Context Mole v0.3.1

Context Mole v0.3.1

This patch keeps indexing progress readable while work is running and makes successful work easier to see.

Stable indexing status

• The live indexing panel has a fixed layout, so rapid item updates do not push project content up and down.
• Inventory and search coverage keep their last verified values while a refresh is pending. A routine refresh no longer clears the cards or flashes a loading state on each poll.
• Progress shows actual successful jobs completed during this app session, with the last successful completion date and time. This is a job count, not a unique-file total, and it resets when the app restarts.
• Queued, running, paused, CPU-waiting, retrying, and failed work remain distinct. Existing indexed coverage can stay unchanged while a repair is in progress.
• Startup text now distinguishes routine index checks from upgrades that are needed.

Install and update

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Existing installed Windows builds can obtain this patch through their normal update check.

This patch does not introduce a database schema or extraction-layout migration. Windows releases remain unsigned, so SmartScreen may show an unrecognized-app warning. Native Linux and macOS packages are not included.

---

Full Changelog: v0.3.0...v0.3.1

Context Mole v0.3.0

Context Mole v0.3.0

Context Mole v0.3.0 adds clearer project status, durable document issues, safer index upgrades, and more reliable evidence across mixed document formats.

Project status and document issues

• Project summaries distinguish available search coverage, active indexing, queued work, and unresolved failures. Stale status is labelled with its last successful check instead of appearing current.
• Document issues are grouped by source file, including failures inside email attachments and archives. Larger issue lists support paging and path filtering while keeping the total scope visible.
• Retry keeps an unresolved issue visible until a successful result replaces it. Hide affects only that issue's visibility, with Undo; unchanged failures remain hidden after a retry, while changed source content can surface a new issue.
• Exclude removes a chosen file's indexed evidence and pending work without modifying the source file. Exclusions persist across restarts and atomic file replacement. Include removes the exclusion and queues fresh indexing.
• Startup and recovery screens keep error details and local recovery paths available. Keyboard project reordering preserves selection and focus through repeated moves.

Model, OCR, connection, and update feedback

• Semantic search uses the pinned Granite Multilingual 97M model. Download, validation, cancellation, and failure states explain their effect on search; existing verified model files are retained safely.
• OCR progress and failures explain which files are affected. Unsupported platforms and unavailable models are identified explicitly rather than presented as complete coverage.
• Optional AI clients remain neutral until chosen or configured. Connection actions preserve unrelated settings, surface conflicts, and explain the local configuration changes they make.
• Installed Windows builds provide clearer update progress, retry, failure details, and the last attempted check. Restart remains deferred while indexing is active.

Search and mixed-format evidence

• Structured search and evidence responses expose clearer source, attachment, location, confidence, and retrieval information, with bounded context and two previews by default.
• Recognized spreadsheet dates include canonical dates alongside raw serial values. Uncached formulas remain labelled as formulas instead of inventing a calculated result.
• Standalone and attached image OCR uses bounded context grouping. Frozen mixed-format fixtures cover documents, spreadsheets, images, email, attachments, and repeated evidence.

Local-data upgrade

This release upgrades the local database to schema 9 and introduces the current extraction preparation. Existing project and folder configuration is retained. Source-backed re-extraction may run in the background; retained active evidence remains available until a successful replacement commits, including through paused projects or a failed rebuild.

Older semantic vectors are used only when both their model policy and extraction preparation are compatible. Keyword and literal evidence remain available while semantic coverage is repaired. Startup removes only allowlisted retired 311M installer-cache files under the managed-cache policy; other models, unknown cache files, and source files are preserved.

These safeguards do not recover evidence already deleted by an earlier application version. The bundled index upgrade and recovery guide describes the boundaries. Older releases may reject an index already upgraded to schema 9; keep an index backup before attempting a downgrade.

Install and update

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Existing installed Windows builds can obtain this release through their normal update check.

Windows releases remain unsigned, so SmartScreen may show an unrecognized-app warning. This release provides a Windows x64 installer and update packages; native Linux and macOS packaging is not included.

