LinguaEmpire — Platform

Bridging the Barriers

The LinguaEmpire translation platform — live on the real backend. AI agents translate, the Chief AI decides, and a human linguist steps in only when the Chief AI routes the job to the Human Gate.

Sign in

Sign in with your LinguaEmpire account. Your password is sent only to the backend on this server and is never stored on this page. The sign-in token is kept only while this tab is open.

No account yet? Create one below. Not signed in? Visit the public LinguaEmpire website — no account needed.

Create account — start a free trial

Create an account, then start a free trial of 25,000 LinguaCredits. Free credits only — no payment is taken and no card is asked for.

Your account type decides which dashboard you land on after sign-in — Freelancer, LSP, Agency or Company. No account type, no account.

Dashboard

Live from the backend right now. "—" means that item has not loaded yet (or needs sign-in / admin, where noted).

Checking…
Backend health
—
public endpoint
Engines ready
—
public endpoint
Languages
—
sign-in required
Priority pairs
—
sign-in required
Queue
—
admin only

API address used by this page:

My account

Your account, subscription and credit balance — all read live from the backend. Sign in to load them.

Sign in to see your account.

My projects

Your real projects from the backend (GET /api/v1/projects), honouring the active workspace when one is set.

TitlePairStatusPipelineCreated
No jobs yet in this session. Create one from the "New job" tab.

New job

This creates a real project on the live backend, uploads your text, and runs the full six-stage AI pipeline. Keep it to 1–3 short paragraphs (separate paragraphs with a blank line). A job can take around 3 minutes — the AI thinks carefully before it translates, so please wait while the page checks progress automatically.

Two paragraphs above = two translation units. One blank line between paragraphs makes a new unit. Or upload a file below instead — the backend extracts and segments it.

Sign in to see the supported file formats.

Your Client ID is filled in automatically when you sign in. It is kept only while this tab is open. Which accounts get a Client ID? Only Company, Agency and LSP accounts get a Client at signup — Freelancer and admin/translator accounts have no client, so New Job cannot run for them. If you signed up as a freelancer or an admin, create a Company, Agency or LSP account to place jobs.

Agent Verified: every new segment is checked and edited by an AI Review agent against your terminology and QA rules. Exact 100%/101% matches are skipped. Segments the AI Review agent checks are labelled “AI Reviewed” and are kept out of model training unless a human expert confirms them. Target delivery: under 3 hours.

Job detail

Full run view for one project: project record, pipeline stages, per-unit LQA, handoffs, segments and the audit trail — all read live from the backend.

Agent Verified:
Export this job:

Human Gate

For a signed-in decider (admin or the assigned reviewer). Only jobs the Chief AI routed to the Human Gate appear here as waiting — everything else is approved directly and never reaches this page.

Waiting gates: "Load waiting gates" lists every job waiting at the Human Gate from the backend (admin). "Scan session jobs" checks only the jobs created in this browser session. Decisions are made only through the real endpoint POST /api/v1/gate/{gate_id}/decide — nothing here is faked.

Admin

All 2,852 priority pairs, loaded live from the backend's pair manifest. Search by language, pair code or linguist name. Every pair works in both directions.

Sign in, then press "Load pairs".

Admin — engines

Live from GET /api/v1/engines (sign-in required).

Sign in, then press "Load engines".

Admin — departments & coworkers

Live from GET /api/v1/departments (sign-in required).

Sign in, then press "Load departments".

Admin — queue

Live from GET /api/v1/queue/metrics and GET /api/v1/queue/failed. Admin only — a non-admin sign-in gets "admin only".

Sign in as an admin, then press "Load queue".

Marketplace

The 2,852 priority pairs from the live pair manifest, with each pair's named Human Gate linguist. Pick a pair to start a job with it. No profiles, photos or rates are shown — the backend does not provide them, so nothing is invented here.

Sign in, then press "Load pairs".

Vendor search

Search the roster-backed linguists for the 2,852 approved pairs — by pair, specialisation, competency, dialect, service and rate. Every result is a named Human Gate linguist from the approved roster; nothing here is invented. Rating comes only from reviews on completed jobs — a vendor with no completed-job reviews shows no rating.

Local time and portfolio pieces cannot be filtered yet: no timezone is recorded for any roster linguist, and no portfolio store exists. Those filters will narrow results only once that data is recorded — until then they are not offered as filters here.

Sign in, then press "Search vendors".

Invitation rate

Set the rate for one invitation, per service and language pair. The rate in effect at invitation time applies to that job; later changes affect only new invitations. Once the supplier accepts, the rate locks and cannot be changed.

Marketplace — verified linguists

Linguists verified in the backend roster (GET /api/v1/roster). Rate shows "—" where the backend has no rate for that linguist.

Sign in, then press "Load roster".
NamePairDialectDomainRateVerified at
Roster not loaded yet.

My Team

Your saved freelancers and agencies. Private notes, client information and your account-private rates are visible only in your own account — never on a vendor's public profile and never to another account. A rate you set here applies only to invitations sent after the change; once a supplier accepts a job, that job's rate locks.

Sign in, then press "Load my lists".

