The persistent Python dispatcher rejects scripts whose feature bundle is not
installed, but the per-request fallback (used when the dispatcher is down, e.g.
restarting right after a model repair) spawned scripts directly and bypassed
that gate. Behavior was therefore inconsistent: a gated script would fail under
the dispatcher but run under the fallback -- the "works once after a repair"
symptom from the original report.
- add packages/ai/src/feature-gate.ts: SCRIPT_BUNDLE_MAP + missingBundleForScript,
mirroring TOOL_BUNDLE_MAP in dispatcher.py, reading the same installed.json and
failing closed exactly like dispatcher._get_installed_bundles()
- runPerRequest now rejects with "feature_not_installed" (the same message the
dispatcher path surfaces) when a gated script's bundle is not installed
- unit tests for the gate, plus a drift test pinning the TS map to dispatcher.py
Closes#327
* fix(deps): patch gray-matter onto js-yaml 4.2.0 (close js-yaml DoS alert)
js-yaml 3.14.2 (quadratic-complexity DoS in merge-key handling, GHSA
patched only in 4.2.0) was kept in the tree by a scoped pnpm override
"gray-matter>js-yaml": "^3.14.1" that exempted gray-matter from the
global js-yaml>=4.2.0 override. gray-matter is a build-time-only
transitive dep of the docs site (vitepress-plugin-llms,
@sugarat/theme-shared) and pinned 3.x because it calls the removed
yaml.safeLoad / yaml.safeDump APIs.
Remove the exemption so gray-matter resolves js-yaml 4.2.0, and add a
pnpm patch renaming safeLoad->load / safeDump->dump (the 4.x
equivalents; load is safe by default). js-yaml 3.x is now gone from the
lockfile.
Verified: gray-matter parse+stringify smoke test passes on 4.2.0; full
VitePress docs build green (177 pages, llms plugin parses all tool
frontmatter with no safeLoad/safeDump error).
* docs(ai): document rembg 2.0.69 pin and advisory non-reachability
The patched rembg 2.0.75 pulls a numpy 2.x closure (numpy>=2.3,
scipy>=1.16, scikit-image>=0.26) that is incompatible with the
numpy==1.26.4-locked AI stack (realesrgan 0.3.0 and codeformer-pip 0.0.4
break on numpy 2.x). Both open rembg advisories are unreachable in this
codebase: rembg is used purely as a library (never the `rembg s`
server), and new_session() only receives allowlisted model names
(remove_bg.py ALLOWED_MODELS), never user-controlled paths. Record this
rationale next to the pin; the Dependabot alerts are dismissed as
not_used.
rembg 2.0.75 requires a numpy incompatible with the pinned numpy==1.26.4
that the rest of the ML stack (onnxruntime etc.) depends on, making
pip-audit's resolution impossible. The rembg <2.0.75 advisory (medium) is
accepted as a residual: it only affects the on-demand background-removal AI
bundle (publishing currently paused) and can't be patched without a numpy
2.x migration across the whole Python sidecar.
rembg 2.0.70+ requires numpy>=2.3.0, but the AI bundle pins numpy==1.26.4
(mediapipe, realesrgan/basicsr, codeformer, paddle all need numpy<2). The
unresolvable rembg==2.0.75 + numpy==1.26.4 combination broke pip-audit's
dependency resolution (CI red) and the background-removal bundle build. 2.0.69
is the newest rembg with an unconstrained numpy requirement. Verified: pip-audit
resolves with no unignored vulnerabilities on Python 3.11.
Bump all workspace package versions and APP_VERSION to 2.0.0, marking the
official 2.0 release. Removes the stale 1.x .release-notes.md artifact
(semantic-release regenerates release notes). The 2.0/multimodality docs
and rebrand already landed on main via #254 and #261, so this carries only
the version designation forward from the rebrand branch.
BREAKING CHANGE: SnapOtter 2.0 - the platform re-architecture (Postgres 17 +
Redis 8 + BullMQ durable jobs, 157 tools across five modalities) is the 2.0
release line, replacing the 1.x SQLite single-container architecture.