---

Full Changelog: v0.2.9...v0.3.0

Context Mole v0.2.9

Context Mole v0.2.9

Context Mole v0.2.9 fixes MCP connections launched by Windows MSIX clients reading a separate, stale copy of the desktop index.

• Detects Windows AppData redirection before accessing the index, broker authentication, or lifecycle files.
• Automatically starts the MCP server in the same user's desktop runtime while preserving the client's stdio connection and environment settings.
• Uses private named pipes, keeps diagnostics on stderr, and releases the worker when the client disconnects.
• Isolates the updated shared broker from legacy brokers that may still be using a private cache.
• Logs the requested and physical data directories and the running MCP version to make future path problems visible.

Your existing index, models, and settings remain in their current directory. Update Context Mole and reload your AI client's MCP connection; no reindex or configuration change is required.

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64.

---

Full Changelog: v0.2.8...v0.2.9

Context Mole v0.2.8

Context Mole v0.2.8

Context Mole v0.2.8 improves database startup and diagnostics, and measures document processing time without queue waits.

Database startup and MCP diagnostics

• The desktop app completes database migrations and interrupted-job recovery before opening its normal interface.
• Migration history is validated before changes begin. Each migration runs in its own transaction, and a failed migration can be retried on the next launch.
• Startup failures show a clear error with the database and log paths, preserving the existing index for diagnosis.
• Logs include application versions, resolved paths, current and expected schema versions, migration outcomes and durations, and database readiness.
• The MCP broker preserves the underlying database error. Missing indexes, incompatible schemas, and access, locking, or corruption failures are reported distinctly.

Document processing time

• Queued documents and files waiting for processor capacity show no running timer.
• Processing time starts when processor capacity has been acquired and excludes availability checks and queue waits.
• The completed average uses successful attempts' processing time for the current session. Failed and canceled attempts are excluded.
• The same timing rules apply when embeddings are refreshed.

Update and reconnect

This release includes updated desktop, MCP server, and broker binaries. After updating, open Context Mole to complete database startup, then reload your AI client's MCP connection so it launches the updated server.

Daily UI and broker logs are available in %LOCALAPPDATA%\ContextMole\logs, or in the logs folder of your custom data directory.

Install

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Releases are currently unsigned, so Windows SmartScreen may show an unrecognized-app warning.

---

Full Changelog: v0.2.7...v0.2.8

Context Mole v0.2.7

Context Mole v0.2.7

Context Mole v0.2.7 improves indexing efficiency, makes project progress easier to understand, and strengthens pause and restart recovery.

OCR and text extraction

• PDF OCR reuses renderers and sends page pixels directly to the OCR engine, reducing repeated work and memory copies while preserving page resolution.
• Page images are allocated only when OCR can process them, keeping queued files from holding large rendered images in memory.
• Long text lines keep their proportions during recognition. A historical scan used in testing now yields 1,455 characters instead of 112.
• Spreadsheet rows are read incrementally, email resources are released promptly, and image/TIFF OCR timeouts appear as current issues.

Clearer project progress

• Separate Up to date, Processing, Queued, and Needs attention counts account for all discovered files.
• Scheduled retries and files waiting for processor capacity are included in the queue. Searchable files are counted separately because previous indexed versions remain available during updates.
• Current issues can be collapsed and browsed in pages of 25. Resolved failures clear, and large projects retain every outstanding issue.
• Unavailable or unreadable folders have a clear status that clears after a successful rescan.

Recovery and performance

• Folder discovery runs alongside indexing, and large bursts of file changes use bounded buffers with a rescan when needed.
• Ordinary change notifications verify file content before repeating extraction. Reindex still forces a complete rebuild.
• Pause, shutdown, and restart retain unfinished jobs and the last searchable revision. Shutdown also waits for discovery cleanup when cancelled during model startup.
• Similar-length passages are grouped for both FP32 and quantized embeddings. Local benchmarks measured 2.2–2.4× faster FP32 processing and about 1.5× faster quantized processing. Quantized retrieval relevance stayed stable or improved overall on the labeled test workload; individual ranking differences and workload limits are documented.
• Large embedding requests use smaller transfers, and outdated vector caches are released when searches switch to streaming.

