On a 28-task document-tool compatibility suite, Wolf passed all 28 tasks, as did the LibreOffice control, including six dynamic-array recalculation tasks. We fixed Wolf against these same tasks over several iterations, so read this as a compatibility result.
Sandboxes / Platforms / Embedded products
Embed the document engine. Keep the receipt.
Wolf Runtime by WolfSDK works with spreadsheets, documents, decks and PDFs inside your environment. On our test files it left untouched package parts unchanged on supported edits, and it gives you a machine-readable receipt and a list of named rendering omissions.
- PreservationPackage preservation
- ReceiptsReceipts and omissions
- IntegrationDocumented shims
- ProcessingLocal processing
Preserve and verify
Inspect the change. Keep the surrounding file.
On supported edits, Wolf lists the parts it changed and counts the parts it preserved, so you can check them against the original file. For renders, Wolf reports each omission, approximation and font substitution before you accept the result.
On our test files, Wolf left all 55 DOCX and all 38 PPTX files byte-identical on a no-op save, and changed no unrelated part in the measured DOCX tracked-change and PPTX text edits. LibreOffice 26.2.5.2 rewrote unrelated parts in every test file.
| Format | Test files | Wolf, no-op save | Wolf, edit | LibreOffice, no-op save |
|---|---|---|---|---|
| XLSX | 45 | No unrelated part changed | Recalculation save: no unrelated part changed | 1,195 part changes |
| DOCX | 55 | 55 of 55 byte-identical | Tracked changes accepted in 52 files: no unrelated part changed | 1,400 part changes |
| PPTX | 38 | 38 of 38 byte-identical | Text edit in the 37 files that contain text: no unrelated part changed | 3,311 part changes |
LibreOffice rewrote unrelated parts in every test file. An office suite that re-serializes on save is expected to, and we use it as a control. The test injected the DOCX tracked changes, and XLSX recalculation rewrites formula sheets and the calculation chain, which we count as intended.
Rendering measured against Office
Compare with native Office output.
On a frozen held-out corpus, Wolf renders closer to Office's own PDFs than LibreOffice does in DOCX, PPTX and XLSX. The PPTX gap is small, and Wolf matches Office's page count on 21 of 39 XLSX files. We fixed earlier failures between scoring passes on this corpus, so treat it as a regression benchmark. We have not run a blind test.
| Format | Documents | Wolf mean distance | LibreOffice mean distance | Wolf page counts match |
|---|---|---|---|---|
| DOCX | 42 | 0.136 | 0.203 | 37/42 |
| PPTX | 51 | 0.107 | 0.120 | 51/51 |
| XLSX | 39 | 0.202 | 0.332 | 21/39 |
Mean SSIM distance to Office's own PDF output; lower is closer. The last column counts documents where Wolf's page count matches Office's. We fixed earlier failures between scoring passes on this corpus, so treat it as a regression benchmark. We have not run a blind test.
Shims first on PATH
Keep supported command lines. Replace the engine behind them.
Put Wolf's documented soffice, pdftoppm, pdftotext, pdfinfo and pdfimages shims first on PATH. If a caller hard-codes an executable directory, place the shims at the documented override location. Check the accepted arguments and output conventions before you integrate.
Select the build.
Check the platform, font policy and supported operations.
Install the shims.
Place the documented shims before upstream tools on PATH, or at the caller's explicit override location.
Run your workflow.
Run your own templates and check the receipts, omissions and refusals.
For explicit integration, call the documented native command-line tool.
Measured image addition
48.4 MiB added, against 357.9 MiB.
The static Linux x86_64 binary is about 50 MB. Optional recognition packages and Python orchestration are extra, and we did not boot a vendor sandbox for this measurement. Wolf reported omissions on some of the test fixtures.
Added to a public code-interpreter sandbox image in a local build, Wolf adds 48.4 MiB (30.7 MiB compressed). LibreOffice plus Poppler adds 357.9 MiB (135.7 MiB compressed). The whole image is 9.0% smaller with Wolf, 3,142.8 MiB against 3,452.4 MiB uncompressed, because the base image dominates.
| Install | Added, uncompressed | Added, compressed | Whole image, uncompressed |
|---|---|---|---|
| Wolf | 48.4 MiB | 30.7 MiB | 3,142.8 MiB |
| LibreOffice plus Poppler | 357.9 MiB | 135.7 MiB | 3,452.4 MiB |
Sandbox posture
Process locally, inside your controls.
