Repository navigation
Releases
v0.4.0-preview.3 - Experimental preview
Experimental, unsigned engineering preview — not stable-release acceptance. GitHub's Latest label identifies the most recent preview download, following this repository's existing convention. Windows 11, local WPR save and long-run WPF retention checks remain open.
Changes since preview 2
- Review report fields before export, with conservative defaults for names, paths, endpoints, command lines and free text. Field selection is not anonymization.
- Scope sampled CPU analysis to captured thread lifetimes and navigate directly to the hottest sampled call path and its evidence.
- Improved recorded-event filters, process-family history, interval comparisons and CPU-stack exploration.
- Monitor cancellation/coverage, keyboard navigation, accessible names and SDK-provenance fixes.
- Official WPF automation item-lifetime compatibility mitigation, with a repeatable actual-app/UIA diagnostic. The original intermittent race remains unreproduced; this is a bounded mitigation with retention risk.
- WPR timeout/cancellation retain bounded native diagnostics. Saving skips optional dynamic .NET PDB generation; managed-symbol availability may be reduced. Capture providers, protected storage, ownership and the 60-second deadline are unchanged.
Download
Extract WindowsOpenIntelligence-win-x64.zip into a local folder and run WindowsOpenIntelligence.exe. Keep all files together; no Visual Studio or .NET SDK is needed. Offline analysis and observation run unprivileged. Guided WPR capture requests administrator elevation for the same Windows account.
The Windows ZIP is the unmodified, hash-verified artifact from the successful PR workflow for commit 5b8115b. Source ZIP and SHA-256 sidecars are included. Checksums establish integrity, not publisher identity; the executable is unsigned.
Verified for this commit
- PR CI and push CI passed: 171 tests each, zero failed/skipped, plus hosted native capture profiles/lifecycle, trace-size limits, desktop/recovery, monitor and replay checks.
- Downloaded package checksum and all 435 dependency asset hashes verified; 596 archive entries had no private/unsafe paths under package rules. Dependency identity/advisory checks had no known advisory matches at verification time.
- Extracted downloaded package passed local Windows 10 desktop smoke with an external accessibility client, 57 accessible names, worker recovery, monitoring, investigations and feature assertions.
- Detailed verification and limits.
Known limitations
Local Windows 10 WPR Resources saving previously timed out; the revised command's retry was cancelled at elevation before recording, so local save acceptance is pending. The WPF change mitigates a proven weak-item failure mechanism, but the original application race, complete window recovery and long-run retention remain open. Historical memory, continuity and observer-cost misses are preserved.
Windows 11, controlled performance/retention, sleep/resume, independent physical accessibility/usability, service-side temporary writes, isolated privileged trust/storage testing and protected installation/signing remain pending. Hosted successes do not close those gates. PR #2 remains draft and unmerged; publishing this experimental snapshot does not imply stable acceptance.
v0.4.0-preview.2 - Experimental preview
Experimental, unsigned preview. This is marked as GitHub's Latest download so it appears in the repository sidebar; it has not passed stable-release acceptance. Independent Windows 11, actual sleep/resume, elevated security validation, signing and repeated long-run gates remain open.
Windows Open Intelligence is a local Windows performance investigation utility. This preview improves the honesty of resource-history charts when collection is interrupted or measurements are unavailable.
Changes since preview 1:
- Charts break across collection pauses, including memory gauges.
- Downsampled history leaves a metric blank throughout any block containing missing measurements; original snapshot evidence remains unchanged.
- Combined read/write I/O stays unavailable when either counter is absent.
- Nine regression cases cover pauses, failed sampling, clock jumps, chart gaps and evidence preservation; the suite now contains 125 tests.
Download WindowsOpenIntelligence-win-x64.zip, extract it, and run WindowsOpenIntelligence.exe. Windows x64 only; no SDK or Visual Studio installation is needed. Observation and offline analysis run without elevation; guided WPR capture requests UAC approval. SHA-256 sidecars verify download integrity, not publisher identity.
This is an unsigned engineering preview. Independent Windows 11, actual sleep/resume, elevated attack validation, signing and repeated long-run acceptance remain open. It is not a stable release.
Validation for source 5ba9bdfe917ec27726210e14249a5652daa3d2dc:
- Release-tag CI and main CI passed every workflow step, including 125 tests, dependency identity/advisory and tampering checks, both WPR profiles, capture recovery, generated ETL scale/rejection, packaged desktop workflow and replay stress.
- Tag-build replay contributor/page p95: 135.01 / 50.31 ms. The generated near-limit trace indexed 4,990,003 events and met all recorded import/reopen/page/memory/derived-size gates.
- Warmed headless observation: 0.312% machine CPU, 48.3 MiB peak including startup/warm-up, no measured gap episodes. Short Windows Server checks do not establish whole-UI, Windows 11, observer-effect or long-run acceptance.
