Where AI becomes your translation infrastructure — at up to 75% lower cost.
Drop the file, get it back translated and reviewed — glossary, translation memory, and 12 QA steps run on every job.
No card. No demo form. No editor.·Backed by a real Account Manager
Real production data from the Fily engine. Updated every 30 seconds.
What you don't do with Fily
The whole category sells you a workflow to use. We take it out of your way.
- No browser editor to open.
- No segment-by-segment approval.
- No pipeline to configure.
- No glossary to copy by hand.
- No broken tags to fix.
- No mediocre MT to clean up.
- No PO to send an agency.
- No 5-day wait.
Upload. Download. Done.
Invisible doesn't mean opaque
What actually happens inside.
Three phases. Twelve quality steps. Every one logged, auditable, attached to your delivery as an HTML report.
Extract
Read the file. Decide what to translate. Pass through what shouldn't change.
01Source parsing
Parse XML / DOCX / MQXLIFF / PDF. Detect format, tags, locked TUs.
02TM match pass-through
100% and 101% matches bypass the AI byte-for-byte.
03Segment translation
Frontier LLM with prompt caching, per-segment context.
Enforce
Apply rules. Repair tags. Replace forbidden terms. Catch what AI gets wrong.
04Glossary enforcement
Mutates target — preferred terms inserted, not just flagged.
05Style guide application
Voice, register, punctuation rules per client / project.
06No-source-bleed check
Detects untranslated chunks that slipped through verbatim.
07Forbidden term replacement
Brand-forbidden words rewritten to approved equivalents.
08Tag count repair
Inline <g>, <x>, {N> realigned. Trados/memoQ accepts on import.
09Consistency check
Same source span → same target across the document.
Deliver
Normalize. Score. Reassemble. Ship the file plus the report.
10Number & date normalization
Decimal commas vs dots, date formats, currencies per locale.
11Quality scoring
0–100 score per segment + per pipeline step. Visible in QA report.
12Bilingual reassembly
Re-pack the file in its original format. Tags intact. Ready to deliver.
Each step has a name, an owner, and a measurable check. The HTML report shows what fired, what passed, and what flagged.
One upload. Any format.
From sdlxliff to scanned PDF.
Tags intact. TM 100/101 untouched. Formatting preserved. The file you upload is the file you get back.
Three branches · One pipeline
Translation. OCR. Transcription. Each one engineered, not generalized.
Most AI tools do one of these. We built dedicated pipelines for all three — because text, scans, and audio fail in completely different ways.
Branch 1
Translation
Native pipelines per format. Tags preserved. TM 100/101 untouched. Glossary that mutates output, not just flags violations.
What makes it different
- Dedicated pipeline per CAT format (SDLXLIFF, MQXLIFF, TXLF, RTF, XLIFF 1.2/2.0)
- Tag count repair + xid normalization — Trados accepts every file we return
- Hard TM 100/101 pass-through (never sent to AI)
- 4-layer glossary cascade: system → org → client → sub-client → job
- DOCX tracked changes round-trip across languages (industry first)
- 12-step QA report on every delivery
Live numbers
46M+ words · 142 pairs · 97.5% QA
Branch 2
OCR
Read what nobody else can read. Phone photos, faxed medical records, blueprints, historical archives — all parsed cleanly.
What makes it different
- Dual-backend vision-language OCR with automatic failover on errors
- Page-level retry — one bad page doesn't kill the document
- Block-level extraction with layout reconstruction (columns, tables, footnotes)
- Handwriting detection + confidence flagging per region
- Works on PDFs (scanned + native), single images, faxes down to ~150 DPI
- Output as searchable PDF or DOCX with reconstructed layout
Live numbers
200+ scanned PDFs · 7.8M words extracted · ~31K pages
Branch 3
Transcription
Audio in, bilingual document out. Diarized. Timestamped. Ready to file with the reviewer or the court.
