v0.3.0

Availability, privacy, and observability improvements since v0.2.0.

Added

• Opt-in structured JSON logging (LOG_FORMAT=json) + a per-request cost/token log line, for shipping to log/metrics aggregation (Loki/Grafana)
• Configurable data-retention windows and PII-redaction controls for stored request data

Changed

• Privacy-forward defaults: request logs auto-expire (30 days), reliability events (90 days), and user input is PII-redacted before storage — all env-overridable
• Streaming concurrency and gunicorn worker/thread counts are now env-configurable

Fixed

• Streaming can no longer exhaust the worker pool — concurrent SSE streams are bounded per worker (excess requests get a fast 503)

Full changelog: v0.2.0...v0.3.0

v0.2.0

First release since v0.1.0 — a full development cycle of admin, observability, and cost-accuracy work.

Added

• Opt-in admin panel with an interactive analytics dashboard — usage, cost, latency, and reliability metrics, persona management, and response-log browsing with retention pruning
• Accurate per-model cost estimation with an input/output token split, priced from a vendored, refreshable LiteLLM cost map (with a configurable override)
• Reliability tracking — error rate, provider-fallback frequency, and content-filter rate, with recent-failure detail
• Token usage now recorded for streamed responses

Fixed

• Streaming reliability, and OpenAPI/Swagger accuracy for the streaming transport

Full changelog: v0.1.0...v0.2.0

v0.1.0 — first public release

First public release of snark — Sarcasm as a Service, an AI-powered humor and utility API where every response is uniquely generated by an LLM.

Highlights

• 28 wit endpoints across humor, developer culture, and workplace comedy
• Multi-provider abstraction (Groq, Gemini, Claude) with automatic fallback on content filters or failures
• Persona system with configurable prompts, tone, temperature, and token limits
• 15 mood modifiers to shift the tone of any endpoint
• Redis caching (5-min responses, 1-hr personas) + anti-repetition logic
• IP-based rate limiting (50 req/hour) — fully public, no authentication
• Health probes (live/ready/status) and OpenAPI/Swagger docs
• Self-contained Docker stack — docker compose up brings the API, PostgreSQL, and Redis together
• Green CI (Black, isort, Flake8, pytest)

Licensed under AGPL-3.0-or-later.