Save an agency by account

Agencies save by their linked account, not by a roster name — enter the agency's user ID. A 422 means that account is not an agency.

Inbox

Your invitations, job updates and messages in one place. This inbox shows only your own items — the backend filters everything to your account, so a client never sees another client's items and a vendor never sees another vendor's.

Accept/decline emails are on by default. Turn this off to stop the emails — the inbox still lists every accept and decline.

Sign in, then press "Refresh".

Chat

Conversations with people you are already connected with. A chat opens only after an invitation is accepted or a job is assigned — there is no way to message someone you are not connected with. When a task ends, the connection — and the chat — stays open. Only the two connected parties can read it.

Sign in, then press "Refresh conversations".

Conversations

No conversations loaded yet.

Pick a conversation
Your messages appear here.

No conversation open.

Invitations

Invite a linguist to a task, or answer an invitation sent to you. Every invitation carries the full offer — pair, tier, scope, deadline and credits — so the decision can be made here, without a message first. Questions before accepting go through Decline with a reason; the chat opens the moment an invitation is accepted.

Send an invitation

Received invitations

Invitations sent to you. Open one to see the full offer, preview the files, and accept or decline.

Sign in, then press "Load received invitations".

Sent invitations

Invitations you sent, with their live state. A declined or expired invitation can be re-sent as a new invitation — the earlier one stays on record.

Sign in, then press "Load sent invitations".

Open tasks

Tasks posted open: any qualified vendor may claim one, and the first to claim it gets it. Claim rules and expiry are shown on every task. If someone else claims a task first, you are told why — and shown the next matching task.

Post a task open

Open now — claim a task

Only tasks you are eligible for are listed here (pair, tier and opt-in, checked by the backend).

Sign in, then press "Load open tasks".

Tasks I posted

Sign in, then press "Load my posted tasks".

Auto-match

The system proposes the best-matched vendors for a pair — by tier, availability, past quality and response rate. It never assigns anyone by itself: you confirm each vendor you want to invite. A score component with no data behind it shows "—", never a made-up number.

Invitation

You opened an invitation link. The result of the link is shown here — it either accepts the invitation, or tells you exactly why it cannot.

Checking the invitation link…

Freelancer dashboard

Your invitations, your tasks, your earnings — only your own data. The backend enforces what each role can read.

My invitations

Invitations sent to you, with their live state.

—

Tasks I can claim

Open tasks you are eligible for, from the open-tasks board.

—

Earnings

What you have earned on completed work.

—

LSP dashboard

Your team, your jobs, your vendors — only your own data. The backend enforces what each role can read.

Team

People invited to your active workspace.

—

Jobs

Your projects, with their live status.

—

Vendors

Vendors you have invited, with each invitation's state.

—

Developer — MCP Server

Use LinguaEmpire inside the AI assistant you already work in — Claude, ChatGPT or any MCP-capable assistant — with the same permissions as your signed-in account. Every call is written to the audit trail with the assistant identified. Only the 2,852 roster-approved pairs run; any other pair is refused.

Endpoint
—
Protocol
JSON-RPC 2.0 over HTTP — methods: initialize, tools/list, tools/call
Auth
Bearer token — the same JWT or API key you use for this platform. An MCP caller can only do what that account can do.

Connect — Claude Desktop

Add a custom connector in Claude Desktop, paste the endpoint below as the MCP server URL, and sign in with your LinguaEmpire account. Start a new chat after adding it, and name LinguaEmpire in your first message so the assistant routes to this server.

Connect — ChatGPT

In ChatGPT settings, add an MCP server with the endpoint below and your LinguaEmpire account. Mention LinguaEmpire in your first message so the assistant routes to this server.

—

Tools

Sign in, then press "Load tools".

Recent MCP calls

Your own calls from the audit trail, newest first — tool, assistant, status and time. Unreachable data shows "—".

Sign in, then press "Load calls".
ToolAssistantStatusTime
Calls not loaded yet.

Agency dashboard

Your team, your jobs, your vendors — only your own data. The backend enforces what each role can read.

Team

People invited to your active workspace.

—

Jobs

Your projects, with their live status.

—

Vendors

Vendors you have invited, with each invitation's state.

—

Company dashboard

Your projects, your vendors, your invoices — only your own data. The backend enforces what each role can read.

Projects

Your projects, with their live status.

—

Vendors

Vendors you have invited, with each invitation's state.

—

Invoices

Your invoices, with their live state.

—

Reporting

Live from GET /api/v1/reports/summary and GET /api/v1/reports/roi. "—" means the backend returned no value for that item.

Sign in, then press "Load reports".

Leave both dates blank for all time. The dates filter the summary, the ROI figures and the CSV download below.

Supplier Payments

Supplier jobs, invoices, balance and bills to clients — all from live API data. Cost is calculated from the agreed rate and the TM net-rate scheme (New 100%, 95–99% 80%, 85–94% 60%, 75–84% 50%, 50–74% 30%, exact/repetition 20% for translation and 100% for review).

Payment is released by LinguaEmpire 7 days after client payment is received. No money moves through this page until a payment provider is connected — this page records jobs, invoices and balances. Express (PayPal/card) is recorded as the chosen payment method only; no Express payment is executed here.