What makes it different
- State-of-the-art ASR + speaker diarization (multi-speaker)
- Verbatim mode for legal — preserves “um”, false starts, pauses
- Per-speaker translation with surrounding context (not segment-by-segment)
- Bring your own template — transcripts in your client's branded DOCX (logo, fields, footer)
- Domain prompting (medical, legal, technical glossaries during ASR)
- Word-level timestamps for subtitle workflows (SRT, CPS-aware)
Live numbers
1236+ hours of audio · 4,098+ jobs processed
All three branches share the same 12-step QA pipeline. See all 15 formats →
What lands in your inbox
A file.
And a report you can hand to your client.
- Auditable segment by segment
- Glossary hits and misses, highlighted
- Forbidden terms flagged with replacements
- TM 100/101 segments marked as pass-through
- Tag count diff per segment
- Quality score per pipeline step
Average QA pass rate across 14,507 jobs: 97.5%
Every job makes the next one cheaper
Your translation memory, built automatically. Reused for free.
Fily builds a Golden TM for your organization with every job you run — confirmed translations, per client, per language pair. No setup, no exports, no TM babysitting.
Exact matches cost $0
100% and 101% TM matches, in-context exacts and locked segments are never billed — rate zero, not a discount. They bypass the AI byte-for-byte and show up as $0 lines in your word-count breakdown.
On repetitive content — forms, contracts, product updates — that routinely cuts the billable count by half or more.
Golden TM, zero maintenance
Confirmed segments are promoted into your organization's Golden TM automatically — scoped per client and sub-client, applied on every future job. Edits your reviewers make in the dashboard feed straight back in.
Bring your own TMX too: upload it once and it pre-fills matches alongside the Golden TM. Inspect, export or prune it any time from your dashboard.
No TM yet? We'll build it from your archive
The TMX Aligner takes a source document and its existing translation — two plain files, no bilingual format required — and returns a ready-to-use TMX plus an extracted glossary.
Years of past translations become leverage on day one, instead of sitting in a folder.
Invisible workflow, full control
Review in the browser. No CAT tool required.
Every job opens in a segment-level review editor. See exactly what changed, fix what you disagree with, and regenerate the final file — your corrections feed the Golden TM so they never come back.
- Real change tracking — a word-level diff against the file you uploaded — changed words in red, with a true % per segment.
- Cross-file consistency — one source translated two ways across a 25-file batch? One click unifies every occurrence.
- AI on demand — per-segment autofix, spellcheck, and “Ask AI to review” with back-translation for languages you don’t read.
- Hold for review — gate delivery until a reviewer approves; the final file regenerates from the approved edits.
The diff nobody else offers
AI speed. Human accountability.
Every paid account has a dedicated Account Manager. Not a chatbot. Not a ticket queue. A human who knows your formats, your glossary, your deadlines — and has a direct line to engineering when something needs to change.
The speed of a tool
AI pipelines run in minutes. Doc3, Smartcat, DeepL get this part. Files in, files out — no human in the loop.
The accountability of a vendor
Agencies have AMs but the AM isthe bottleneck — every PO, every change goes through them. Fily AMs unblock, they don't gatekeep.
Direct line to engineering
When standard isn't enough, your AM brings engineering in. We've built custom pipelines for specific clients in days, not quarters. Not a sales conversation — actual code shipped.
Available on every paying tier. Not an enterprise-only upsell.
When standard isn't enough
Custom pipelines, built by engineers. Not configured by a sales rep.
The 15 format pipelines are the baseline. When a client's workflow doesn't fit, we build the pipeline. Here are three we actually shipped.
docx_tracked_delta_v2
Tracked changes that survive translation
An LSP came to us with this: "Client edits the EN doc with Word tracked changes, we already translated v1 to Spanish. They want v2 in Spanish with the same tracked changes applied — not a fresh re-translation." No tool on the market did this.
We built a delta-aware DOCX pipeline that ingests EN + redlines + ES_v1 and emits ES_v2 with the equivalent insertions and deletions in Spanish. Reviewer keeps the audit trail. Hours instead of days per round.
