Skip to content

CHANGELOG.md

Latest commit

 

History

History
105 lines (85 loc) · 4.71 KB

CHANGELOG.md

File metadata and controls

105 lines (85 loc) · 4.71 KB

Changelog

All notable changes to this project will be documented in this file.

[3.3.0] — 2026-03-01

Added

  • Release prep automation script: scripts/prepare_pypi_release.ps1 validates version/changelog parity, builds distributions, runs twine check, and performs local wheel install/import smoke checks.
  • PyPI operator checklist: docs/PYPI_RELEASE_CHECKLIST_v3.3.0.md with one-command prep and manual finalization steps.
  • Advanced intelligence SDK methods for direct /v1/* parity: counterfactual, trajectory_scan, instability, phase_coherence, interaction_matrix, cascade_risk, mechanism_probe, action_surface, multi_horizon_outlook.

Changed

  • Version bumped to 3.3.0.
  • Release metadata aligned to WaveGuard server v3.3.0 deployment surface.

[3.1.0] — 2026-02-26

Added

  • Live crypto market data: New server endpoint /v1/market/{action} with 7 actions (token_data, price_history, ohlc, top_coins, search, dex_token, dex_search) powered by CoinGecko and DexScreener
  • 10th MCP tool: waveguard_market_data for AI agents to fetch live crypto prices
  • Solana multi-chain support: x402 payments now support 7 networks including Solana mainnet (Base, Ethereum, Polygon, Arbitrum, Optimism, Solana)
  • A2A protocol: Google Agent-to-Agent discovery at /.well-known/agent.json (protocol v0.3.0, 4 skills)
  • Bazaar discovery: x402 extensions now include JSON Schema inputSchema/outputSchema for decentralized API marketplace discovery

Changed

  • Version bumped to 3.1.0 across all files and registries

[3.0.0] — 2026-02-25

Added

  • Fingerprint endpoint (POST /v1/fingerprint): Raw 52-dim physics embeddings
  • Compare endpoint (POST /v1/compare): Structural similarity between datasets
  • 5 crypto risk tools: token_risk, wallet_profile, volume_check, price_manipulation, market_data — all available via MCP
  • x402 crypto payments: Pay-per-call with USDC ($0.01/call) on Base mainnet
  • 6 MCP prompts: Guided workflows for anomaly analysis, monitoring, crypto due diligence

Changed

  • MCP tools expanded from 3 to 9 (now 10 in 3.1.0)
  • Server architecture split: CPU web function + GPU compute function on Modal

[2.3.0] — 2026-02-24

Added

  • Automatic retry with exponential backoff: max_retries=2 by default. Retries on 429, 500, 502, 503, 504, connection errors, and timeouts. Respects server Retry-After header.
  • Environment variable auto-read: SDK reads WAVEGUARD_API_KEY from env if no api_key is passed. Matches stripe/openai SDK pattern.
  • Debug logging: logging.getLogger("waveguard") for request-level diagnostics. Enable with logging.basicConfig(level=logging.DEBUG).
  • MCP console script: pip install WaveGuardClient[mcp] now installs waveguard-mcp command. Works with uvx --from WaveGuardClient waveguard-mcp.

Fixed

  • MCP server was excluded from PyPI package — mcp_server/ now included
  • Non-JSON 200 responses now raise WaveGuardError instead of silently returning zeroed results

Improved

  • Error messages now suggest WAVEGUARD_API_KEY env var on auth failures
  • Clearer timeout error messages include retry count

[2.2.0] — 2026-02-24

Added

  • Multi-resolution scoring: per-feature local energy measurement alongside global fingerprint
  • PCA dimensionality reduction when n_samples < n_dims
  • Adaptive per-feature z-score thresholds (Bonferroni-style correction)
  • New response fields: global_anomaly, feature_anomaly, anomalous_features

Improved

  • IoT F1: 0.30 → 0.87 (+190%) — subtle per-feature anomalies now detected
  • Average F1: 0.65 → 0.76 (+17%) across 6 benchmark scenarios
  • WaveGuard now wins 4/6 scenarios vs sklearn (was 1/6)

[2.0.0] — 2026-02-24

Added

  • Initial public release of the WaveGuard Python SDK
  • WaveGuard client class with scan(), health(), and tier() methods
  • Full exception hierarchy: AuthenticationError, ValidationError, RateLimitError, ServerError
  • MCP server for Claude Desktop & AI agent integration (stdio + HTTP transports)
  • 7 runnable examples covering all major use cases
  • Documentation: getting started, API reference, MCP integration, Azure migration guide
  • Test suite with mocked HTTP responses (runs offline)
  • pip install WaveGuardClient with minimal dependencies (just requests)

Key Features

  • One-call API: Send training + test data → get anomaly scores back
  • Fully stateless: Nothing persists between calls
  • Any data type: JSON, numbers, text, time series — auto-detected
  • MCP support: Built-in server for Claude Desktop and AI agents
  • Azure migration: Drop-in replacement for Azure Anomaly Detector (retiring Oct 2026)