Make the full pnpm test:docker suite pass the env-dependent tests (~85 failures):
- Dockerfile.test: ENV LD_LIBRARY_PATH=/usr/local/lib so the built libheif 1.21 is not shadowed by the base image's older system libheif (heif-dec failed with an undefined-symbol error -> 'No HEIF decoder found' on 72 HEIF tests); add libjxl-tools (JXL) and ghostscript + the ImageMagick policy.xml EPS allow-edit.
- docker-compose.test.yml: SYNC_WAIT_MS=30000 so sync-wait image tools do not fall back to 202 under single-container contention (10 tests).
- install_feature.py: guard tarfile.extractall(filter='data') behind Python>=3.12 (bookworm ships 3.11); the manual entry guards already protect.
- feature-status.test.ts / docker-file-secrets.test.ts: skip the two cases that cannot hold inside the container (/.dockerenv always present; root bypasses chmod). Verified on host: all still pass.
Resolve the actionable Dependabot alerts via pnpm overrides (for transitive deps) and a Python pin bump.
- pnpm overrides: esbuild >=0.28.1 (the lone high-severity alert), @babel/core >=7.29.6, @opentelemetry/core >=2.8.0, js-yaml >=4.2.0, qs >=6.15.2, uuid >=11.1.1, yaml >=2.8.3
- rembg 2.0.62 -> 2.0.75 in requirements.txt and requirements-gpu.txt
Verified: pnpm install, typecheck, lint, and full build all pass.
NOT included: the astro advisory requires Astro 5 -> 6 (a major, breaking framework upgrade), which warrants its own migration PR rather than a security bump.
* fix(pdf): never enlarge on compress, honor redact case, hide same-format convert
- compress-pdf: guard both modes so output is never larger than the input; low-DPI scans could be upsampled and grow. Falls back to the original bytes.
- doc_redact.py: caseSensitive=true now filters PyMuPDF's case-insensitive search to exact-case hits, so the toggle works instead of always over-redacting.
- convert-{document,presentation,spreadsheet}: omit the input's own format from the output dropdown; the backend already rejects same-format conversions.
Verified end-to-end against an isolated Docker stack during a full visual QA sweep of all 37 PDF tools.
* fix(ui): show real multi-file preview thumbnails per modality
The bottom multi-file preview strip rendered a raw <img src=blobUrl> for every file, so audio/video/PDF inputs showed a broken-image icon plus the filename. ThumbnailStrip now branches on FileEntry.previewKind: images use <img> (icon fallback on error), video shows a captured first frame, PDF shows a pdf.js page-1 render, and audio/other show a type icon + extension. Fixes the multi-file preview across all modalities.
Verified in the browser for image/PDF/audio/video.
* fix(modality): make pipeline, batch validation, save/upload, previews & UI modality-aware
The app grew up image-only; several paths still assumed image. They now dispatch on the tool/file modality (image/video/audio/document/file):
- pipeline /execute + /batch: validate+decode input via inputHandlerFor(modality) instead of validateImageBuffer, so PDF/audio/video/data pipelines work (were rejected 'Invalid image').
- batch: non-image inputs now get per-modality validation (ffprobe/qpdf) before the worker instead of passing through unchecked.
- files /upload, user-files /save-result + /thumbnail: accept non-image files (MIME from extension; video-poster / pdf-first-page thumbnails).
- postprocess CONTENT_TYPE_TO_EXT: cover video/audio/pdf/text/zip so output extensions are corrected for all modalities.
- worker pipeline-finalize: attach result payload to the complete SSE event so the sync-window-timeout fallback still delivers a download.
- frontend: batch-ZIP blob MIME by extension (not svg-only); modality-neutral fallback labels/filenames; 'smaller file' not 'smaller image'.
Found via a codebase-wide image-only-assumption audit. Verified: PDF/audio/video pipelines + batch now work; image paths unchanged. canBrowserPreview kept image-only by design (non-image is rendered by dedicated displayMode viewers).