Industry first — read the DOCX page
pdf_qa_12step_v2 + image_qa
Dual-backend OCR for impossible scans
Healthcare LSPs were hitting walls with 150-DPI faxed medical records that generic OCR failed on silently — returning blank pages without warning.
We built a fallback chain: a frontier vision-language OCR as primary, automatic failover to an alternate backend on truncation or JSON errors. Page-level retry, confidence flagging, handwriting detection in the QA report.
See the scanned-PDF pipeline
4-layer cascade
Multi-tenant glossary cascade
An LSP with 40+ end-clients needed a glossary system where org-level terms, client-level overrides, and sub-client (brand) overrides all stack — without re-attaching glossaries to every job.
We built a 4-layer cascade: system → org → client → sub-client → job. Forbidden terms inherit and override. Style guides cascade. PMs configure once, every future job picks up the right rules automatically.
See the LSP use case
Bring your own template
Transcripts in your client's branded document
Sonix, Rev, Otter hand you their default DOCX. Then your PM spends an hour re-formatting into the client's template: logo, case number, signature block, footer.
Upload your client template once. Every future transcript for that client lands inside it — speaker labels, timestamps, translations placed exactly where the template expects them. Zero re-branding.
See the audio pipeline
Light post-editing
Already translated? We only fix the errors
Send a bilingual file that's already translated — by a human, another vendor, or an MT engine you don't trust.
Fily corrects only errors— accuracy, terminology, grammar, numbers — and applies your glossary. It never retranslates or over-edits style, so your linguist's voice survives. Billed at a reduced rate.
Semantic audit · +5%
Back-translation you can hand to compliance
For healthcare, legal and regulated content: every segment is translated back into the source language by an independent pass and compared for semantic drift.
You get an auditable report per segment— plus per-segment back-translation inside the review editor, so a reviewer who doesn't read the target language can still sign off.
Human QA
A professional linguist, one toggle away
Flip “Human QA” on any job and a professional reviewer works through the AI output segment by segment before delivery — with the QA report and back-translation as their starting point.
Quoted upfront in credits, tracked in your dashboard. AI speed for the 95%, human judgment where it matters.
cross-file consistency
One term, one translation, across 50 files
In a big batch, the same source sentence used to come back translated two or three different ways across files — and reviewers caught it by hand.
A tag-safe consistency pass runs across the whole batch, flags every span with divergent targets, and unifies them in one click from the review editor — inline tags untouched.
Got an edge case Doc3, Smartcat, or your current vendor can't handle?
Talk to our Account Manager teamBuilt for the way you actually work
Whether you're a translator, a PM, or a buyer.
Translate. Don't post-edit.
Drop your sdlxliff, mqxliff, or txlf. Get it back with your glossary applied, tags intact, and TM matches untouched. You review the QA report, not 4,000 segments.
AI speed. Your judgment on what to keep.
Your PMs stop being dispatchers.
Bulk uploads, client-specific glossaries cascading from org → client → subclient, batch ZIP delivery, per-role access control for your team, and a QA report your reviewer can stand behind.
Dedicated Account Manager. Real human. Not a chatbot.
No agency. No emails. No 5-day wait.
Upload directly. Receive the translated file plus an auditable report. HIPAA-ready. On-prem available. SLA on Enterprise.
Account Manager for every paid tier. Backed by humans.
Trusted by translators and the teams that hire them
“I stopped opening Trados for first-pass translation. Fily returns the sdlxliff with my glossary already enforced. I review, I deliver.”
— Translator, healthcare specialist
“We process 200K words/week through Fily before they hit our reviewers. Our PM team grew zero last year while volume doubled.”
— PM, mid-size LSP
“For our 18-language community communications, Fily is the only tool that handles every format we need in one upload.”
— Director, K-12 multilingual programs
14,507 jobs processed · 142 language pairs · 1236+ hours of audio
Volume-tiered · Same quality for everyone · PAYG or subscription
Best quality always. Pay less as you grow.
