Skip to main content
Guardian
DocsChangelog
FAQContact
Download
Guardian

AI-generated code governance for small engineering teams. Enforce team policies and approve releases with confidence.

Product

  • Changelog
  • Docs
  • FAQ
  • Contact

Resources

  • Getting Started
  • Security
  • Configuration
  • Guru AI
  • Monitoring

Legal

  • Privacy Policy

© 2026 Guardian Project. All rights reserved.

Designed & Developed bySenol Dogan

Changelog | Guardian

Release History

Guardian changelog

All release notes are fetched from release records in real time.

April 2026

v1.3.0

v1.3.0 - AI Pipeline Overhaul + MCP Server + Developer Ecosystem

Published: Apr 19, 2026

Release notes

Highlights

  • Evidence-Based Findings: Critiques now include `line_start`, `line_end`, `evidence_snippet`, `category`, and `confidence` fields for verifiable, actionable results.
  • Few-Shot Prompts: System prompts upgraded with concrete good/bad examples, chain-of-thought reasoning, and language-specific rules (React hooks, Rust ownership, Go error handling, Python type hints).
  • Prompt Template System: Embedded prompts via `include_str!` with `{{VAR}}` substitution and workspace-level overrides (`prompts/` directory).

Release notes

  • Evidence-Based Findings: Critiques now include `line_start`, `line_end`, `evidence_snippet`, `category`, and `confidence` fields for verifiable, actionable results.
  • Few-Shot Prompts: System prompts upgraded with concrete good/bad examples, chain-of-thought reasoning, and language-specific rules (React hooks, Rust ownership, Go error handling, Python type hints).
  • Prompt Template System: Embedded prompts via `include_str!` with `{{VAR}}` substitution and workspace-level overrides (`prompts/` directory).
  • Import Context Analysis: Critiques now receive `build_import_context()` data showing local module relationships and type signatures.
  • Performance Tuning: Batch size 3→5, content limit 6K→8K chars, concurrency 1→2, flush interval 5s→3s.
  • CLI Prompt Parity: CLI batch scan now uses the same high-quality system prompt as desktop.

🖥️ UI Enhancements

  • Line Numbers: Findings display exact line ranges (`L42-L58`).
  • Category Badges: Color-coded `security`, `performance`, `maintainability`, `correctness`, `style` tags.
  • Confidence Indicator: Visual percentage showing AI certainty per finding.
  • Evidence Snippets: Expandable code evidence directly in the critique row.
  • DiffViewer: Side-by-side diff display for suggested fixes.
  • Category Filter Bar: Filter findings by category with one click.

🔌 MCP Server (NEW)

  • Guardian MCP server (`guardian-mcp`) with 5 production tools:
  • `scan_file` — Scan a single file against policy with metadata
  • `get_scan_policy` — Retrieve workspace or default scan policy
  • `classify_paths` — Batch classify files by policy (up to 100)
  • `list_critiques` — Guidance for accessing stored critiques
  • `apply_fix` — Guidance for applying suggested fixes
  • Works with Cursor, Claude Desktop, and any MCP-compatible client.

🔒 Security Hardening

  • Thread-Safe Env: Replaced `std::env::set_var` UB with `SecureEnvStore` atomic store.
  • Secret Masking: API keys wrapped in `SecretString` (zeroed on drop).
  • Scoped FS: Tauri permissions narrowed from global to workspace-only access.
  • CSP Tightened: Removed `unsafe-eval` from Content Security Policy.
  • IPC Auth Gates: Added auth checks to `apply_fix`, `apply_fix_now`, `undo_fix`, `confirm_fix`.
  • Redaction Patterns: Expanded from 5 to 19+ patterns (OpenAI, Anthropic, AWS, GCP, Stripe, Slack, npm, JWT, Private Keys, Database URLs, Email, Phone).
  • CLI Redaction Parity: CLI now matches desktop's full 19-pattern redaction.
  • Transcript Detection: 2-tier scoring (strong markers instant-reject, weak markers need ≥2) to avoid false positives.
  • Path Safety: CLI no longer leaks absolute paths on `strip_prefix` failure.

