Skip to content

CHANGELOG.md

Latest commit

 

History

History
165 lines (106 loc) · 6 KB

CHANGELOG.md

File metadata and controls

165 lines (106 loc) · 6 KB
Raw

Changelog

v2.6.1

  • Public distributions contain only the Python wrapper and thin npm launcher.
  • Removed GUI, TUI, Electron, editor-extension, enterprise-preview, and legacy JavaScript code.
  • Added synchronized PyPI, npm, and official MCP Registry metadata.

v2.6.0

Fixed

  • Preserve quoted arguments on Windows by launching native executables without a second cmd.exe parsing pass.
  • Add --cwd and SAGE_WORKSPACE_CWD so Electron and other desktop hosts can supply the real project directory.
  • Make foreground telemetry sync non-blocking; proof snapshots remain handled by the detached sender.
  • Keep the npm launcher and PyPI core compatible by upgrading an older Python core before delegation.
  • Synchronize Python, npm, lockfile, exported, and MCP version metadata.

Packaging

  • Keep npm install passive and publish only the thin launcher/bridge files.
  • Publish PyPI first and npm second from one GitHub release workflow using trusted publishing.
  • Add release version checks, npm launcher CI, package dry-runs, and a PyPI-availability gate before npm publish.

v2.4.3

Fixed

  • Public proof success rate now uses real SAGE run counts instead of token-usage rows, fixing bogus ratios like 15/7,502 successful.

Added

  • Ollama is now included in public model and agent savings rows.
  • sage run -- ollama ... is attributed to the Ollama row in public proof snapshots.

v2.4.2

Changed

  • sage api users now shows the saved display name or hostname by default, while still hiding hash-like internal usernames.
  • Connected-user rows now show short machine/install IDs so duplicate runner machines can be distinguished.

v2.4.1

Fixed

  • sage api users now hides names, usernames, hashed runner identities, and raw telemetry-install counters by default.
  • The private admin users endpoint now returns sanitized connected-user rows by default; raw identity fields require ?raw=1.
  • Added regression tests so raw identity output remains opt-in.

v2.4.0

Added

  • sage install now performs system-wide local AI-agent enforcement without requiring cloud connection.
  • sage init now installs project-local AGENTS.md, CLAUDE.md, SAGE.md, and Claude Code hook/settings files.
  • Claude Code enforcement hook blocks bare shell commands, direct file/search/edit tools, and subagents without SAGE instructions.
  • Live command output is capped by default so very noisy commands do not flood agent context before the SAGE summary.
  • sage telemetry off sets local-only telemetry level 0, and sage telemetry preview --level 0 is accepted.

Changed

  • README now makes sage install and sage init explicit first-class setup steps.
  • sage connect copy now describes optional public proof sync instead of implying local SAGE requires cloud auth.

v2.3.1

Changed

  • First-run flow: ML V2 prompt (y/n) runs before sage connect.
  • Connect ending now confirms all detected AI agents use SAGE compulsorily.
  • Clearer ML prompt explains V1 keeps learning if user skips V2.
  • Connect prints "sage run -- pytest" as next step after setup completes.

v2.3.0

Changed

  • ML V2 dependencies (torch, sentence-transformers, faiss-cpu) are now optional.
  • Install base SAGE without CUDA/torch: pip install psycgod-sage.
  • Install with ML V2: pip install psycgod-sage[ml].
  • Post-install prompt asks users if they want ML V2 features.

Fixed

  • Fix PowerShell built-in commands not working on Windows (sage now detects PowerShell vs cmd.exe).
  • Resolve version inconsistencies across README and CHANGELOG.

v2.2.2

Fixed

  • Stability fixes for context compression stats persistence.
  • Telemetry sender background thread reliability.

v2.2.0

Added

  • File operation tools: sage read, sage grep, sage write, sage edit, sage glob, sage tree.
  • sage call for tracked agent tool-calls.
  • sage show --raw to recover exact stored output.
  • MCP tools for file operations (sage_read_file, sage_grep, sage_write_file, sage_edit_file, sage_glob, sage_tree).

Changed

  • MCP tool surface expanded from run/explain/suggest/agents/workflow/history to include file and search tools.

v2.1.0

Added

  • LSP server (sage lsp) for editor and AI agent integration.
  • Agentic retry loop with circuit breaker for auto-fix workflows.
  • MCP tools: sage_agentic_run, sage_agentic_fix, sage_agentic_session.
  • sage.toml configuration for agentic autonomy level and LSP transport.

Changed

  • Agent firewall now supports interactive approval prompts.
  • Compression strategies are selectable per-command.

v2.0.1

Fixed

  • Package build verification for clean virtual environment installs.
  • Cloudflare Worker dashboard auto-refresh interval (now 10s).
  • Public proof snapshot includes estimated_savings_usd and savings_by_agent.

Documentation

  • README updated with live proof stats from public dashboard.
  • Install path clarified for PyPI/npm first installs.

v2.0.0-cli-public

Public CLI-first release candidate for SAGE.

Changed

  • Rename the public PyPI distribution to psycgod-sage while keeping the installed command as sage.
  • Mark the package as beta because the public package is CLI-first and the GUI is not included yet.
  • Position SAGE as a local-first command wrapper for AI coding agents.
  • Document the public proof model, telemetry behavior, and known limitations.

Security

  • Store new SAGE API keys in the operating system keyring when available.
  • Keep raw command output and logs local by default.
  • Keep public dashboard data aggregate-only.

Documentation

  • Add install instructions for pip install psycgod-sage.
  • Add release notes for v2.0.0-cli-public.
  • Add SECURITY.md and CONTRIBUTING.md.
  • Add terminal capture assets for sage run --, sage context report, and sage mcp install.

Known Limitations

  • The GUI is not public yet.
  • SAGE machine authentication is required for connected proof/dashboard sync.
  • Telemetry above local-only is opt-in and policy-limited.
  • The public dashboard publishes aggregate proof metrics only.