Sign in, then press "Load supplier payments".

Supplier rates

Only an admin sets rates. Anyone signed in can list their own rates; leave the supplier ID blank to list your own.

Revenue & Margin

Live from GET /api/v1/finance/revenue-margin. Revenues, COGS (money paid to suppliers), margin in absolute terms and as a percentage of COGS, revenue/cost/margin per client, and weekly payments received, expected and overdue. "—" means the backend returned no value for that item.

Figures are grouped by currency because no exchange-rate source is configured. An invoice unpaid 30 days after issue counts as overdue.

Sign in, then press "Load finance report".

Translation Memory

Search the shared memory with syntax-aware matching, patch fuzzy matches to fit the new segment, recall phrase fragments and repair them. Every suggestion comes from a real memory row, and every repair shows what changed and why.

Patch and repair a fuzzy match

Terms found while translating

Memory Builder

Turn a project's confirmed segments into reusable memory, fragments and terms. Draft segments are never converted.

Penalty profiles

Lingua CAT Tool — In-editor TM maintenance

Edit or delete TM records from the editor (manager and above). Concordance relevance floor is 70%: matches below 70% are not shown as concordance results. Changes can be undone until you leave this panel.

Lingua CAT Tool — Penalty profile editor

Per-penalty values, default-profile selector and per-project override. Changing a value re-runs the editor's match display (the penalty-adjusted score is computed server-side).

Translation Memory

The workspace-wide shared memory: every confirmed segment enters it at once, so a fix reaches the next translator immediately. Match percentages include 102% context matches. Minimum match, minimum quality tier and minimum word count are configurable below.

Rules

Import & export

Import accepts .tmx, .xlsx (source | target columns) or tab-separated .txt/.tsv. Export writes .tmx or .xliff. Rows already in the memory are updated, not duplicated.

Reference material

Add a source document and its translation: the platform aligns them sentence by sentence and offers the pairs as reference matches while you work. A single document stays searchable through concordance, and a pair that does not line up is reported as unaligned, never forced. Reference material is not a translation memory: its matches are labelled "Reference", ranked below exact translation-memory matches, and never counted as memory scores. Attach material to a project with its project ID, or leave it blank for client-wide material.

Concordance

Screening tasks

Screen a vendor with a paid test task — a small scoped assignment at a stated rate, with a deadline and an evaluation rubric. The vendor completes it here, you score it against the rubric (LQA-based) and decide hire or no-hire. The result and score stay on the vendor's profile.

Payment for a screening task is not collected here yet: no payment gateway is connected, so no payment has been made or recorded. The task, its score and the hire decision are recorded; payment waits on the gateway decision.

Sign in, then press "Load screening tasks".

Create a screening task

Your screening tasks

Purchase orders

Costs from linguists' rate cards — generate a purchase order per linguist and download it as a PDF. The freelancer keeps 100% of the agreed rate; commission is charged to the client on top, never deducted here. Payment of a purchase order needs a payment gateway; none is connected yet, so no payment has been made or recorded.

Sign in, then press "Load purchase orders".

Rate cards

Create a purchase order

TM bands and net rates: New 100%, 95–99% 80%, 85–94% 60%, 75–84% 50%, 50–74% 30%, exact/repetition 20% for translation and 100% for review.

Generate after delivery

One purchase order per linguist with completed work on a delivered project, costed from their rate card and the words in their completed segments. A linguist with no verified per-word rate card is listed as skipped, never costed from an invented rate.

Purchase orders

Vendor performance

Translator Accuracy by match-score range, Top Members, Task Usage, on-time rate, LQA trend, jobs completed and earnings — computed from live records only. "—" means the backend returned no value for that item.

Sign in, then press "Load vendor performance".

Developer

Create and manage API keys. A new key's full value is shown once, right after creation — copy it then. It is kept only on this page while this tab is open and is never saved to a file.

Your API keys

Sign in, then press "Load keys".
Key prefixNameScopesCreatedRevoked
Keys not loaded yet.

Developer — API reference

Formats available on this backend (GET /api/v1/formats) and the main endpoints. Replace le_YOUR_KEY with a real key from above.

Example

curl -H "Authorization: Bearer le_YOUR_KEY" \
  /api/v1/projects

Main endpoints

  • GET /api/v1/projects — list projects
  • POST /api/v1/projects — create a project
  • GET /api/v1/projects/{id} — one project
  • GET /api/v1/projects/{id}/pipeline — pipeline run
  • POST /api/v1/translate — translate (see /api/v1/formats for accepted formats)
  • GET /api/v1/pairs/languages — 150 languages
  • GET /api/v1/engines — translation engines
  • GET /api/v1/formats — accepted file formats
  • GET /api/v1/webhooks/events — webhook events + retry policy
  • GET /api/v1/connectors — connector registry
  • GET /api/v1/developer/cli — CLI reference
  • GET /api/v1/models — model registry
  • GET /api/v1/models/capabilities — capability table
  • GET /api/v1/byo-llm — BYO LLM configurations

Developer — BYO LLM

Bring your own large language model (Enterprise / Enterprise Plus): OpenAI, Anthropic, Google, Amazon, DeepL, Gemini, or a custom endpoint. The credential is never stored here — give the name of the environment variable the server reads it from (for example OPENAI_API_KEY). A key typed below is used for one connectivity check only and is never saved, shown again, or sent anywhere else.

