PathKeep is a local-first desktop app for long-term archiving and intelligence for browser history. Built with Tauri 2, Rust, React 19, TypeScript, Vite, and Bun. Publicly validated support currently covers Google Chrome, Microsoft Edge, Firefox, ChatGPT Atlas on macOS, Perplexity Comet on macOS, and Safari on macOS with Full Disk Access; additional Chromium / Firefox-family adapters remain implemented but not yet publicly promised.
Most people assume their browsing history is always there when they need it. It isn't.
Most Chromium-based browsers — including Chrome, Edge, Brave, and Arc — inherit an upstream default that automatically deletes local browsing history older than approximately 90 days. This isn't a bug; it's hardcoded in the Chromium source. Safari only keeps local history for one year by default unless you manually reconfigure it. Even Firefox, which is more generous, eventually prunes history based on database size.
Cloud sync doesn't save you either. Syncing is designed for multi-device convenience, not as a long-term archive. Firefox Sync only uploads the last 30 days and expires synced history after 60 days. Brave Sync only covers URLs you manually typed. Arc doesn't sync history at all.
If you rely entirely on Chrome and never manually back up your browsing history, chances are high that your local browsing history older than three months is already gone. The pages you visited last year, the searches you ran during that important moment in your life, the rabbit holes you went down while learning something new — all silently erased from your machine.
📋 How long does your browser actually keep history? (reference table)
Data sourced from official documentation, help centers, and source code as of April 2026. "Cloud sync" refers to the browser's first-party sync service. Retention often depends on user settings; defaults are shown below.
| Browser | Engine | Local Default | Local Max | Cloud Sync | Effective Max |
|---|---|---|---|---|---|
| Google Chrome | Chromium | ~3 months (Chromium default) [1] | No built-in setting found | Google Account may retain longer depending on Web & App Activity settings [2] | ~3 months locally; longer if synced |
| Microsoft Edge | Chromium | ~3 months (Chromium default) [1] | Enterprise policy can shorten only [7] | Privacy dashboard retains some synced data; TTL undocumented [5] | ~3 months locally; cloud TTL unknown |
| Brave | Chromium | ~3 months (Chromium default) [1] | No keep-forever setting found | Only typed URLs; server deletes after 12 months inactive [8] | ~3 months locally |
| Arc | Chromium | ~3 months (Chromium default) [1] | No keep-forever setting found | Does not sync history [12] | ~3 months locally |
| Opera / Opera GX | Chromium | ~3 months (Chromium default) [1] | No keep-forever setting found | Syncs history, but cloud TTL is undocumented [17] | ~3 months locally |
| Vivaldi | Chromium | 3 months [9] | Forever (user setting) [9] | Syncs history, but cloud TTL is undocumented [10] | Forever (if configured) |
| Dia | Chromium-based | Undisclosed (likely ~3 months) | No keep-forever setting found | E2E encrypted sync exists, but no published history TTL [28] | Undisclosed |
| Safari | WebKit | 1 year (default) [14] | Configurable (no hard cap found) | iCloud syncs history across devices; cloud TTL undocumented [15] | Configurable |
| Firefox | Gecko | Capacity-driven (no fixed day limit) [18] | Configurable via max_pages; can retain years [18] |
Only last 30 days uploaded; expires after 60 days in cloud [19] | Years locally; ~60 days via Sync |
| LibreWolf | Firefox-based | Firefox-like (unless clear-on-close is on) [21] | Same as Firefox [18] | Firefox Sync available manually [21] | Firefox-like |
| Floorp | Firefox-based | Likely Firefox-like | Not documented | Not documented | Likely Firefox-like |
| Waterfox | Firefox-based | Likely Firefox-like | Not documented | Not documented | Likely Firefox-like |
| Zen Browser | Firefox-based | Likely Firefox-like | Not documented | Supports Firefox Sync for history; Window Sync is local only [27] | Likely Firefox-like locally |
| ChatGPT Atlas | Chromium | Undisclosed (likely ~3 months) | Undisclosed | Web History controls exist; cloud TTL undocumented [30] | Undisclosed |
| Perplexity Comet | Chromium-based | Local history stored; TTL undocumented | Undisclosed | E2E encrypted sync for history; cloud TTL undocumented [32] | Undisclosed |
Key takeaway: Unless you use Firefox (capacity-driven) or have manually configured browsers like Safari or Vivaldi, your browser is almost certainly deleting history on a schedule you never agreed to.
Note: The ~3 month Chromium limit is inferred from the upstream Chromium source code. While downstream browsers can alter this, most do not document a different retention policy.
You might wonder: does it actually matter?
