Skip to main content
All notable changes to Attesta are documented here. This project follows Semantic Versioning. Until v1.0.0, minor versions may include breaking changes.
Attesta is in early development (v0.1.x). APIs may change between minor releases. Pin your dependency to a specific version in production.

​
v0.1.0 — Initial Release

​
Added

  • @gate decorator — protect any Python function with human-in-the-loop approval. Supports sync and async functions with automatic event loop bridging.
  • gate() wrapper — TypeScript equivalent of the @gate decorator with full generic type safety.
  • Attesta class — high-level entry point for production use with shared defaults for risk scoring, rendering, audit logging, and trust.
  • Attesta.from_config() — load configuration from YAML files with auto-detection of rich vs. legacy formats.
  • DefaultRiskScorer — 5-factor heuristic risk scorer analyzing function name, arguments, docstring, caller hints, and novelty.
  • CompositeRiskScorer and MaxRiskScorer — combine multiple scorers with weighted averaging or conservative max.
  • Four risk levels — LOW, MEDIUM, HIGH, CRITICAL with configurable challenge mappings.
  • Four challenge types — auto-approve, confirm, quiz, and teach-back.
  • Multi-party challenge — CRITICAL actions can require approval from multiple independent reviewers.
  • Adaptive trust engine — Bayesian trust scoring that adjusts risk based on agent history. Trusted agents see reduced friction; incidents revoke trust immediately.
  • min_review_seconds — enforce minimum review duration to prevent rubber-stamping.
  • TerminalRenderer — rich terminal UI with color-coded risk panels and interactive prompts (requires pip install attesta[terminal]).
  • Pluggable protocols — RiskScorer, Renderer, AuditLogger, TeachBackValidator, and ChallengeProtocol interfaces for custom implementations.
  • JSONL audit logger — append-only audit trail with SHA-256 hash chaining.
  • attesta CLI — init, audit verify, audit stats, audit rubber-stamps, trust show/list/revoke, and version commands.
  • Domain profile framework — DomainProfile, DomainRegistry, DomainRiskScorer, and register_preset() for custom industry-specific risk patterns.
  • Risk amplifiers in attesta.yaml — regex patterns that boost risk scores for matching action names.
  • MCP proxy (MCPProxy class and attesta mcp wrap CLI) for wrapping any MCP server with Attesta approval.
  • attesta_tool_handler decorator for custom MCP servers written in Python.
  • LangChain integration — AttestaToolWrapper for gating LangChain tool executions.
  • OpenAI Agents SDK integration — approval hooks for OpenAI function calling.
  • Anthropic Claude integration — tool-use interception for Claude API responses.
  • CrewAI integration — AttestaCrewAIToolWrapper for gating CrewAI tool executions.
  • Vercel AI SDK integration — gatedVercelTool and createAttestaMiddleware for TypeScript (TypeScript only).
  • No-code nodes — n8n (n8n-nodes-attesta), Flowise (flowise-attesta), Langflow (langflow-attesta), and Dify (dify-attesta) integration packages.
  • TypeScript SDK (@kyberon/attesta) — full implementation with gate(), Attesta, evaluate(), and all core protocols.
Run attesta audit verify after installation to confirm the audit chain is working. The hash-chained JSONL format is tamper-evident — modify any entry and every subsequent hash breaks.