```
and choose your Pro/Max plan when prompted.
Make sure ANTHROPIC_API_KEY is not set. If that environment variable is present, the CLI uses it (API billing) instead of your subscription. Tusk's Tomes strips it from the call it makes, but it's cleanest to leave it unset in your shell.

Enabling it

Note

Nothing to install and no restart — the app detects whether the CLI is present and signed in.

Open Settings → Providers & models. The Claude Code row shows whether the CLI is detected and logged in. Click Recheck after logging in.
Assign it to individual phases in the routing rows, mixing it with other connections — or pick a subscription rung in guided routing to put every phase on it at once.

Choosing a model

Per-phase models are set in the routing rows, exactly like every other connection. Claude Code accepts the aliases sonnet, opus, and haiku as well as full model IDs. The aliases are listed first, and deliberately: each resolves to whatever the CLI currently calls that tier, so a CLI upgrade moves them forward on its own. A pinned ID does not, which is how a configuration ends up still naming a superseded model months later.

How it differs from a metered API key

	An API key (e.g. via OpenRouter)	Claude Code (your subscription)
Auth	API key stored in the app's encrypted keystore	Your own `claude login` session — nothing stored by the app
Billing	Per-token API credit	Your Claude subscription
Rate limiting	Header-driven pacing	None — the subscription self-throttles; the CLI surfaces its own limit messages
Cost estimate	Per-token estimate	Shown as $0, covered by your subscription

Troubleshooting

"CLI not found" — claude isn't on your PATH. Reinstall Claude Code or restart your terminal/app so the updated PATH is picked up.
"Login not detected" — run claude login. On macOS the session may live in the Keychain and not be detectable from here; if runs work anyway, you can ignore the warning.
A run fails with a usage/limit message — you've reached your subscription's limit for headless usage. Wait for it to reset, or switch a phase to another connection in the routing rows. The run pauses and checkpoints rather than failing, so you can resume at the exact chunk it stopped on.

A note on terms of use

Using Claude Code's headless mode with your own subscription is a documented feature. Because this integration is distributed publicly, review Anthropic's Commercial Terms and the Use the Claude Agent SDK with your Claude plan article to confirm your usage fits. The integration is intentionally hands-off: it only invokes the CLI you authenticated yourself.