Think about what browser history really is. It's a record of how you learned things, how you researched a major decision, what you were reading the week something changed your life. It captures what you cared about as a student, before you took that job, before you moved cities, before things were different. For many of us, a significant chunk of life plays out online — and browser history is the closest thing we have to a journal of that experience.
Those years of history are gone, not because you have chosen to delete them. They're gone because your browser decided they weren't worth keeping.
Until recently, it didn't matter much that browsers delete old history, because there was no practical way to extract meaning from literally tens of millions of raw URLs anyway. But that's no longer true. Local AI inference is now fast and cheap enough to run agentic analysis on years of browsing data, right on your own machine. Concepts like Andrej Karpathy's "LLM Knowledge Bases" — where LLMs compile and maintain a personal knowledge base — are becoming reality.
The question is no longer whether you can extract meaning from decades of history. The question is whether you’ll still have the data when that future arrives.
The data you save today is the raw material for intelligence tomorrow. But you can't analyze what you've already lost.
PathKeep runs quietly on your machine and incrementally backs up browsing history from all your browsers — automatically, on a schedule, without manual effort. It never reads live browser databases directly; instead, it stages safe copies, deduplicates, and appends to a local archive that you fully own and control.
On top of that archive, PathKeep gives you powerful recall (full-text search, regex, timeline, filters, export) and deterministic Core Intelligence from local archive facts. Semantic search, the AI assistant, MCP, and other AI-backed surfaces are v0.2 roadmap items, not v0.1.0 promises.
Use Chrome with Google Sync enabled? PathKeep supports Google Takeout import, letting you recover extended history (often ~18 months, depending on your Google account settings) instead of just the local ~3 months.
Download the latest release from GitHub Releases.
- macOS: open the
.dmg, movePathKeep.appto/Applications, then open it. Safari Browser Direct import requires granting Full Disk Access to PathKeep before scanning Safari history. - Windows: install the unsigned
.msior-setup.exerelease package. Windows will showUnknown Publisher, and SmartScreen may require More info -> Run anyway until PathKeep has publisher reputation. Scheduled backups use Windows Task Scheduler. The installer bundles the WebView2 offline installer, but Windows Server Core / headless Server environments are not valid GUI acceptance hosts. - Linux: Support is on the way. Build from source if you want to try it out right now.
install the.AppImage,.deb, or.rpmartifact. Linux scheduled-backup support is still preview/manual-review because desktop keyring andsystemd --userbehavior varies by distribution.
- Remove any installed schedule from System -> Scheduled Backup Settings.
- Quit PathKeep, then delete the app or uninstall it with your OS package manager.
- Optional local data removal: delete the PathKeep app-data directory only if you also want to remove the archive, config, audit artifacts, and derived indexes. On macOS that directory is
~/Library/Application Support/com.yi-ting.pathkeep.
PathKeep is organized around three functional pillars, built in order of priority:
┌─────────────────────────────────────────────────────┐
│ INTELLIGENCE │
│ Core insights now · AI surfaces on v0.2 roadmap │
├─────────────────────────────────────────────────────┤
│ RECALL │
│ Full-text search · Timeline · Filters · Export │
├─────────────────────────────────────────────────────┤
│ ARCHIVE │
│ Incremental backup · Schedule · Security · │
│ Import · Audit · Encryption │
└─────────────────────────────────────────────────────┘
The foundation. Everything else depends on a trustworthy archive.
- Incremental backup — staged database copies (never reads live browser DBs), append-only archive, automatic deduplication
- Multi-browser discovery — auto-detects installed browsers and profiles; you choose which to back up
- Scheduled backups — native install / status / remove support for macOS (
launchd) and Windows (Task Scheduler), with Linuxsystemd --userpreview kept manual-review - Google Takeout import — preview, import, revert, restore, and repair flows with full dry-run
- Encryption — plaintext or SQLCipher-encrypted archive, with re-key preview and audit trail
- Audit ledger — every backup, import, rollback, and restore leaves an immutable run record with manifests and artifacts forming a hash chain
- Rollback — any write operation is reversible; user-visible facts use soft-hide visibility, not destructive delete
Finding what you've seen before, across years of history.
- Full-text search — FTS5-powered keyword search across URLs, titles, and search terms
- Regex search — optional regex mode for advanced pattern matching, post-filtered on canonical results
- Interactive timeline — year → month → day drill-down with density visualization and virtual scrolling for millions of records
- Composite filters — by browser, profile, domain, time range, page type, visit source, or import batch
- Export — filtered result sets exportable to HTML, Markdown, plain text, or JSONL
Understanding your browsing patterns, built on top of a solid archive. AI features are deferred for v0.1.0 — PathKeep works from local archive facts without any AI provider.