* fix(pipeline): generate a modality-aware preview for pipeline results
processPipelineFinalize now derives the output content type from its extension and runs generatePreview (video poster / pdf first page / image thumb), sets previewRef on the result, and surfaces previewUrl in the /execute sync response and the SSE complete event (via buildLegacyResultPayload). Pipeline outputs get a preview like single-tool results instead of always returning previewUrl: undefined.
Verified: PDF pipeline -> previewUrl returns a valid PNG first-page render; png pipeline correctly has no previewUrl; audio/video/multi-step pipelines all 200.
* fix(worker): auto-save a new library version when processing a library file
The worker hardcoded savedFileId = undefined ('No auto-save') even though the whole versioning feature was wired around it: the frontend sends fileId for library files and reads result.savedFileId, tool-factory threads fileId into ToolJobData, and autoSaveToLibrary implements the new-version save -- but the worker never called it (dead code from the tool-first-workflow merge). processToolJob now calls autoSaveToLibrary with data.fileId; without a fileId it is a no-op, so tool-first uploads are unchanged.
Verified: processing a library PDF with fileId creates version 2 (parent linked, toolChain appended, savedFileId returned); processing without fileId saves nothing.
* fix(library): ownership check + modality-aware dimensions in autoSaveToLibrary
- Only create a new version when the requester owns the parent (parent.userId === opts.userId); prevents versioning another user's file via a known fileId.
- Dimensions are modality-aware: sharp for images, ffprobe (probeMedia) for video, null for audio/document. Previously sharp-only, so non-image versions always got null dims.
* fix(ai): thread fileId + real userId through the 16 AI tool routes
AI custom routes parsed neither the fileId multipart field nor the authenticated user (they hardcoded userId: null), so processing a library file via an AI tool never created a new version, and AI jobs were unattributed. Each route now parses fileId like clientJobId and passes getAuthUser(request)?.id as userId to enqueueToolJob.
Verified: ocr-pdf on a library PDF creates a new version (v2); the ownership check still denies cross-user versioning.
* feat(tracing): add OpenTelemetry dependencies and --import preload flag
* feat(enterprise): add distributed_tracing feature gate
* feat(tracing): add SDK bootstrap with enterprise gating
* fix(tracing): correct test coverage for enterprise-unavailable path and prevent double-init
Test 2 now mocks @snapotter/enterprise to throw an import error, exercising
the catch block in the preload. Test 3 imports with no endpoint so the preload
is a no-op, avoiding leaked SDK from double-initialization. Added idempotency
guard to initTracing() as a safety net.
* feat(tracing): add Pino trace mixin and shared logger
When OTel tracing is active, every Pino log line now includes traceId,
spanId, and traceFlags fields for log-to-trace correlation. The mixin
is a no-op when no SDK is registered (community users).
* feat(tracing): add _otel to ToolJobData and inject trace context at enqueue
Add optional _otel carrier field to ToolJobData for W3C trace context
propagation across BullMQ job boundaries. When an active OTel span exists,
propagation.inject() writes traceparent/tracestate into the job data before
queue.add(). When no SDK is registered (community edition), the carrier
stays empty and _otel remains undefined -- zero overhead.
* feat(tracing): extract trace context and create spans in BullMQ worker
* feat(tracing): inject trace context into Python sidecar calls
* feat(tracing): add trace context extraction to Python sidecar
* feat(tracing): add shutdownTracing to graceful shutdown sequence
* feat(tracing): enrich HTTP spans with tool_id and user_id attributes
* docs: add OpenTelemetry env var documentation to .env.example
* test(tracing): add lifecycle integration tests for trace propagation
* fix(tracing): inject trace context into pipeline and batch flow jobs
* fix(tracing): add sidecar.execute Node-side span and remove unnecessary comment
Wraps PythonDispatcher.run() with a sidecar.execute span on the Node
side so traces show the full round-trip (Node span -> Python span).
Also removes an obvious comment from logger.ts.
rembg 2.0.75 pulls in numpy>=2.3.0 which conflicts with our pinned
numpy==1.26.4 and would break the entire AI dependency chain. The two
rembg CVEs (SSRF + path traversal) are in its server/CLI components
which we don't use; they're already in the pip-audit ignore list.
