SAGE v2.4.4

Adds sage ml soak, a Ctrl+C-safe ML soak proof command that streams terminal logs, writes JSON/TXT/JSONL/log proof files, and uses an isolated SAGE data directory by default so corrupted local databases and test failures do not poison public proof metrics. Bumps psycgod-sage to 2.4.4.

SAGE v2.4.3

Public dashboard patch: fix success-rate totals by using live aggregate run counts, add Ollama to model/agent proof rows, and attribute sage run -- ollama commands to Ollama.

SAGE v2.4.2

Admin usability patch: sage api users now shows saved display names/hostnames and short machine IDs while keeping hash-like internal usernames hidden.

SAGE v2.4.1

Privacy patch: sanitize sage api users output by default, make raw admin identity fields opt-in, and deploy sanitized Cloudflare admin users endpoint.

SAGE v2.4.0

SAGE v2.4.0 hardens mandatory local AI-agent enforcement. Adds sage install system-wide agent instructions/hooks, sage init project enforcement, live output caps, telemetry off/local-only CLI support, and verified PyPI build artifacts. Verified with compileall, full pytest suite, build, and twine check.

v2.2.2 — ML prediction counter now live

ML Prediction events_with_prediction now includes all commands analyzed (training data + live usage). Number grows with every command run.

v2.2.1 — Live dashboard auto-refresh

Dashboard proof snapshot now auto-sends every 10 commands. Sender debounce reduced from 90s to 30s. No more stale numbers on the public dashboard.

Release 2.2.0 — One-Command Install

What's New

Zero-friction install — just pip install psycgod-sage and run any sage command. Connection happens automatically on first use.

Auto-Connect Flow

1. pip install psycgod-sage
2. Run any command (e.g. sage run -- python test.py)
3. SAGE auto-connects: asks expiry (30/60/90 days), asks your name, authenticates via hardware fingerprint (zero interaction), stores key in keyring

Auth Priority

• Hardware auth (instant, no interaction) — uses machine UUID
• GitHub CLI (silent) — if gh is logged in
• GitHub OAuth (browser) — last resort

Other Changes

• /v1/whoami endpoint for server-side key verification
• /v1/machine-login endpoint for hardware-based registration
• sage api whoami now verifies connection with server
• Detects installed AI agents (Claude Code, Codex, Aider, Cursor, Windsurf)
• Connected users count in admin API
• Display name prompt for admin identification

Full Changelog: v2.1.0...v2.2.0

Release 2.1.0 — LSP Server + Agentic Loop

What's New

LSP Server

• sage lsp — start LSP server for editor integration (stdio)
• sage lsp --tcp --port 19473 — TCP mode for AI agent connections
• Custom methods: sage/predict, sage/explain, sage/fix, sage/session
• VS Code extension scaffold with full LSP client

Agentic Loop Engine

• Session state — tracks command history, failure streaks, error patterns
• Decision engine — 3 autonomy levels: suggest, ask, auto
• Auto-fix pipeline — 9 error patterns (ModuleNotFoundError, port in use, permission denied, etc.)
• Intent detection — recognizes deploy, fix_bug, setup, refactor, publish workflows
• Circuit breaker — exponential backoff, loop detection, max retries
• Verification system — re-runs commands to confirm fixes worked

MCP Integration

• sage_agentic_run — run with full retry loop
• sage_agentic_fix — auto-fix last failed command
• sage_agentic_session — get session context for AI

Configuration

• sage.toml support for agentic + LSP settings
• Configurable: autonomy level, max retries, fix patterns, never-auto-fix list

Testing

• 33 new tests (LSP protocol + agentic loop)
• All 204 tests pass

Install / Upgrade

pip install --upgrade psycgod-sage

Live Dashboard

sage.api.marketingstudios.in/dashboard

Release 2.0.7

Performance fix + ML prediction daemon