Comparison
Fily vs wxrks (Bureau Works)
Both translate documents with AI. They take different paths. Here's the honest comparison so you can pick the right one.
wxrks (Bureau Works) wxrks — Bureau Works until March 2026 — is an AI-first translation management system born inside an LSP. It manages the whole operation: workflow steps, vendor assignment, price lists, payables, 15+ CMS/repo connectors, and AI layers (Sous Chef agents, Smells verification, Model Routing) on top.
Fily is not a TMS. It's the translation engine itself: you upload the file, a 12-step pipeline translates it with glossary enforcement, tag repair, and TM respect, and you download the finished file with an audit report. The workflow disappears instead of being managed.
If you run a network of freelance linguists and need software to orchestrate and pay them, wxrks is built for that. If you want the translation done — without operating the workflow — Fily is built for that.
Side-by-side
| Capability | Fily | wxrks (Bureau Works) |
|---|---|---|
| Workflow model | File in → file out, no workflow to run | Full TMS — steps, vendors, tasks, finance |
| What the price includes | The translation, done (all-in per word) | Software license — linguists billed separately |
| TM 100/101 matches | $0, byte-for-byte pass-through — optional re-check pass that catches inherited TM errors | 0–1% net rate — the TM is trusted as-is |
| Audio / video translation | 1,200+ hours processed (ASR + diarization + translation) | .mp3 / .mp4 / .wav listed as unsupported |
| Scanned PDF (OCR) | Dual-backend OCR with auto-fallback, page-level retry | PDF→DOCX conversion with page caps |
| Glossary in the output | Enforced — 3-layer autofix mutates the target | Detected — Smells flags issues for a human to fix |
| Tag handling (CAT files) | Dedicated codecs per dialect + 4-layer repair with rollback | Fix Tags AI action + manual drag & drop |
| Client-side reviewer role | Yes — your client's editors see only what's assigned | No client-side reviewer role |
| Vendor management & payables | Not the product — Fily is the vendor | Core strength: price lists, POs, payables |
| CMS / repo connectors | 15+ (AEM, Contentful, GitHub, Figma…) | |
| SSO / SOC 2 | On the roadmap | OIDC SSO, SOC 2 Type II |
| Human QA as a deliverable | In-house linguist pass with mandatory back-translation | Marketplace of vendors you manage |
| QA report deliverable | HTML audit report on every job | In-app dashboards |
| API + webhooks | Private beta | v3 API |
| Seats | Not seat-priced | Unlimited seats |
| Pricing model | Per delivered word, all-in | Words/year subscription + AI tokens + API calls |
When wxrks (Bureau Works) is the right choice
- You run your own network of freelance linguists and need software to assign, track, and pay them.
- You need continuous localization through CMS/repo connectors (Contentful, AEM, GitHub…).
- Your security team requires SSO and SOC 2 today, not on a roadmap.
- You want unlimited seats on a software license and you supply the labor.
When Fily is the right choice
- You want translated files delivered, not a translation operation to manage.
- Your content includes audio, video, scans, or faxes — not just text formats.
- You need the glossary enforced in the output, not flagged for someone to fix later.
- You don't fully trust the TM you inherited — Fily can re-check 100% matches instead of billing them at 0% and passing errors through.
- Your client's own editors need to review jobs without seeing the rest of your operation.
- You'd rather compare all-in cost per delivered word than license + labor + tokens.
The philosophical difference
wxrks manages the work: people, steps, vendors, money. It's a system of record for a translation operation — the AI is a layer on top of the workflow.
Fily is the work: the pipeline that produces the translated file. There is no workflow to operate, because the workflow is the product's job, not yours.
Not the same product competing on features — two products competing for the same buyer. Decide whether you want to run the operation or receive the output.
See for yourself.
Upload the same file to both. Compare what comes back.