Proxmox LXC and other headless Linux installs lack libgl1, causing
all AI tools to show a misleading "install opencv-python-headless"
error even though the pip package is already installed. Detect the
libGL ImportError and suggest `apt-get install -y libgl1` instead.
Path resolution for the feature manifest and install script was hardcoded
to /app/..., which only works inside the Docker container. Native installs
(e.g. Proxmox at /opt/snapotter) hit "No such file or directory" errors.
Resolve both paths relative to the source file location via import.meta.url
so they work regardless of where the project is installed.
Also loosen mediapipe==0.10.21 to >=0.10.21 in requirements.txt and
requirements-gpu.txt to match the feature manifest. The exact pin has no
cp313 wheel, so it fails on Python 3.13 (Debian 13 default). mediapipe
0.10.35 ships py3-none universal wheels that resolve cleanly.
Reported-by: MickLesk (community-scripts/ProxmoxVE#14720)
- Expose birefnet-hr-matting in UI (People/Ultra) and fix model defaults
(People/Max now uses birefnet-matting for true alpha matting)
- Add output format selector (PNG/WebP/AVIF) with lossless alpha support
- Add edge smoothing post-processing (Off/Light/Medium/Strong) via
morphological mask refinement to reduce gray halo artifacts
- Add color decontamination to remove background color spill from
semi-transparent edge pixels
- Thread new settings through full stack: frontend -> API schema ->
Python sidecar -> Sharp effects pipeline
- Add i18n keys for all 21 locales
- Add unit tests for new option serialization (3 tests)
- Add integration tests for new settings validation (4 tests)
HuggingFace snapshot_download had no retry logic, causing lama-onnx and
codeformer-onnx installs to fail on transient network errors. Direct URL
downloads already had 3 retries with exponential backoff -- this adds
the same pattern to HF downloads (3 attempts, 10s/20s backoff).
Process images in 512px tiles instead of all at once, drastically
reducing peak VRAM usage. If OOM still occurs, retry with 256px tiles
after clearing the CUDA cache. Covers both upscale and face enhance.
Closes#191
Add support for models defined via downloadFn/args (rembg_session,
hf_snapshot) in bundle verification, recovery, and uninstall paths.
Previously only path-based models were tracked, so bundles using
rembg or HF snapshot downloads appeared broken after install.
Also improve pip install error messages with user-friendly hints for
common failures (basicsr build issues, OOM, disk full) and add better
error context for rembg session download failures.
Handle OOM kills (exit code 137) with actionable memory guidance,
filter ANSI/progress noise from error output, add --no-cache-dir to
pip installs, reduce download concurrency to 2, and bump default
container memory from 4g to 6g.
- Bump all workspace package versions to 1.17.0
- Update APP_VERSION constant and OpenAPI spec
- Update AI tool count from 15 to 16 across docs and i18n
- Update tool table with AI Canvas Expand, Meme Generator, Beautify
- Add image editor, OIDC, and 20 languages to README features
- Add release notes for v1.17.0
- Add JSON-LD structured data and SEO improvements to landing/docs
- Filter known client-error noise (rate limit, empty body, unsupported
media type, content-length mismatch, premature close) from Sentry
via beforeSend to stop 644 events of non-actionable noise
- Sanitize x-output-filename header to prevent TypeError on non-ASCII
filenames in optimize-for-web preview (23 events)
- Handle EPIPE on Python dispatcher stdin write with graceful fallback
to per-request spawning instead of crashing (NODE-W)
- Map EACCES on storage directory/file write to proper 503 status
instead of generic 500 (NODE-P, 3 events)
The GPU detection in gpu.py had two issues preventing GPU usage in
containers (especially rootless podman with CDI):
1. When torch was installed but torch.cuda.is_available() returned
False, the function returned immediately without trying the
ONNX Runtime + nvidia-smi fallback. This meant a CPU-only torch
build (installed before GPU was available) would block all GPU
detection, even for ONNX-based tools.