- The provided source archive builds independently with zero warnings/errors. The Windows ZIP is the unmodified release-tag CI artifact.
Windows Open Intelligence 0.4.0 preview 1
Windows Open Intelligence helps you investigate recorded Windows activity, compare resource contributors, and inspect the events behind each finding — locally on your PC.
Engineering preview · Windows x64 · MIT. This is a prerelease for experimentation and feedback. It is unsigned and is not a stable or independently certified diagnostic/forensic tool.
In this preview
- Open ETL recordings, select an interval, rank CPU/disk contributors, and inspect process lifetimes, events and captured stacks.
- Observe recent resource history without elevation, save investigations, and compare equivalent attempts with explicit evidence limits.
- Use bounded, fixed WPR profiles through a separate elevated helper; protected collector storage stays on the application's fixed local NTFS volume.
- Import/reopen caches and save investigations in deeply nested Windows paths.
- Package privacy exclusions and dependency identity/advisory checks run before publication.
- See genuine Summary, Explorer and Evidence views in the README gallery.
Validation
The complete Windows verification run passed on commit 02770d2e5d3cacaeaf837ee8ff54772579caf7b9: 116 tests, independent CPU/disk reference comparisons, package privacy and dependency checks, generated ETLs near/above the five-million-event limit, both shipping capture profiles, deadline/cancel/disconnect/crash-recovery/coexistence, aggregation stress, packaged desktop recovery, monitor continuity and warmed measurement, and replay stress.
The near-limit trace indexed 4,990,003 events and passed its recorded import/query/memory/derived-file targets. Replay handled 228 requests and 50,000 paged events with two restarts; contributor/first-page p95 was 130.04/66.13 ms on this runner. These are fixture measurements, not performance guarantees for every recording or machine.
The Windows application ZIP below comes from that successful CI run. Its checksum, private-file exclusions and actual dependency bytes were checked again after download. The source ZIP was exported from the same commit and independently built with zero warnings/errors. Native capture passed on the hosted Windows Server runner; independent Windows 11 and privilege-boundary acceptance remain open.
The automatic preview-tag workflow also passed every step. Both published ZIPs were downloaded anonymously and their checksums verified. The downloaded Windows app passed local Windows 10 launch, worker recovery, observation, saved-investigation and comparison smoke checks.
Run
Download WindowsOpenIntelligence-win-x64.zip, verify its companion SHA-256 file, extract to a local folder, and open WindowsOpenIntelligence.exe. Keep all runtime files together; no SDK or Visual Studio is needed. The source ZIP includes build instructions and dependency locks.
Start the UI from an ordinary desktop session. Recording requests same-account administrator elevation, uses a fixed local NTFS volume, and requires at least 2 GiB free. Review traces, screenshots and reports before sharing them: they can contain process names, paths, command lines and endpoints.
Remaining limits
Windows 11 x64 is the release target; independent desktop acceptance and elevated same-SID storage attack validation remain open. Service-side temporary writes and portable helper signing/trust still need review. SkiaSharp 2.x is unmaintained upstream; see the dependency assessment.
Recorded resource attribution is not proof of why an application froze. Missing providers, incomplete stacks, unavailable estimates and event loss remain explicit. Broader performance, long-run retention, sleep/resume and usability gates are documented in validation.
Created and maintained by ard12. Reproducible issues, Windows compatibility reports and contributions are welcome.
Windows Open Intelligence 0.3.0 preview
Windows Open Intelligence 0.3.0 preview adds bounded guided WPR recording with a separate elevated helper, local import, lifetime-aware contributor analysis, and evidence/report export. The self-contained Windows ZIP includes the UI, worker, and helper. The source ZIP contains the reviewed GitHub source set and dependency locks.
Local verification: 58 tests passed; the extracted Windows ZIP passed desktop rendering, worker recovery, and damaged-trace exit checks. Real ETL stress covered 30 cycles, 578 requests, 150,000 paged events, cancellation and crash recovery. Five million generated observations passed aggregate totals and exact-interval checks.
This is a prerelease because native WPR capture-to-report testing could not complete locally after UAC was canceled. GitHub Actions did not start a runner: GitHub reported a failed account payment or spending-limit block. The workflow remains configured to require native capture validation before publishing a CI artifact. Contributor queries measured 514 ms p95 in the broader 12-cycle stress run, above the 300 ms target; one older ETL differs from native CPU-weight estimates by 1.55%, above the 1% agreement target. Windows 11 and maximum-envelope ETL testing remain pending.
The original ETL is never included in the release. Reports can contain captured paths, command lines, endpoints, and addresses; review them before sharing. SHA-256 sidecars are attached for both ZIPs.