Every tier gets the full stack — translation, OCR, transcription, 12-step QA, glossary cascade. The only thing that changes is volume.
USD base
Solo
Individual translators · small projects
$90 / mo
10,000 credits / month
1 credit = 1 word · one pool, mix products freely
- All features unlocked
- 12-step QA report
- Glossary cascade
- API access
Pro
Freelancers, PMs, growing teams
$400 / mo
50,000 credits / month
1 credit = 1 word · one pool, mix products freely
- Everything in Solo
- Batch ZIP delivery
- Org/client glossary cascade
- Email support
Studio
LSPs, agencies, in-house loc teams
$1,750 / mo
250,000 credits / month
1 credit = 1 word · one pool, mix products freely
- Everything in Pro
- Dedicated Account Manager
- Custom templates
- Priority queue
Scale
High-volume LSPs · multilingual content ops
$6,000 / mo
1,000,000 credits / month
1 credit = 1 word · one pool, mix products freely
- Everything in Studio
- Net-30 terms available
- Concurrent jobs boost
- Dedicated reviewer pool
Enterprise
Custom volume · compliance · on-prem
Custom
2.5M–10M+ credits / month
1 credit = 1 word · one pool, mix products freely
- Custom volume + SLA 99.9%
- HIPAA / SOC 2 / on-prem
- Dedicated workers + KMS
- Audit logs + data residency
No subscription? Buy a credit pack
Pay as you go — bigger packs, cheaper credits
One-time purchase, no card on file, no monthly commitment. Credits never expire while your account is active, and every product draws from the same pool.
| Pack | Credits | Price | Per credit |
|---|---|---|---|
| starter | 2,000 | $20 | $0.0100 |
| small | 5,000 | $50 | $0.0100 |
| medium | 11,000 | $100 | $0.0091 |
| large | 28,000 | $250 | $0.0089 |
| xl | 58,000 | $500 | $0.0086 |
| xxl | 120,000 | $1,000 | $0.0083 |
| mega | 310,000 | $2,500 | $0.0081 |
| ultra | 650,000 | $5,000 | $0.0077 |
What credits buy (pay-as-you-go rate — subscribers spend at their tier's lower rate):
Optional review layer · Stack on any product
Add a review pass when it matters
Semantic Audit
+$0.01 / word · +$0.10 / page · +$5 / hrAI back-translation check on every segment. Detects omissions and meaning drift.
Language Target
+$0.03 / word · +$0.30 / page · +$30 / hrHuman native reviewer. Bilingual file with tracked changes. 24–72h turnaround.
Enterprise add-ons · Infrastructure & compliance
Pick what your team needs
Dedicated workers
$250 / moExclusive worker pool + private queue + latency SLA.
99.9% uptime SLA
$500 / moContractual SLA with automatic credits for downtime.
Data residency EU/US
$500 / moClient data kept in specific region (GDPR / sovereignty).
Audit logs (SOC2/HIPAA)
$1,000 / moCompliance reporting for regulated industries.
Custom KMS
$750 / moBring-your-own encryption keys for storage at rest.
White-label UI
$2,000 / moclient.fily.pro subdomain + logo + reports without Fily branding.
On-prem LoRA hosting
$3,000 / moLoRA fine-tuned on client corpus hosted in dedicated infra.
Dedicated AM
$2,500 / moAssigned Account Manager + weekly call + quarterly QBR.
All enterprise add-ons available from the Scale tier and up. Custom enterprise contracts negotiated individually with regional discount applied.
Free tier available — 15K words / 10 OCR pages / 30 min audio per month with watermark.
Annual commitment: −15% on any subscription. Net-30 from Scale tier (with credit check).
Frequently asked
Upload. Download. Done.
Try Fily with one file. No card. No form. No editor.
15 formats · 142 language pairs · 97.5% QA pass · Auditable on every job