Release notes

  • App.tsx: Decomposed from 1355→390 lines into focused components.
  • SettingsModal: Split from 1779→160 lines into 6 sub-components.
  • ProviderSpec Trait: Extracted provider logic from monolithic AI client.
  • Zustand Stores: Created `authStore`, `workspaceStore`, `uiStore`, `toastStore`.
  • useSettings: Split into 6 focused sub-hooks.
  • i18n: Split inline translations into separate locale files.

🌐 Website Updates

  • 6 New Doc Pages: CLI guide, MCP Server setup, Secret Redaction reference (EN + TR).
  • Ecosystem Section: New homepage component showcasing CLI, MCP, VS Code, Redaction, Multi-Provider, Evidence.
  • FAQ Expansion: 2 new general questions + "Developer Tools" category (5 questions) in EN and TR.
  • 7 New Feature Descriptions: Added to i18n for CLI, MCP, VS Code, rule engine, redaction, evidence, multi-provider.
  • Translation Fixes: Corrected Turkish translations.

Release notes

  • 6 New E2E Groups: Category Filters, Guru Chat, Keyboard Shortcuts, Evidence Display, Settings Extended, Accessibility.

🛠️ Developer Experience

  • 52 IPC Commands Documented: Full API reference in `docs/IPC_COMMANDS.md`.
  • Cargo Workspace: All 5 crates unified under workspace.
  • SQLite WAL: Enabled WAL mode for concurrent read performance.

📦 Ecosystem

  • guardian-cli: Production-ready CLI scanner with full prompt parity.
  • guardian-mcp: MCP server for IDE integration.
  • guardian-vscode: VS Code extension scaffold with diagnostics and MCP client.
  • guardian-scan-policy: Shared scan policy between desktop and CLI.
v1.2.6

v1.2.6 - ## [1.2.6] - Approval Policy + Audit Compatibility + Review Precision

Published: Apr 04, 2026

Release notes

Highlights

  • Release approval policy documented with role boundaries for release managers, incident commanders, and reviewers.
  • Legacy `.guardian/release_decisions.jsonl` rows now remain readable during upgrades, so manual approval state is preserved.
  • Audit precision hardened for local/dev-only findings and recent-fix context reuse to reduce stale duplicate reports.

Governance

  • Release gate behavior and approval state rules now link directly to the release approval policy document.

Dependencies

  • Updated `jspdf` and `next` patch versions and added dependency overrides to reduce critical/high npm audit exposure.

March 2026

v1.2.5

v1.2.5 - Cross-Platform Release + UX Polish

Published: Mar 15, 2026

Release notes

Highlights

  • Cross-platform release flow stabilized for macOS + Windows with strict release gate enforcement.
  • Website UX polish on homepage section transitions (softer section boundaries in light/dark themes).
  • Documentation release-note sync improved and mobile readability refinements continued.

Release notes

  • `actions/checkout@v6`
  • `actions/setup-node@v6`
  • `actions/upload-artifact@v6`
  • `actions/download-artifact@v8`

Governance

  • Release-note synchronization remains changelog-driven; `1.2.5` entry added for deterministic release note rendering.
v1.2.4

v1.2.4 - Release Readiness: Website + Gate Stability

Published: Mar 15, 2026

Release notes

Highlights

  • `guardian.policy.yaml` warning threshold tuned to `pass_max_warnings: 10` for strict gate stability.
  • `release_all_local.sh --gate-only` now passes with `PASS_WITH_WARNING` on current repo state.
  • Scan noise reduced in governance core:

Security / Quality

  • Website dependency audit fixed (`npm audit fix`): high/moderate vulnerabilities reduced to zero.

February 2026

v1.2.3

v1.2.3 - ## [1.2.3] - Quality-First: Faster Scans + Project-Aware Audits + One-Click Fixes

Published: Feb 18, 2026

Release notes

Highlights

  • Full EN/TR localization: Desktop + website UI now support English and Turkish. Guru + Monitor AI outputs follow your selected language (severity tokens remain stable).
  • One-click Apply + Undo: apply fixes immediately from Monitor and Guru, with a per-file Undo history.
  • Guru notifications: when a reply arrives and you are on another tab, you get a toast + a sidebar badge count.