Your BYO LLM configurations

Sign in, then press "Load configurations".
ProviderModelEndpointCredential variableLast check
Configurations not loaded yet.

Model Builder

Build and manage custom models, plug in third-party language models, and assign a model to a project. Only the per-client adaptation layer updates immediately from your translation memory and glossary — no foundation model is ever retrained here. Third-party models are refreshed by a manual Retrain, which rebuilds the adaptation snapshot from the live Data Sources.

Create a model

Your models

Sign in, then press "Load models".
NameKindProviderBase modelUse caseStatusReal-time adaptation
Models not loaded yet.

Model Builder — Capability table

Which models fine-tune on segments, which on terminology, and which adapt in real time. Only the per-client adaptation layer adapts in real time; every third-party model needs a manual Retrain.

ModelSourceKindProviderFine-tune on segmentsFine-tune on terminologyReal-time adaptation
Capability table not loaded yet.

Model Builder — Data Sources & Retrain

Attach live translation-memory entries and glossary terms to a model as its Data Sources. Retrain rebuilds the adaptation snapshot from those entries and the client's current totals — it never retrains a foundation model. Per-client models adapt in real time already, so they have no Retrain button.

KindSource IDAttached
No Data Sources loaded yet.

Model Builder — Assign to a project

A project takes a model OR an AI profile, never both. Assigning one while the other is set is refused — unassign the other first.

AI profile (one per project, as an alternative to a model)

Webhooks

Register a URL to receive an event every time a project or document changes state, an export finishes, or an LQA report is ready. Every delivery is signed (HMAC-SHA256) and every event has its own delivery status below — not just a list of past errors.

Deliveries are signed with a key held in the server environment. If that key is not set, deliveries are held as failed and nothing is sent unsigned. Retry policy: first attempt at once, then after 1 minute, 5 minutes, 30 minutes, 2 hours and 12 hours (6 attempts in total); after that a delivery is marked dead until replayed.

Your webhooks

Sign in, enter a client ID, then press “Load webhooks”.
URLEventsActive
Webhooks not loaded yet.

Deliveries — status per event

EventStatusAttemptsLast errorNext retry
Deliveries not loaded yet.

Git integration

Connect a repository and pull source resource files in — a pull enqueues them for translation through the pipeline. Pushing translated files back is a manual action from this page once a project is translated; there is no automatic push when a project finishes or a workflow stage is reached.

Your provider token is supplied with each pull or push and is used once — it is never stored. GitHub pull/push is live. GitLab and Bitbucket connections can be registered now; their pull/push follows the same pattern and is enabled once verified against a live account.

Your connections

Sign in, enter a client ID, then press “Load connections”.
ProviderRepositoryBranchSource filePush back
Connections not loaded yet.

Pull / push

Sync history

Connectors

Connectors that work are shown as live. Entries still being designed show the decision they need — they have no connect button.

Public URL import — live

Import a .docx, .pdf or .txt file from a public URL into a project. Only public http(s) URLs are accepted; private-network addresses are blocked.

Developer CLI

The LinguaEmpire CLI wraps the same API for CI/CD — login, init, import, export, upload, download, pull and push — with a per-folder config file. It needs only Python 3 (no install step) and signs in with an API key or a JWT, never a hard-coded credential.

python3 cli/linguaempire.py login --api-url https://YOUR-HOST --api-key le_YOUR_KEY
python3 cli/linguaempire.py init --api-url https://YOUR-HOST --client-id YOUR-CLIENT-ID --pair en/fr
python3 cli/linguaempire.py import --file messages.json
python3 cli/linguaempire.py export --project PROJECT-ID --format txt
python3 cli/linguaempire.py pull --connection CONNECTION-ID --project PROJECT-ID
python3 cli/linguaempire.py push --connection CONNECTION-ID --project PROJECT-ID

Coworkers

The Content Creator generates real content through the backend's content coworker (a live LLM call). If the LLM is unavailable, the page says so — it never shows canned text as generated content.

Content Updater

Rewrites or updates existing text through the backend's content coworker (a live LLM call). If the LLM is unavailable, the page says so — it never shows canned text as updated content.

Converse

Live spoken conversation between two people in different languages. Each spoken turn is transcribed, translated through the translation engine, and spoken back. Usage is counted per session and device — never per minute, and nothing is charged.

Speech status

Speech-to-text
—
Voice output
—
Deployment
—

Sign in, then press “Check speech status”.

Start a conversation

Both directions of the language pair must be among the approved roster pairs. Tone controls: formal/informal register for Korean and Spanish; gender marking for Spanish.

Live conversation

Speak into your microphone. The transcript, the translation and the measured time for each turn appear below. If speech-to-text or voice output is not configured on the server, the page says so — it never shows a made-up transcript or audio.

No conversation started yet.

Turns

No turns yet.

Conversation history

Your conversations, with every turn and its translation. Only the two parties in a conversation can open it.