2. The failure logged a generic "torch loaded but CUDA not available"
with no diagnostic information, making it impossible to debug
whether the issue was a CPU-only build, missing libraries, or
device permissions.
The fix restructures gpu_available() into three detection tiers
(torch -> ONNX Runtime -> nvidia-smi) that always fall through on
failure. When torch CUDA fails, it now checks torch.version.cuda to
distinguish CPU-only builds from CUDA builds that can't access the
GPU, and logs LD_LIBRARY_PATH, torch.cuda.init() errors, and
nvidia-smi results.
Also fixes two env var passthrough bugs in buildMinimalEnv():
- SNAPOTTER_GPU was never passed to the Python subprocess, so the
user-facing GPU override env var had no effect
- MODELS_DIR was a dead entry (never set as env var); replaced with
MODELS_PATH which the Dockerfile sets and Python scripts read
Closes#134
Auth: login rate limit 30/min (was 500), global rate limit 1000/min (was
unlimited), password/username max lengths on all Zod schemas, session
invalidation on role change, API key legacy scan bounded to 100 keys.
SVG: hardened regex sanitizer with CDATA stripping, XML entity decoding,
set/animate/iframe/embed blocking, comprehensive data: URI blocking,
use element external href blocking. 11 attack payload fixtures added.
SSRF: fixed DNS rebinding TOCTOU by pinning resolved IPs via custom
HTTP/HTTPS agents. Added 6to4 and NAT64 to blocked IPv6 ranges.
Docker: capability dropping (cap_drop ALL + minimal cap_add), resource
limits (4g/8g mem, 512/1024 pids), healthcheck timeout, password
removed from startup banner, default password warning comments.
Network: CSP and HSTS applied in all environments (not just production),
stack traces removed from all error responses, internal paths stripped
from error details, per-route rate limits on uploads (60/min) and URL
fetches (200/hour).
Files: exclusive temp file creation (O_EXCL), disk space circuit
breaker, per-user storage quotas, settings payload 64KB size guard.
Python sidecar: script name allowlist in dispatcher, minimal environment
for subprocess spawns.
Dependencies: fixed 6 production CVEs (drizzle-orm, fastify, fast-uri,
@fastify/static, next, archiver/lodash). Pinned all GitHub Actions to
SHA hashes.
114 security tests added. Full OWASP Top 10 penetration test matrix
verified against production Docker container (30/30 pass after
hardening).
- Two-gate threshold: Otsu >= 60 uses Otsu; 40-59 uses fixed 100
(catches strong scratches on borderline images)
- Remove morphological OPEN after component filtering: it was eroding
thin scratch lines that were correctly detected
- Lower Otsu gate from 60 to 40 to avoid false-negating borderline images
Add TIER_PARAMS dict with fast/balanced/high presets controlling band
size, mask dilation, seam strip width, and Telea pre-inpainting. Parse
tier from sys.argv[7] with balanced fallback. Conditional Telea and
seam refinement steps skip cleanly for fast tier. Progressive outpaint
now accepts band_size and progress bounds for tier-appropriate scaling.
- Fix dispatcher pipe deadlock: drain stdout pipe in a background thread
to prevent blocking when ONNX runtime output exceeds 64KB pipe buffer
- Add 5-minute SSE stall timeout so the UI shows an error instead of
hanging forever when async AI processing stalls
- Guard CPU colorization: skip for images >2MP on CPU and when DDColor
model is not installed, with clear user-facing messages
- Add AVIF decode fallback via ImageMagick for bitstream variants that
Sharp's bundled libheif cannot decode (affects all tools)
AVIF (and other Sharp-native formats) were written as raw bytes to a
.png temp file, causing PIL to fail with "cannot identify image file".
Every other AI module wrapper already converts via sharp().png().toBuffer()
before writing; face-landmarks was the only one that skipped this step.
- Refactor use-tool-processor and use-pipeline-processor hooks
- Enhance dropzone component with improved UX
- Improve seam carving with better error handling and tests
- Add JXL format encoding support to format-encoders
- Update tool routes for consistent format handling
- Add dropzone unit tests