Changed

  • Project Map context now respects the persisted Scan Scope (`source|extended|full`) for consistency with monitoring.
v1.2.2

v1.2.2 - Balanced Performance + Smart Scope

Published: Feb 17, 2026

Release notes

Highlights

  • Desktop + CLI alignment: shared scan policy core to avoid drift in what gets scanned and skipped.
  • Guru stays responsive while monitoring: adaptive AI queue with fairness between audit traffic and chat.
  • Web search upgrades (Tavily): URL-aware extract when a link is present, plus configurable search depth.

Fixed

  • Website rate limiting: release pages prefer the `releases.json` snapshot to avoid GitHub API 403/rate-limit spam.

Changed

  • Settings: added Scan Scope and Web Search depth controls; copy clarified around search vs extract behavior.
  • Release hardening: stricter CSP allowlist and secret-scan gate integrated into `npm run verify`.
v1.2.1

v1.2.1

Published: Feb 11, 2026

Release notes

Highlights

  • Batch processor default flush threshold increased from `2` to `3` files (`GUARDIAN_MAX_BATCH_SIZE` default).

Highlights

  • Batch processor default flush threshold increased from 2 to 3 files (GUARDIAN_MAX_BATCH_SIZE default).

Changed Desktop watcher and guardian-cli file filtering are now aligned to the same code-first scan policy. Version sync updated across app, Tauri, and website to 1.2.1.

v1.2.0

v1.2.0 - STABILIZATION

Published: Feb 11, 2026

Release notes

Highlights

  • Ollama Offline UX: Ollama provider no longer requires an API key to launch; requests omit `Authorization` header when no key is configured.
  • Batch Audit Safety: Normalized/whitelisted `file_path` for batch critiques to prevent ghost findings and state inconsistencies.
  • AI Context + Reviews UX: Replaced the long accordion layouts with an inspector split view (list/filters on the left, preview/actions on the right) to reduce scroll fatigue and speed up triage.

Fixed

  • Ollama Offline UX: Ollama provider no longer requires an API key to launch; requests omit `Authorization` header when no key is configured.
  • Batch Audit Safety: Normalized/whitelisted `file_path` for batch critiques to prevent ghost findings and state inconsistencies.

Changed

  • AI Context + Reviews UX: Replaced the long accordion layouts with an inspector split view (list/filters on the left, preview/actions on the right) to reduce scroll fatigue and speed up triage.
  • Settings (Updates) About Links: Added "Built by Senol Dogan" and quick links to the website + contact page.
v1.1.1

v1.1.1 - Hotfix: Updater Signature & UI

Published: Feb 10, 2026

Release notes

Highlights

  • Fixed updater error
  • Improved dark mode colors
  • Added changelog website link

v1.1.1 - Hotfix Release

Bug Fixes:

  • Fixed updater error
  • Improved dark mode colors
  • Added changelog website link

New Features (from v1.1.0):

  • Semantic search for similar code patterns
  • 92% token usage reduction with diff-focused context
  • Guardian.lock for consistent team scans
  • Reviews queue for managing fix proposals

Version: v1.1.1 Updated: Feb 10, 2026

v1.1.0

v1.1.0 - Phase 6 Complete

Published: Feb 10, 2026

Release notes

Highlights

  • Find similar code patterns across your codebase
  • Local vector database with sqlite-vec
  • Support for OpenAI, Ollama, or offline mode

v1.1.0 - Phase 6 Complete

Semantic Search:

  • Find similar code patterns across your codebase
  • Local vector database with sqlite-vec
  • Support for OpenAI, Ollama, or offline mode

Performance Improvements:

  • 92% reduction in AI token usage
  • Faster analysis with diff-focused context

Team Features:

  • Guardian.lock ensures reproducible scans
  • CLI enforcement modes (strict/warn/off)

Reviews System:

  • Review, apply, or reject proposals

Version: v1.1.0 Released: Feb 10, 2026

v1.0.0

v1.0.0

Published: Feb 09, 2026

Release notes

No notes were provided for this release.

Archive

Download
April 2026March 2026February 2026