Validation and benchmarks

The Release build and all 282 automated tests passed locally. Five checked-in public fixtures cover native PDF, Word, scanned PDF, JPEG, and multilingual TIFF extraction, with source attribution, licenses, hashes, and quality checks. All five fixture checks passed with the pinned OCR models.

See benchmark results and commands to reproduce the measurements. Performance varies with the documents and machine.

Use Reindex on existing projects to apply the improved OCR to files already indexed.

Install

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Releases are currently unsigned, so Windows SmartScreen may show an unrecognized-app warning.

---

Full Changelog: v0.2.6...v0.2.7

Context Mole v0.2.6

Context Mole v0.2.6

Context Mole v0.2.6 makes OCR deadlines fair under processor contention and clarifies how project dashboard totals relate to one another.

OCR reliability

• OCR timeouts now begin only when a request has processor capacity, owns the serialized OCR slot, and is ready to run inference.
• Time spent downloading or preparing models, waiting behind another OCR request, or waiting for processor capacity no longer consumes the OCR deadline.
• Cancellation from pausing, shutdown, or the indexing pipeline continues to interrupt waits and active OCR promptly.

Clearer project totals

• The dashboard now labels successfully indexed files as Searchable now and error records as Unresolved issues.
• A concise explanation and detailed tooltips clarify that total files is a unique file count, searchable files can overlap queued or processing work while their last good index remains available, and one file can have multiple unresolved issues.

Download

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Releases are currently unsigned, so Windows SmartScreen may show an unrecognized-app warning.

---

Full Changelog: v0.2.5...v0.2.6

Context Mole v0.2.5

Context Mole v0.2.5

Context Mole v0.2.5 keeps meaning-based search available while projects transition between embedding policies and makes partial semantic coverage visible and repairable.

Mixed embedding search

• Semantic and hybrid searches continue using documents whose embeddings match the active model instead of disabling semantic retrieval for the entire project.
• Hybrid keyword retrieval still covers the full searchable index while the semantic branch searches only the compatible subset.
• MCP responses report semantic_partial_coverage with compatible, excluded, and repair status details when coverage is mixed.
• If no compatible embeddings exist, semantic search remains unavailable and hybrid search safely falls back to keyword results.

Coverage status and repair

• The project screen shows how many indexed files currently have compatible embeddings.
• The status distinguishes partial coverage from repairs that are already queued in the background.
• A focused repair action queues only files with missing, incomplete, or outdated embeddings; keyword data and source files are not rebuilt or modified.
• Background policy refresh now evaluates compatibility against the active embedding model so mixed generations converge automatically.

Reliability

• Vector snapshots and streaming semantic retrieval are scoped to one explicit embedding policy, preventing incompatible vectors from entering the same search branch.
• Automated coverage verifies that compatible documents remain semantically searchable while legacy documents remain available through keyword search and await repair.

Install

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Releases are currently unsigned, so Windows SmartScreen may show an unrecognized-app warning.

For the complete search schema, examples, data behavior, and manual MCP setup, see the README.

---

Full Changelog: v0.2.4...v0.2.5

Context Mole v0.2.4

Context Mole v0.2.4

Context Mole v0.2.4 restores CPU-driven indexing parallelism, keeps native-memory cleanup without admission throttling, and fixes retry and development-build reliability.

Indexing throughput and memory cleanup

• Document parallelism is derived from the selected CPU profile and the machine's logical CPU count, without a separate memory-admission limiter reducing work to one file.
• OCR remains serialized while temporarily receiving the full processor capacity allowed by the selected profile.
• Per-operation resources are disposed promptly, OCR and embedding sessions still unload after idle periods, and disposable vector caches are cleared under genuine system-memory pressure.
• Indexing activity now reports processor-capacity waits without misleading memory-admission or exclusive-file states.

