v2.12.0

[2.12.0] - 2026-07-29

Core System Updates

• Token Usage & Cost Tracking: Added per-request token usage and estimated cost tracking across paid inference providers, with prompt/completion/total/reasoning token fields, pricing metadata, and automatic audit-schema migrations.
• Pricing Service: Added config/pricing.yaml and local pricing resolution with exact, glob, and provider-default matching while distinguishing unpriced models from explicit free local-model rates.
• Usage Reporting Pipeline: Added tracked generation paths and a capability-gated usage sink across OpenAI, Anthropic, Gemini/Vertex, Ollama, and OpenAI-compatible providers, including streamed web-search branches and reasoning/thinking token visibility.
• SGLang Inference: Added SGLang inference service support.

Chat-app & UI Improvements

• Admin Costs: Added admin-panel usage and cost visibility through expanded audit fields and a new Costs tab backed by aggregate usage metrics.
• Admin Panel: Improved admin panel UX and general interface polish.
• General UX: Applied additional user experience refinements across the client.

Bug Fixes & Technical Improvements

• Usage Aggregation: Added GET /admin/observability/usage aggregation support across SQLite, Postgres, MongoDB, and Elasticsearch backends.
• Pricing Drift Checks: Added a read-only pricing drift checker that compares local pricing against LiteLLM and OpenRouter sources for mismatches, missing models, and retired exact-key entries.

Documentation & Configuration

• Cost Tracking Docs: Added token usage and cost tracking documentation covering architecture, provider wiring, pricing updates, and known follow-up work for non-token-billed media/OCR services.
• Pricing Config: Added the new pricing configuration file for local rate management.