Claude Code statusline behind LiteLLM and other gateways

How the Sprag statusline reads spend, budgets and cache TTL when Claude Code runs through LiteLLM or another API gateway, and why the cache countdown can look frozen.

🌐 Behind a gateway (Bedrock / Vertex)

A gateway reports the cache-creation total but never the 5m/1h split. That left the tool unable to tell "nothing cached yet" from "this provider does not say", and the fallback assumed an hour — for a window that is really five minutes on Bedrock, overstating it twelvefold.

Since v3.26.0 the gateway is detected from the model ids in the transcript, which fixes:

• The countdown falls back to 5 minutes, labelled 5m?. Three grades of certainty get three labels: measured (5m), inferred (5m?), unknown (?).
• In a 5-minute bucket the countdown colour follows absolute time rather than a percentage. 30% of five minutes is 90 seconds, and green there promised comfort that was not there.
• The ⚠ 5m TTL warning finally reaches these users — with different advice, since no subscription plan changes a gateway's TTL.
• Extra cost if 5m-only is only asked of sessions that have 1h writes to lose. Elsewhere the arithmetically honest +$0 read as an endorsement of the bucket you are already stuck in.
• Delegated runs dropped for an unpriceable model id show as 🔀 N unresolved instead of nothing, which used to be indistinguishable from never having delegated.
• Environment variables set to a foundation-model ARN now resolve. An opaque application-inference-profile id still does not: guessing at it is how wrong prices enter the ledger.

Cost and routing savings on a gateway use the gateway's prices. A Claude session with a gateway ANTHROPIC_BASE_URL is billed by the gateway, so Actual cost, Sep spend and Routing saved are computed from its /model/info prices (refreshed with the model map, valid for seven days). A model the gateway does not price stays unpriced: the session is left out and counted (+N unpriced) instead of being filled in from the Anthropic list prices. Sessions that do not go through a gateway keep using the list prices below. A delegation to a pricier model is recorded as a loss (-$0.12), not clamped to zero.

If the detection is wrong, pin it with sprag mode ttl=5m (or ttl=1h). An explicit value outranks the measurement.

LiteLLM: your key budget stands in for the missing 5h/7d caps (v3.35.0)

Behind a LiteLLM proxy (Bedrock and friends), Claude Code's stdin never carries rate_limits, so the ✦ current / 📅 weekly gauges simply do not exist. LiteLLM does track per-key budgets, so the statusline draws a budget gauge in their place.

• Detection: ANTHROPIC_BASE_URL points somewhere other than the official endpoint. The key comes from ANTHROPIC_AUTH_TOKEN / ANTHROPIC_API_KEY, or — when neither is set — from running the apiKeyHelper script in settings.json. The helper only runs inside the 5-minute refresh child, never on the render path (v3.43.0). That call sets APIKEY_HELPER_NONINTERACTIVE=1, so a helper that checks it fails fast instead of opening a login window when it has no cached token; the refresh runs every five minutes with nobody at the keyboard, and would otherwise stack login windows while you are away.
• A 404 from /key/info no longer aborts the refresh: custom-auth deployments (Okta JWT and friends) always 404 there, and the budget lives in /user/info. Only a failure of both endpoints counts as a failure (v3.43.0).
• When the gauge falls back to the internal-user budget it is labeled budget (user). LiteLLM enforces the team-membership limit, which can differ from the user limit by orders of magnitude (v3.43.0).
• The proxy is asked via GET /key/info and GET /user/info — only the calling key's own data. Renders read a cache file; a detached background process refreshes it every 5 minutes (same shape as the update check), so the statusline never waits on the network.
• Budget source priority follows real-world usage: the team-membership budget (team_memberships[].spend + its linked budget table row) first, then the key's own max_budget, then the internal-user budget. Verified against a Dockerized LiteLLM, including memberships whose budget diverges from the team max into a separate budget-table row.
• Unlimited keys (no max_budget) get no gauge. The 💵 monthly-spend segment still shows, since it comes from session logs.
• Inspect with sprag litellm-budget (gauge with used/remaining amounts), --json (raw cache), or --refresh (query now).

claude.ai Enterprise: usage-credit budget gauge (v3.93.0)

Enterprise users who sign in with claude.ai do not get rate_limits in the statusline input either. The statusline now reads extra_usage from GET https://api.anthropic.com/api/oauth/usage and draws 💳 budget ▰▱ 54% $271/$500 in the same place.