Sign in, then press “Load conversations”.

Usage

Counted per session and device. No per-minute counting and no charge.

Devices
—
Sessions
—
Turns
—
Charges
—

QA

Runs the backend QA checks on one project (POST /api/v1/projects/{id}/qa). Per-segment checks and a summary, read live from the backend.

AI Quality

Quality-based routing, predictive typing, Language Guard, Translation Smells, Adaptive MT and content-based routing — all read live from the backend (/api/v1/ai-quality/…). Nothing here is canned: where a feature needs the language-model lane and it cannot be reached, the page says so.

Quality-based auto-routing (QPS)

A Quality Performance Score (0–100, from the same checks as the QA tool) routes each segment automatically. High scores skip post-editing; low scores route to post-editing; the middle band goes to LQA review. An open Human Gate always wins over any score.

Predictive typing

Type a translation and press Tab to accept the suggestion. Suggestions come first from this client's own translation memory; only when the memory has no match does the language-model lane complete the text — and if the lane cannot be reached, the page says so instead of guessing.

Language Guard

Checks a translation for biased or harmful phrasing, tone, and glossary breaches — the exact violation is shown. The guard also runs on every project QA run and on every saved segment.

Translation Smells & Proofread

Automated quality assessments. A smell is a probable-error indicator — each one carries the evidence it was raised on. Proofread gives contextual fixes; the language-model suggestions are marked when the lane cannot be reached.

Review Assessment

Compares a reviewer's edit with the previous translation: what kind of edit it is, and whether it made the segment better or worse — judged from the QA checks it resolved or introduced.

Adaptive MT

The engine learns from corrections as you work: an accepted correction updates this client's translation memory and glossary, and the next suggestion uses the corrected wording. This is a per-client adaptation layer — it does not retrain any foundation model.

Content-based auto-routing

Analyses incoming content and selects the engine and workflow before translation starts — legal, medical and financial content always goes to human post-editing plus LQA; technical content goes to the self-hosted engine first.

Workspaces

Group projects into workspaces. The active workspace is kept only while this tab is open, is shown in the navigation bar, and is honoured by New Job and My projects.

Your workspaces

Sign in, then press "Load workspaces".
NameIDParentCreated
Workspaces not loaded yet.

Workspace invites

Invite someone to the active workspace. Invites appear here in the platform — no email is sent.

Invites for the active workspace

Set an active workspace, then press "Load invites".
EmailRoleStatusCreated
Invites not loaded yet.

My invites

Invitations sent to you. Accept or decline each one — the decision is recorded through the backend.

Sign in, then press "Load my invites".

Templates

Reusable project templates: workflow and stages, languages, translation memories, glossaries, AI rules, deadline offset, project name mask, custom fields, assignment template, reference files and folder — in 3 steps, not 9. Editing a template never changes projects already created from it.

Sign in, then load your templates. Your client ID is taken from the sign-in card at the top of the Dashboard.

Step 1 — Basics and workflow

Step 2 — Linguistic assets and translation rules

Pre-fill rules, live from the backend: 100% matches are inserted and confirmed, 75–99% fuzzy matches are inserted unconfirmed, and segments with no match are translated by AI and left unconfirmed.

Step 3 — Assignments, deadlines, folder and custom fields

Your templates

New project from template

Pick a template, a language pair and a service. The project is created end-to-end through the backend — the same roster check applies, so a pair with no linguist cover is refused, not created.

AI Translation Profiles

A profile sets which translation memory, glossary and engine a project uses, plus custom instructions — and is assigned directly to a project. A project takes a profile OR a template, never both.

Your AI profiles

Assign a profile to a project

Translation rules

Rules that pre-fill target text before human work. Set them per project or per template; the most specific rule wins (project, then template, then client default).

Pre-fill a project

Runs the rules over a project's untranslated segments now. Segments that already have target text are never touched.

Folders

Organise projects into folders. Custom fields set on a folder are the defaults for the projects inside it, and are saved in templates.

Your folders

Move a project into a folder

Deadlines

Project deadline plus an individual deadline per workflow stage. Deadlines are visible to everyone who can view the project, and overdue state is computed from the live clock — never invented. A completed stage auto-verifies after 15 days if nobody reviews it; completing a later stage verifies the earlier one. The 15-day figure is the current default and is confirmed at go-live.

Complete a stage

Completing a later stage verifies every earlier completed stage that is still unverified.

Auto-verification sweep

Admin only: verifies every stage that completed at least 15 days ago and is still unverified.

Chief of Staff

The daily brief, live from GET /api/v1/chief-of-staff/brief. Exactly what the backend returns — nothing added.

Sign in, then press "Load brief".

Reviewer

Runs the backend reviewer on one project (POST /api/v1/projects/{id}/review). Per-segment issues with severity and a suggested fix, plus a summary — read live from the backend. Checks up to 100 segments per run.

Learning loop

Predicted translations in the CAT editor: accept, amend or reject each prediction and the per-client adaptation layer learns from it at once — the accepted wording updates this client's translation memory and glossary, and the next prediction uses it. The layer scans the client's historical translation memory for consistency. Terminology, tone and brand voice are learned per client only. No foundation model was retrained.

