Changelog

All notable changes to llm-toll are documented here.

v0.13.0

• Round all costs to 10 decimal places (COST_ROUND_PLACES) to prevent floating-point drift
• Web dashboard for cost trends and usage analytics

v0.12.0

• PostgreSQL backend for team-wide cost tracking
• PostgresStore with ThreadedConnectionPool and row-level locking
• create_store() factory function with URL-based backend selection
• LLM_TOLL_STORE_URL environment variable support
• Graceful fallback to SQLite on connection failure

v0.11.0

• Auto-updating pricing registry from remote source
• update_pricing() function with local caching (24h TTL)
• --update-pricing CLI command
• load_remote_pricing() for bulk pricing updates

v0.10.0

• LangChain callback integration (LangChainCallback)
• Pre-call budget check in on_llm_start
• Post-call usage logging in on_llm_end

v0.9.0

• LiteLLM callback integration (LiteLLMCallback)
• Model name normalization for LiteLLM provider prefixes
• Zero-decorator cost tracking for all LiteLLM calls

v0.8.0

• Hardening batch 3: TPM retry_after calculation, prefix boundary matching, bounded dynamic cache
• Protected user-registered models from cache eviction
• Fixed prefix resolution race condition

v0.7.0

• Hardening batch 2: TOCTOU documentation, mid-stream break test, coverage CI
• Hardening batch 1: version single source of truth, context manager support, pricing validation

v0.6.0

• CLI dashboard for viewing costs and usage statistics
• --stats, --reset, --export csv commands
• --project and --model filters
• Color-coded cost output with NO_COLOR support

v0.5.0

• Async decorator support for async functions and async generators
• asyncio.to_thread for non-blocking SQLite operations
• Transparent async stream wrapping

v0.4.0

• Gemini SDK auto-parser
• Local/Ollama provider support with $0 cost tracking
• Rate limiting for local models

v0.3.0

• RateLimiter with sliding-window RPM and TPM enforcement
• LocalRateLimitError with retry_after attribute
• Pre-call rate limit checks in the decorator

v0.2.0

• Streaming support for OpenAI and Anthropic
• StreamAccumulator for chunk-based token tracking
• Character-based token estimation fallback

v0.1.0

• Initial release
• @track_costs decorator with auto-detection for OpenAI and Anthropic
• PricingRegistry with built-in model pricing
• SQLiteStore for local persistence
• CostReporter with color-coded terminal output
• BudgetExceededError for budget enforcement
• GitHub Actions release pipeline