• On by default. Turn it off with sprag mode no-oauth-budget (back on with sprag mode oauth-budget), or set SPRAG_OAUTH_BUDGET=0.
• It stays out of the way for everyone else: a gateway, ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN, or Bedrock, Vertex or Foundry skips it before any token is read. A Pro or Max login (subscriptionType in the stored credentials) skips the request and rests for 24 hours, since those plans get rate_limits.
• Source priority: ① rate_limits from the statusline input, ② the LiteLLM budget, ③ /api/oauth/usage. When an earlier source supplies a window, this one does not run at all.
• The endpoint is not in the official documentation. If the response has a different shape (no extra_usage, extra usage disabled, a non-USD currency), nothing is drawn.
• The OAuth token is read from CLAUDE_CODE_OAUTH_TOKEN, the macOS keychain item Claude Code-credentials, or .credentials.json in the Claude config directory, only inside the background refresh process. It is held in memory and never written to the cache, logs or arguments. Refreshing an expired token stays with Claude Code.
• The response carries no reset time, so the gauge shows none.
• If the token cannot be read (a denied or timed-out keychain prompt included), the refresh rests for 6 hours so permission dialogs do not pile up every five minutes.
• Inspect with sprag oauth-budget, --json (raw cache), or --refresh (query now, works even when turned off).

One related non-bug: if your session model is already sonnet, a sonnet-delegation (T1) rule can never save anything, because there is no price gap to capture. That is correct, but route-scan rules displayed it identically to "no delegations yet", so it now says outright that the rule does not apply at the current default model.

LiteLLM: model aliases come from the gateway (sprag profile-map --refresh)

Behind a LiteLLM gateway, sprag profile-map --refresh reads model aliases straight from the gateway's own GET /model/info, so route-scan no longer needs to wait on learned votes or a hand-edited profile-map.json. The refresh uses the same 5-minute check and 24-hour cache as the LiteLLM budget gauge. It never stores your auth token: each refresh reads it from apiKeyHelper at call time and lets it expire on the helper's own schedule. The three-step resolution order and the manual profile-map.json format are in route-scan.

Pricing (Jul 2026)

Per million tokens (USD), as used by the cost estimator:

Source: Anthropic pricing docs. Sonnet 5 stays at $2/$10; the planned rise to $3/$15 was cancelled. Haiku 5.5 prices each request by its own prompt length (input + cache writes + cache reads), so a request over 100k tokens pays the long rates. Versions ≤ 2.16.x priced Fable 5 at the Sonnet tier (~3× under-estimate) — upgrade to 2.17.0+.

Cache TTL by plan

FAQ

Is this the same tool as claude-token-saver? Yes. Sprag is the new name and the npm package is sprag-cli. The old claude-token-saver package stopped updating at 3.55.0, so move to sprag-cli to keep receiving releases.

Does it call an LLM or need an API key? No. Everything is post-hoc analysis of the session logs Claude Code already writes on your machine. No extra model calls, no key, no telemetry leaves your computer.

Will it change how I prompt? No. It installs hooks and a statusline once, then works from what you already do. Rules, delegation, and warnings show up inside your normal sessions.

Where do the savings numbers come from? Each delegated run writes a ledger entry with the actual price difference. The tier criteria behind the routing are benchmarked on public data; the method is in docs/BENCHMARK.md.

How it works · Environment

Claude Code logs every API call to ~/.claude/projects/<dir>/<session>.jsonl. This tool dedupes streaming chunks by requestId and aggregates cache_read_input_tokens / cache_creation.ephemeral_5m/1h_input_tokens by day and session.

Node.js ≥ 20 · macOS / Linux / Windows / WSL · zero dependencies.

npm uninstall -g claude-cache-monitor && npm i -g sprag-cli