Skip to content

Project Status

Current package version: 0.8.0a1. The repository is an educational open-source project and remains pre-1.0.

Area Current state Main limitation
Platform setup Public alpha Published on PyPI; pre-1.0 compatibility
Loop Guard Locally usable Only sees routed traffic; single-node state
MCP Firewall Technical preview Task-based execution deferred; local approvals
Replay Technical preview UI and export contracts may still evolve
Benchmark Lab Technical preview Starter dataset is synthetic and small
Sandbox Technical preview Requires Docker; not a certified boundary
Agent Playground Public alpha Deterministic scenarios are intentionally small
IssuePilot Public alpha Planning is deterministic and does not replace review
ReproLab Technical preview Arbitrary execution requires Docker Sandbox
VS Code extension Alpha Publicly available in the VS Code Marketplace

Implemented

  • OpenAI- and Anthropic-compatible guarded proxy routes
  • exact request, tool, error, and sequence loop detection
  • Shadow/Enforce modes, limits, pause/resume, session export
  • stdio and Streamable HTTP MCP proxying with policy, approvals, validation, and audit
  • Replay SDK, automatic model traces, JSON/JSONL/OTel exchange, comparison, cost and failure tags
  • 30-task versioned benchmark dataset, three adapters, deterministic/custom scorers, paired bootstrap gate
  • copied-workspace Docker runner with diff, selective apply, discard, and export
  • deterministic Playground scenarios persisted as Guard and Replay records
  • GitHub/local issue import, implementation planning, explicit Git apply, and export
  • local bug reproduction packages with environment manifests and Docker-only execution
  • VS Code Activity Bar integration and workspace setup
  • Alembic-managed SQLite schema, backup, restore, cleanup, diagnostics
  • no product telemetry and metadata-only logging by default

Not complete

  • MCP task-based execution
  • fully integrated v1.0 dashboard and cross-module workflow
  • broad real-world benchmark datasets and adapter ecosystem
  • hardened multi-user or network deployment
  • macOS/Windows parity for real Sandbox security tests
  • formal security audit or certification

Release policy

Each milestone should remain installable, documented, and tested. Features may change before 1.0, but versioned event and export formats should change only through an explicit schema version.

Project priorities

Correctness, learning value, tests, privacy, and honest limitations come before monetization. The project has no paid tier, hosted service, SLA, or closed features. Optional donations do not affect architecture or issue priority.