The embedded Wolf binary has no network client, runs no subprocess, interprets no macros, resolves no external references and collects no telemetry. Apply operating-system resource limits to untrusted documents.
In a recorded Linux check, Wolf produced the same task results with an empty environment, no HOME, networking disabled, a read-only root filesystem and fonts supplied from a declared directory. One suite passed 87 of 87 in both runs. Another passed 19 of 22 in both, with the same three failures.
The native command-line tool uses bundled fonts and declared directories. The compatibility shims default to system fonts, so pin fonts when you need repeatable output.
Rights by distribution band
Match the license to the deployment.
You sign one written Order for the linked office-document engines and rendering. Fees are flat for each annual term, with no per-seat, per-instance or per-document charge.
Internal
Run on infrastructure you control and provide outputs to end users.
From $15,000 a year
One flat annual fee.
Product
Embed and distribute with the named licensed product.
From $60,000 a year
Includes one licensed product. Each added product is $15,000 a year.
Platform
Preinstall in sandbox or compute-platform images for third-party developers.
From $150,000 a year
One flat annual fee.
For a standalone document-engine service, repackaged library or SDK resale, ask about separate terms. Wolf Runtime has no online checkout.
A free evaluation of Wolf Runtime is available on request. It runs for 60 days with one 30-day extension, for non-production use.
Platforms
Documented release targets on Linux, macOS and Windows.
The documented release targets are Linux x86_64 and arm64, macOS x86_64 and arm64, and Windows x86_64. Our measured output comparison covers Linux x86_64, Linux arm64 and macOS arm64. Confirm the platform, package and operation for your deployment.
Measured on three platforms.
On an eight-input suite with fixed inputs, bundled fonts, clock, locale and seed, Wolf builds for those three platforms produced identical PDFs, PNGs, extracted text and render reports. A larger run found small JPEG decoding differences between macOS arm64 and Linux x86_64 in PDFs that contain JPEG images.
We plan .NET hosts first and Java hosts next over the C ABI. Package names, prices and dates are not announced.
Tell us about your integration.
Tell us where Wolf Runtime will run, which formats you need and how you ship, and we will send a quote.
Notes
- Preservation: same-format operations on finite sets of test files, run on Linux x86_64 with wolf 0.1.0. The DOCX edit is tracked-change acceptance on revisions the test injected. One image-only PPTX deck has no text and is excluded from the text-edit count. PDF edits are not measured. LibreOffice 26.2.5.2 is the control.
- Receipts and omissions: the command and receipt contract is documented, and operations differ by format. Receipts and omission reports are operation-specific and do not prove every semantic outcome.
- Rendering: a frozen held-out corpus of 42 DOCX, 51 PPTX and 39 XLSX files, scored as mean SSIM distance to Office's own PDF output against a stored LibreOffice 26.8.0.3 baseline. Wolf reported omissions on 12, 18 and 8 files respectively. In DOCX, nine earlier Wolf render failures were fixed before this scoring pass. No PDF-input ranking exists.
- Compatibility: a pinned static Linux build. Wolf's pass count went from 19 of 22 to 20 of 22 to 28 of 28 across iterations against these same tasks, so the result is not from a held-out test. The six dynamic-array tasks use references captured from Excel. This was not a live autonomous run, and the suite's publisher has not endorsed the result.
- Footprint: a local build from a public code-interpreter sandbox template, against LibreOffice 25.2.3 headless from Debian with 89 added packages. The added-layer figures are uncompressed unless marked compressed. No vendor sandbox was booted and no cold-start claim is made.
- Sandbox posture: a design statement for the embedded binary and no hostile-file or certification claim. Optional recognition runs a separate executable, and sessions and the control plane have their own process lifecycles. The isolation check ran on Docker linux/amd64 with static Wolf 0.1.0 and matched the unsandboxed run; the 19 of 22 suite has three failures in both.
- Platforms: the release targets are a documented list and carry no per-feature qualification. The eight-input comparison has no Windows run. A larger run of 577 cases found 8 PDF differences in files with JPEG images between macOS arm64 and Linux x86_64 (maximum channel delta 3), so no byte-identity claim is made across platforms.
- Licensing: Wolf Runtime is licensed under a written Order with SynthGL, Inc. Starting prices are list prices for new Orders and the final terms are in the Order.