llm_meter library

Classes

BatchingSink: Wraps another MeterSink and flushes events in batches.
ConsoleSink: Prints every event with debugPrint (no-op in release unless printInRelease is set).
Currency: Currency that costs can be displayed in. USD is the canonical unit costs are recorded in; everything else is converted at display time.
LlmMeter: The meter singleton.
LlmMeterHud: Drop-in widget that overlays a small floating card showing the live cost, latency, and cache stats reported by LlmMeter.
MeterConfig: One-line setup passed to LlmMeter.init.
MeteredCall: Wrap any LLM Future + extractor so the meter records it automatically.
MeteredStream: Wraps a Stream<C> of LLM chunks so the meter records the full call once the stream completes.
MeterEvent: Immutable record of a single LLM API call.
MeterSink: Receives every MeterEvent that the meter records.
MeterStats: Snapshot of stats computed over the events currently held by LlmMeter.
MeterUsage: Token usage extracted from a provider response.
MixpanelSink: Sends events to Mixpanel's /track ingestion endpoint.
ModelPricing: Per-token price for a single model variant, expressed in USD per token.
ModelStats: Per-model summary.
PosthogSink: Send events to PostHog's /capture/ endpoint via plain HTTP.
RetryingSink: Wraps another MeterSink and retries on failure.
RingBuffer<T>: Fixed-capacity, O(1) push ring buffer. Drops the oldest entry when full.

HudCorner: Corner anchor for the HUD card.
SymbolPosition: Whether the currency symbol prints before or after the number.

defaultPricing → const ModelPricing: Fallback when a model id is unknown to both the built-in table and any user overrides.

builtInModels → List<String>: The set of model ids known to the built-in pricing table.
no setter

anthropicUsage(Map<String, Object?> response) → MeterUsage: Anthropic Messages API usage shape: { usage: { input_tokens, output_tokens, cache_read_input_tokens, cache_creation_input_tokens } }.
gdprScrub(MeterEvent event) → MeterEvent: Redacts user-correlatable fields from event for GDPR-safe transmission.
geminiUsage(Map<String, Object?> response) → MeterUsage: Google Gemini generateContent usage shape: { usageMetadata: { promptTokenCount, candidatesTokenCount, cachedContentTokenCount } }.
openAiUsage(Map<String, Object?> response) → MeterUsage: OpenAI Chat Completions / Responses API usage shape: { usage: { prompt_tokens, completion_tokens, prompt_tokens_details: { cached_tokens } } }.
priceForModel({required String model, required int tokensIn, required int tokensOut, int cachedTokensIn = 0, Map<String, ModelPricing>? overrides}) → double: Pure cost calculation. Use this when you want a one-shot quote without recording an event.
pricingFor(String model, {Map<String, ModelPricing>? overrides}) → ModelPricing: Lookup the bundled ModelPricing for model.

ChunkUsageExtractor<C> = MeterUsage? Function(C chunk): Signature for accumulating per-chunk usage info during a stream.
UsageExtractor<T> = MeterUsage Function(T response): Signature for converting a provider response into a MeterUsage.