Repository navigation
Releases
Releases · qwertymuzaffar/llm-budget
v0.2.1
Patch Changes
- 49cad16: Resolve prices by the full model id before dropping the provider prefix. A price table keyed by a prefixed id such as
openai/gpt-oss-120bwas never matched, so calls to that model were silently recorded at $0 (or rejected underunknownModel: 'throw'). A provider-specific key now also wins over a bare one for the same model.
v0.2.0
Minor Changes
- 7655f73: New: RedisStore (node-redis and ioredis clients), reservations (
reserve/settle/release, andguard({ reserve })) so concurrent calls cannot collectively overshoot a limit, anonRecordledger hook for billing, and streaming support (budget.meter(principal, stream),meterStream,StreamUsageTracker) that meters OpenAI and Anthropic streams when they complete.
Patch Changes
- 13625bb: Expanded README: why per-user budgets, concepts (principal, limits, decision, windows), how the sliding rate limit works, and worked examples for plan tiers, usage meters, team + user budgets, token estimation, streaming, error handling, and testing with an injected clock. Design notes on record-after semantics, floating-point dollars, and failed calls.