CAT editor

Load a job's segments, get a prediction for a segment, then accept it, amend it, or reject it. Every decision feeds the adaptation layer immediately, and the effect is visible in the next prediction or in the learning feed below.

Client: —

Adaptation

What this client has learned, from which correction, and when — terminology and tone signals included. Agent Verified outputs appear here as excluded until a human expert or secondary reviewer confirms them. This is a per-client adaptation layer — it does not retrain any foundation model.

SLA

Service terms for each plan, live from GET /api/v1/sla. Terms marked "to confirm" are not firm yet — they are still to be confirmed.

Governance

Audit trail and your roles, live from the backend. Filter the audit trail by entity type and action.

Certificate events

Quick filters for Certificate of Translation audit events. Only events that really happened are listed — a filter with no rows means nothing of that kind was recorded yet.

Governance — my roles

Your roles, live from GET /api/v1/users/me/roles.

Sign in, then press "Load my roles".

Client Portal

Request a quote, review the estimate a PM sets, approve it, track the work and download the multilingual CSV — all from here. Payment inside the portal is not available yet: no payment gateway is connected, so there is no Pay Now button and paying returns a clear message instead.

Client Portal — quote actions

Load a quote to review it, set or edit the estimate (PM), approve or decline it, track its projects, download the multilingual CSV, or check payment status.

Client Portal — branding

Set the portal branding for a client: colours, subdomain and logo address.

Inter-company

Assign a whole project to another company as if it were its own. The vendor sets the service and rate on accept, attaches a quote or invoice the client downloads, and cannot pass the work on to a third company. The client never sees the vendor’s linguists, and revision authors are never shown here.

Inter-company — assignment actions

Accept with a rate, decline, cancel, assign a linguist, attach a quote or invoice, or check progress. Progress shows counts and pipeline status only — no linguist names.

Stakeholder review

Share a review link with a stakeholder: they review in the browser with no install, no seat and no account. They can comment and approve; approval uses up the link. A used or expired link is refused.

Stakeholder review — open a link

Paste the review link key to open a review, comment on a segment, and approve it.

Drive

The in-platform file library. Admin, Project Manager and Resource Manager see Drive content; Language Lead and Linguist see only files linked to projects they are assigned to. Files are stored on the platform — this is separate from any external storage connector.

Drive — link a file to a project

Linking a file to a project lets the linguists assigned to that project see it.

Compliance

Compliance checks run against rules your workspace admin defines — forbidden terms, required disclaimers and restricted patterns. The platform invents no compliance rule of its own: if no rule is defined, nothing is checked. The wider design of what a Compliance Coworker should check is a decision for the Chief Executive, and is recorded separately.

Compliance — run a check

Check a project’s content, or any text, against the active rules. The report is stored and can be opened again later.

Custom coworkers

Build a coworker around your own workflow: a brief, tools and workflow steps. Running it calls the live language model once per step, chained — the output of one step is the input of the next. Enterprise Plus only. If the language model is unavailable, the run fails openly and returns no content.

Custom coworkers — run

Run a custom coworker on your input.

Chief of Staff skills

Personal skills apply to your own Chief of Staff brief; Global skills are shared across a workspace. A skill is an instruction that is added to the brief the Chief of Staff generates — it changes what the brief says. Counts are limited by plan; the limits shown are proposed figures, to confirm at go-live.

Help Center

Guides for the LinguaEmpire platform, by topic. Every article describes the product as it really works today — including its limits. For a question in conversation, use the Lingua Help chatbot on the public website.

Roles & permissions

Who can do what in the active workspace. Roles are enforced by the backend on every request — changing a role here changes what that person can actually do. Account type (Freelancer, LSP, Agency, Company) and workspace role (Administrator, Project Manager, Resource Manager, Language Lead, Linguist) are two different things, and both are shown for every member.

Set an active workspace in the Workspaces view first.

Sign in, set an active workspace, then press "Load roles".

Permission matrix — preset roles

PermissionAdministratorProject ManagerResource ManagerLanguage LeadLinguist
Roles not loaded yet.

This matrix is LinguaEmpire's own. It follows LinguaEmpire's verified access rules: Administrators, Project Managers and Resource Managers see Drive content, while Language Leads and Linguists see only documents assigned to their tasks; Administrators and Project Managers edit and approve glossary terms, Linguists suggest only.

Custom roles

No custom roles loaded yet.

Members

NameEmailAccount typeWorkspace roleChange role
Members not loaded yet.

Drive access for my role

What you can see in Drive depends on your workspace role: Administrators, Project Managers and Resource Managers see all Drive content; Language Leads and Linguists see only documents assigned to their tasks.

Sign in, set an active workspace, then press "Load Drive access".

Glossary suggestions

Members who can suggest but not edit glossary terms propose them here. An Administrator, Project Manager or Language Lead approves or rejects each suggestion; approving adds the term to the glossary.

Suggestions in this workspace

Suggestions not loaded yet.

Notifications

Your notifications, and which emails you receive. Email is on by default for every event. The in-platform list below always works — it does not depend on email.

Email preferences

Sign in, then press "Load preferences".

