Product updates

What's new in FPGAi

Approved release notes for customer-facing features, workflow improvements, and important platform changes.

CLI agent

v0.1.5View release

VS Code extension

v0.4.5View release
Improved

PDF imports preserve readable text

FPGAi now keeps useful text from PDFs even when a document also contains diagrams, scanned pages, or image-only content.

PDF understanding

  • PDFs with selectable text and visual pages now keep both the recovered text and vision-based descriptions, so diagrams do not replace searchable content.
  • Scanned PDFs can retain OCR-recovered text alongside visual-page captions, while genuinely image-only PDFs remain available through vision indexing.

Recovery

  • If the first PDF enrichment pass times out or returns incomplete output, FPGAi makes a bounded text-recovery attempt before deciding how to index the document.
  • The ingester reports recovered page counts and parser details when recovery is needed.
Improved

Large PDF retries keep completed work

When a long PDF run is interrupted, FPGAi now resumes completed visual-page work instead of sending those pages again.

Recovery

  • Completed visual-page classifications and captions are checkpointed locally, so an unchanged PDF retry only calls pages that still need work.
  • The ingester Jobs panel lets you remove failed history rows without touching ready packages.

Control

  • The source timeout shown in the ingester follows the server's admin ingestion setting when connected.
Improved

Knowledge-base imports recover more reliably

Large knowledge-base imports now handle temporary processing interruptions more gracefully, so successful work is less likely to be lost during a handoff.

Reliability

  • Temporary vision-service interruptions are retried automatically before a page is marked unsuccessful, and successful pages remain available when a later page fails.
  • Package pushes retry temporary server or network interruptions and keep the package ready when a retry still needs attention.

Package handoff

  • An uploaded .fpgrai package is treated as a complete snapshot, so it can be pushed directly without rerunning local source cleanup.
Improved

Vision ingestion now explains what happened

Large PDF ingestion runs now process up to 1,000 visual pages and show clear diagnostics when a page is skipped, capped, or fails.

Visibility

  • The local ingester now keeps a persistent activity log with timestamps, pipeline stages, errors, and reasons a step did not run.
  • Package inspection shows how many visual pages were detected, rendered, captioned, failed, or left outside the configured limit.

Capacity

  • Vision indexing supports up to 1,000 visual pages per PDF by default; when a document is larger, completed pages are retained and the remaining pages are clearly marked.
Fixed

Website error screen after platform updates

Some visitors could see an "Unexpected Application Error" screen when opening fpgaitool.com shortly after we shipped an update. The site now recovers automatically.

Fixed

  • Browsers no longer keep an outdated copy of the site after we deploy an update, which was the cause of the error screen.
  • If a browser still holds an outdated copy from before this fix, the site now reloads itself once and continues normally instead of showing a raw error page.
Improved

RAG ingestion is easier to monitor and recover

Large local and Google Drive ingestion runs now show what is happening file by file, remain usable while work continues, and give you control when a run needs to stop.

Visibility

  • Google Drive downloads show a live progress bar before parsing, so you can tell whether a file is still downloading or already being indexed.
  • Each file reports its current pipeline stage, completion details, and its own error instead of hiding a batch failure behind a frozen page.

Control and recovery

  • Stop an entire ingestion run or cancel selected files while the rest of the batch continues.
  • If the local ingester restarts, unfinished work is clearly marked for re-ingestion rather than disappearing silently.
  • Existing server documents are recognized more reliably, including files whose local staging name previously contained an internal identifier.
Improved

Reliability and security hardening across the platform, CLI, and VS Code extension

Sign-in with Google or GitHub is more dependable, multi-file apply in VS Code is more resilient, and the platform received a broad round of security hardening.

CLIv0.1.5VS Codev0.4.5

Improved

  • Signing in with Google or GitHub no longer fails after spending a long time on the provider’s consent screen.
  • A chat request that fails before the answer starts streaming no longer counts against your monthly quota.
  • Platform-wide security hardening based on an in-depth review.

VS Code extension 0.4.5

  • Applying a multi-file response without the CLI installed now applies every valid file and reports a problematic file individually, instead of stopping the whole apply.

CLI agent 0.1.5

  • Security hardening and reliability improvements for generated-file apply and the editor bridge.
Maintenance update

Public website availability restored

A website hotfix restores public page loading after the prerender rollout while keeping the site security policy strict.

CLIv0.1.3VS Codev0.4.4

Fixed

  • Public pages now load normally again in browsers that enforce the production security policy.
  • The prerendered pages keep their crawlable initial content without relaxing script protections.

Compatibility

  • CLI and VS Code extension versions are unchanged for this maintenance update.
Improved

Public FPGAi pages now render content before JavaScript loads

The public website now serves crawlable page content immediately, improving access from stricter corporate networks and link preview systems that do not run JavaScript.

CLIv0.1.3VS Codev0.4.4

Access

  • Public pages now include real rendered content in the initial HTML.
  • Corporate filters and crawlers that skip JavaScript no longer see an empty app shell.

Compatibility

  • Authenticated dashboard, account, and admin pages remain client-rendered and private.
  • CLI and VS Code extension versions are unchanged for this update.
Product release

FPGAi is now available in Visual Studio Code

Use FPGAi from the native VS Code Chat view for HDL questions, code review, image-assisted analysis, and generated-file workflows.

CLIv0.1.3VS Codev0.4.1

New

  • Chat with @fpgai without leaving the editor.
  • Attach supported screenshots and diagrams when using a vision-capable model.
  • Apply generated files with guardrails, backups, and one-click undo.

Local workflow

  • The extension works without the Python CLI for hosted chat, knowledge-backed answers, and generated-file apply.
  • Install the optional CLI agent to enable local simulation, diagrams, waveforms, preflight lint, and diff workflows.
Looking for current availability? Visit the system status page.