Failed-file retry

• Retry now distinguishes files newly queued from files that are already queued or processing, instead of reporting that no failed files needed retrying.
• Error cards remain until the queued retry succeeds, and the status message now explains that behavior.
• Errors for files removed from the project are cleared so they cannot leave a permanently ineffective Retry button.
• Transient embedding-refresh interruptions remain eligible for automatic recovery.

Development and build reliability

• Development brokers run from immutable staged snapshots outside project build output, preventing the shared broker from locking assemblies during subsequent builds.
• Concurrent clients continue to share one broker process, while rebuilt development payloads receive a fresh staged deployment.
• The repository now pins the .NET 10.0.400 SDK used by local and release builds.

Install

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Releases are currently unsigned, so Windows SmartScreen may show an unrecognized-app warning.

For the complete search schema, examples, data behavior, and manual MCP setup, see the README.

---

Full Changelog: v0.2.3...v0.2.4

Context Mole v0.2.3

Context Mole v0.2.3

Context Mole v0.2.3 fixes background-start window restoration on Windows and restores safe parallel document parsing.

Windows startup and tray

• Automatic launches still start quietly in the system tray, but the initial hide is now one-time.
• Showing Context Mole from the tray now keeps the window visible, including after it has previously been shown and closed back to the tray.

Parallel indexing

• PDFs, images, Office files, email messages, and archives can now enter extraction concurrently when the selected CPU profile and available memory allow it.
• Memory admission reserves one shared OCR peak instead of serializing every document that might contain OCR-capable content.
• OCR peaks remain serialized and account for every active document reservation, preserving the operating-system safety reserve.
• OCR workers yield their normal processor slot before requesting full OCR capacity, preventing processor and memory admission from waiting on each other.
• Activity details now distinguish serialized OCR waits from ordinary process or system-memory limits.

Install

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Releases are currently unsigned, so Windows SmartScreen may show an unrecognized-app warning.

For the complete search schema, examples, data behavior, and manual MCP setup, see the README.

---

Full Changelog: v0.2.2...v0.2.3

Context Mole v0.2.2

Context Mole v0.2.2

Context Mole v0.2.2 substantially reduces memory use when several AI clients are connected and makes background indexing easier to understand and control.

Shared AI connection runtime

• Each configured AI client now launches a lightweight MCP adapter that connects to one on-demand broker shared by the current Windows user.
• Granite model sessions and the vector-index cache are shared instead of being duplicated across every connected client.
• The broker unloads idle model and cache state after two minutes and exits after ten minutes without requests or work.
• Packaged and locally staged MCP deployments include one version-matched broker with guarded startup, shutdown, and upgrade handling.

Memory-aware indexing

• Indexing admits work through one global memory budget using file type, file size, parsing, OCR, and embedding estimates.
• Large files no longer cause all indexing to stop: at least one eligible job can continue while additional work waits for safe capacity.
• Memory and processor admission are coordinated so waiting jobs do not hold scarce resources unnecessarily.
• Active-indexing details now distinguish admission queues, memory waits, processor waits, and real parsing work, including estimates and available-memory context.
• Native OCR and embedding runs respond to cancellation and release temporary memory more aggressively.

Pause, retry, and project status

• Pausing a project immediately requests cancellation of its running indexing work and durably returns interrupted files to the queue without recording false failures or consuming retry attempts.
• Late parser, embedding, and storage callbacks cannot commit work after the project has been paused.
• Project summaries now separate files actively processing from queued files and scheduled retries, so the totals remain meaningful.
• Retrying means a retry is running now; Retry scheduled means it is waiting for its next eligible time.
• Error cards describe failed attempts explicitly instead of showing an ambiguous attempt number.

Install

Download JPCodeCraft.ContextMole-win-x64-Setup.exe for Windows 10+ x64. Releases are currently unsigned, so Windows SmartScreen may show an unrecognized-app warning.

For the complete search schema, examples, data behavior, and manual MCP setup, see the README.

---

Full Changelog: v0.2.1...v0.2.2