"Projects managed by me" filters email only — the list below still shows every notification for you.

Your notifications

Sign in, then press "Refresh".

Notification settings

Choose how much each project notifies you: all notifications, mentions only, or muted. In-app notifications stay on — a project can be muted, but the in-app channel itself cannot be switched off. Email is on by default.

Projects I watch

Watch a project to be notified when its source text changes. You get one grouped email after the first change — the email itself is only sent once a mail provider is connected; the in-platform notification always arrives.

Expertise

Performance-derived expertise for the roster linguists on one pair, live from GET /api/v1/human-intelligence/expertise. Every score comes only from recorded performance — completed jobs, LQA scores, Human Gate decisions, on-time delivery and recorded screening outcomes. A linguist with no recorded performance for the pair shows "—", never a made-up score. Admin only.

LinguistJobsDeliveredAvg LQAOn-timeGateScreeningSuitabilityWhy
Expertise not loaded yet.

Expertise — one linguist

All pairs for one linguist, live from GET /api/v1/human-intelligence/expertise/{id}. Admin, or the linguist themself.

Routing

Suitability-ranked proposals for a pair, domain and risk level, live from GET /api/v1/human-intelligence/routing. Routing proposes — it never assigns anyone. Assignment happens only through the invitation / Human Gate decision. Only roster-verified linguists for the pair are ever proposed.

LinguistSuitabilityRecommendedJobsAvg LQAWhy
Proposals not loaded yet.

Calibration

Agreement between judges on the same item, live from GET /api/v1/human-intelligence/calibration. Disagreement is recorded as diagnostic data, not as a failure — it never changes either judgment. A drift flag appears when a judge's agreement rate falls below 0.70 across 5 or more comparisons. Admin only.

Benchmarks

Multilingual benchmarks and model evaluation, live from GET /api/v1/benchmarks/suites. Every score comes from real stored judgments and real LQA records under one rubric — a locale with no judgments shows “—”, never a filled-in number. Disagreement is shown as a diagnostic, never averaged away.

Sign in, then press “Load benchmarks”.

Benchmarks — new suite

Create a suite, then add anchors and gold items to it. Gold items are real stored items created from real translations and judgments — nothing is filled in for you.

Train

Governed training data — SFT, RLHF and RLVR — built by credentialed contributors only, with full lineage and governed delivery. Every number below comes from a live API call; unreachable data shows “—”.

Train — new dataset

Expert task design: a brief and rubric for one language pair, inside one customer scope. The pair must be one of the approved roster pairs. Exactly one scope is required — an organisation or a client you belong to.

Train — task design and annotation queue

Assign an annotation task to a credentialed contributor only — a verified linguist pair for this pair and domain, with a real performance record (LQA 90 or a Human Gate approval). Unvetted assignees are refused by the server.

Train — quality measurement

Agreement analysis, rater calibration against anchor items, and drift detection — computed from real judgments. Empty data shows “—”, never an invented figure.

Train — governed delivery

Export a dataset as a real file (JSONL or CSV) with full metadata, per-item lineage and documentation. Data never leaves the dataset scope without an explicit export decision recorded here first.

Certificates

Certificates of Translation, live from the backend. A certificate stamps one delivered language version with who approved it and when, the LQA score and the audit-log reference. Certificates are never changed once issued — a correction issues a new version.

Certificates — project audit trail

The full audit log for the project above: creation, view, edit, comment, approve, export and delete events with who did each one and when. Only events that really happened are listed — types with no rows yet are shown as missing, never filled in.

Book a demo

Book a live walkthrough of the LinguaEmpire platform with our team. No sign-in is needed to book.

Bookings

Bookings made through this page. Admin only — a non-admin sign-in gets "admin only".

Sign in as an admin, then press "Load bookings".

Lingua CAT Tool

Edit one project segment by segment: source on the left, translation on the right. Protected items (variables, serial numbers, email addresses) cannot be dropped from a translation; a segment being worked on is locked to everyone else until it is confirmed; every confirmed segment enters the shared Translation Memory at once. Live quality flags show on each row.

Draft: Synced

Shortcuts inside the editor: Ctrl/⌘+Enter confirms the active segment · Ctrl/⌘+F focuses Find · Ctrl+G concordance on the selection · Ctrl+H find & replace · Ctrl+K command palette · F9 next unconfirmed · F7 next terminology error · Esc dismiss. Every shortcut also has a visible button.

Lingua CAT Tool — QA panel

All issues in the open file. “Fix All” only fixes the auto-fixable checks (double spaces, end punctuation copied from the source, repeated words); every other check has no auto-fix and its Fix All button is disabled with the reason.

Open a project to see its QA issues.

Lingua CAT Tool — Changes

Tracked changes. Track changes is ON by default in Review mode. Accept or Reject per change, or in bulk.

No changes loaded yet.

Lingua CAT Tool — Preview

Direction: left-to-right

Open a project, then choose a preview tab.

Lingua CAT Tool — Context

Screenshots with a red outline, file context, labels and max length for the active segment. Plural forms (CLDR) are built second — see the note below.

Select a segment to see its context package.

Lingua CAT Tool — Keyboard shortcuts

Nothing here is shortcut-only: every shortcut has a visible button doing the same thing.