- Deterministic insights — browsing rhythm calendar heatmap, search activity, domain deep-dive, sessions, search trails, query families, refind pages, activity mix, and periodic summaries — all computed from archive facts, no AI required
- Semantic search — deferred to the v0.2 roadmap; v0.1.0 ships keyword and regex recall only
- AI assistant — deferred to the v0.2 roadmap until provider, retrieval, evidence, and lock-state flows are truth-tested
- MCP server — deferred to the v0.2 roadmap; v0.1.0 does not expose browsing history to external AI tools
- Insight cards — topic timelines, task/thread detection, browsing rhythm, explore-vs-exploit patterns, source effectiveness, and contrastive summaries from deterministic local models
- Remote backup — Preview → Manual → Execute flow for S3-compatible remote bundles, with checksum and restore-readiness verification
PathKeep separates implemented adapters from publicly validated support. The README only promises what has been independently verified.
| Status | Browsers |
|---|---|
| Validated | Google Chrome; Microsoft Edge / Edge Dev; Firefox; ChatGPT Atlas (macOS); Perplexity Comet (macOS); Safari (macOS, requires Full Disk Access) |
| Implemented | Chromium, Brave, Vivaldi, Arc, Opera, Opera GX, LibreWolf, Floorp, Waterfox |
Implemented browsers appear in discovery and archive data but are not yet in the public support promise. See the adapter playbook for the promotion gate.
| Platform | Status | Notes |
|---|---|---|
| macOS | Primary | Signed / notarized builds; Touch ID session unlock; Safari support requires Full Disk Access |
| Windows | Preview | Unsigned MSI / NSIS installers; Unknown Publisher is expected; Task Scheduler supported |
| Linux | Preview | AppImage / .deb / .rpm builds available; keyring behavior varies by desktop environment |
| Layer | Choice | Why |
|---|---|---|
| Desktop framework | Tauri 2 | Cross-platform, Rust core, lightweight |
| Core logic | Rust workspace (vault-core, vault-worker, vault-platform) |
High performance, safe, cross-platform |
| Browser parsing | browser-history-parser (standalone Rust crate) |
Reusable, community-publishable parser |
| Frontend | React 19 + TypeScript + Vite | Modern, type-safe |
| Toolchain | Bun | Package management and scripts |
| Canonical storage | SQLite (optional SQLCipher encryption) | 20-year durability, local-first |
| Full-text search | SQLite FTS5 | Core recall, no external service |
| Vector / semantic | Deferred for v0.1.0 | Future replaceable sidecar, not in default build |
| AI inference | Deferred for v0.1.0 | Future user-configured providers |
- Bun
- Rust
1.94.1withclippy,rustfmt, andllvm-tools-preview - Git
- Tauri 2 platform prerequisites
Linux (Debian / Ubuntu) development packages:
sudo apt-get update
sudo apt-get install -y \
pkg-config libglib2.0-dev libgtk-3-dev \
libwebkit2gtk-4.1-dev libayatana-appindicator3-dev \
librsvg2-dev patchelf rpmbun install
bun run dev # Browser-only Vite preview (127.0.0.1:1420)
bun run desktop:dev # Full Tauri desktop appbun run build # TypeScript + Vite bundle
bun run desktop:build # Release desktop bundlebun run check # All mainline quality gates
bun run build # TypeScript + Vite bundle
bun run test:unit # Vitest unit tests
bun run test:e2e # Playwright end-to-end tests
bun run coverage:js # JS coverage gate
bun run coverage:rust # Rust coverage gate
bun run mutation:js # Desktop-contract JS mutation gate
bun run mutation:js:full # Full JS mutation deep sweep
bun run mutation:rust # Full Rust mutation deep sweep
bun run verify # check + debug desktop build rehearsalFor the full gate matrix, deep checks, and release signoff commands, see TESTING.md.
| What you need | Where to look |
|---|---|
| Contributor workflow | CONTRIBUTING.md |
| Local environment and repo layout | DEVELOPMENT.md |
| Test surfaces and command matrix | TESTING.md |
| Release runbook and artifact matrix | RELEASE.md |
| User-facing troubleshooting | TROUBLESHOOTING.md |
| Support and bug-report expectations | SUPPORT.md |
| Browser support and adapter promotion | docs/architecture/browser-support-and-adapter-playbook.md |
| Product vision, features, and design | docs/ |
PathKeep uses Conventional Commits, colocated tests, and doc-first updates. Start with CONTRIBUTING.md.