Repository navigation
Releases
TRACTOR 0.5.1 — Browser Search
TRACTOR 0.5.1 — Browser Search
Tractor 0.5 adds the local recursive search service used by Orc and a public in-browser results page.
0.5.1 fixes a service startup race: when a previous instance's service still held the shared local port (a predecessor from another data directory cannot be joined, only outlived), a starting service failed instead of waiting for the port to free. Startup now retries binding and registering for up to 30 seconds, and the local token file's creation race is handled the same way. It also fixes a startup hang on machines whose hostname does not resolve (common on macOS): binding no longer reverse-resolves the hostname before the port opens.
- DuckDuckGo provides general web result snippets. Destination websites are not crawled by the search adapter.
- Wikimedia Commons provides image results with thumbnails, file pages, creators, and license details.
- Search jobs use the existing recursive investigation engine and save progress to Tractor's local history.
- The service binds to loopback, uses a per-user local token, and shuts down after the Orc processes that started it close.
- Tractor's application icon is a flying saucer, including the Windows and macOS native application assets.
Provider availability, result ranking, and coverage can vary. The engine does not claim exhaustive web coverage.
Native bundles and checksums are built only after the platform checks complete.
TRACTOR 0.5.0 — Browser Search
TRACTOR 0.5.0 — Browser Search
Tractor 0.5 adds the local recursive search service used by Orc and a public in-browser results page.
- DuckDuckGo provides general web result snippets. Destination websites are not crawled by the search adapter.
- Wikimedia Commons provides image results with thumbnails, file pages, creators, and license details.
- Search jobs use the existing recursive investigation engine and save progress to Tractor's local history.
- The service binds to loopback, uses a per-user local token, and shuts down after the Orc processes that started it close.
- Tractor's application icon is a flying saucer, including the Windows and macOS native application assets.
Provider availability, result ranking, and coverage can vary. The engine does not claim exhaustive web coverage.
Native bundles and checksums are built only after the platform checks complete.
TRACTOR 0.4.0 — Coverage
TRACTOR 0.4 — Coverage
This release broadens source discovery while keeping original evidence, explicit limits and resumable searches.
- Optional Brave, Mojeek and current Kagi v1 search APIs; Marginalia with shared-development/private credentials.
- Expanded SearxNG upstream metadata, visible partial failures and resumable pages.
- Common Crawl capture metadata for known URLs/domains, without destination or WARC downloads.
- Provider capabilities, grouped Settings, profiles, language/freshness controls, configurable bounded budgets and a privacy screen.
- Local provider health, independent-index discovery counts, retained duplicate discovery paths, expanded query candidates and 25-language Wikidata labels.
- CLI provider diagnostics, coverage reports, refresh and expanded exports; versioned offline benchmark.
- Authenticated caches stay in memory; keys are redacted from storage, logs and exports. Existing database history migrates automatically.
Download the native archive matching your system and architecture, extract it fully, and open Tractor. Native builds are unsigned; Tor remains an optional external dependency. Source archives and a Python wheel are also attached. SHA256SUMS.txt covers each artifact.
Known limits: paid APIs were contract-tested with fixtures when keys were unavailable. The public Marginalia key can be rate-limited. Search snippets are not full documents, index agreement is not factual confirmation, and no search is exhaustive. Claims, source-independence analysis, document/PDF retrieval, timelines, translation integration and large-scale persistence redesign remain later milestones. See docs/roadmap.md.