Lingua CAT Tool — Command palette (Ctrl+K)

Esc closes. Ctrl+Shift+K shows the shortcut list; Ctrl+Shift+L shows Key Assist.

Key Assist

Esc closes.

Concordance

Look up how a phrase was handled before, in context, without leaving the segment. Searches the shared Translation Memory first, then live project segments.

Glossaries

Approved term translations, in as many languages as a glossary needs. The memory is checked first — a segment from an exact memory match keeps its translation and the glossary is not applied to it; otherwise the glossary is enforced and the quality check flags segments where an approved translation was not used. Admins approve terms; linguists can only suggest them.

Add a term

Glossary check

Flags every segment of a project where an approved glossary translation was not used.

LQA & Quality

MQM-based LQA profiles: each error carries a category, a weight and a severity — Critical 25, Major 5, Minor 1, Neutral 0 — producing a 0–100 score per project against the unchanged pass mark of 90. The Translation Quality Score applies the same maths to AI output that is not yet confirmed, and lists the areas review should focus on.

Label an error & score

Comments

Notes on a whole document or on one segment, with threaded replies and a status of open, in progress or resolved. A document note reaches every language of the document; a segment note reaches the same segment in every other language.

Contact us

Send a message to the LinguaEmpire team — a person replies; nothing here is automated. Your message is stored as a contact request for the admin team.

Style Guides

A stored style guide for a workspace or client, applied as instructions to Reviewer and AI jobs and checked by the Reviewer. This is LinguaEmpire's own style-guide store: the role-model study references style guides as a Reviewer input but no standalone style-guide editor of theirs was verified, so this is built as LinguaEmpire's own implementation.

Apply a guide to a project

Website Translator

Translate any website with one line of JavaScript — no change to the site's CMS or backend. Give the platform a public URL: it fetches the page, translates every visible text node through the six-stage pipeline, and serves a translated preview with in-context review. Add the snippet to the site and visitors see the translated site.

CDN delivery and the SEO proxy are a later phase: the snippet translates in the visitor's browser, so search engines see the source language. Only public pages are fetched — private, loopback and link-local addresses are refused, and pages over 2 MB are refused.

Software localization

Localize app and website strings: import keys (JSON, YAML, Android XML, iOS Strings, LocJSON), translate them through the six-stage pipeline, add screenshots as context for each key, and export the translated keys — pull them from the command line, or push a new keys file from this page after import.

Plural handling: Android plurals stay grouped (one / few / many / other) and ICU plural values are carried whole, never split. Screenshots are reviewer context; they are not fed to the translation engine.

Video & subtitles

Translate video and audio transcripts with a subtitle editor: upload SRT or VTT (or a plain transcript), edit each cue and its timing, and export SRT or VTT. Every cue runs through the six-stage pipeline, with LQA and the Human Gate per cue.

AI dubbing / voice-over is not available yet — no voice engine is approved — and this view says so rather than faking dubbed audio. Transcription is not built either: supply the transcript file; only the transcript is translated.

Image translation

Upload an image: the platform reads its text, translates it through the six-stage pipeline, and re-renders the image with the translated text in place — preview first, layout preserved.

OCR engine status loads when you open this view.

SCORM & eLearning

Translate SCORM packages and XLIFF exports: upload a SCORM zip, translate every text node through the six-stage pipeline, and download the rebuilt LMS-ready package. Subtitles (.srt/.vtt) inside a package and job-aid files are not translated yet. The course builder that turns slides and documents into interactive SCORM with quizzes is a later phase.

XLIFF import

An XLIFF export (1.2 or 2.0) can also start a SCORM project — one segment per trans-unit — and runs the same pipeline.

Orchestrator

Visual no-code workflows: drag a trigger and an action onto the canvas, save, and activate. Every block that can execute does execute for real — a block that cannot run yet is marked with what it comes with, and the run record says so.

Canvas — drag blocks here

Drag a block from the palette into the canvas, or press “Add”. Triggers fire the workflow; actions run when a trigger fires. Blocks marked “comes with…” cannot execute yet and are never shown as running.

Palette

Trigger: File uploaded
Trigger: Project delivered
Trigger: LQA below bar
Trigger: Job created
Action: Create project
Action: Notify
Action: Route on LQA
Action: Parallel — comes with parallel workflows
Action: Branch — comes with parallel workflows
Action: Slack notify — comes with Slack notifications

Canvas

No blocks yet — drag blocks here.

Your workflows

Sign in, then press “Load workflows”.

Runs

OTA delivery

Over-the-air translation delivery: mobile and web apps fetch new strings without a new release. Publish a bundle — every string is translated through the pipeline first — and apps pull the update. Native iOS, Android and Flutter SDKs are the follow-up phase; the delivery API and the Python reference client are what ship now.

Publish a bundle

Your OTA apps

Sign in, then press “Load OTA apps”.

Inspect bundles

List every published bundle for one app, then open a bundle's strings. Strings are fetched with the same read key the app uses (?key=); leave the key blank for a public bundle.

Branching

Git-like branching for translations: branch a project — optionally matching a dev branch — edit translations in the branch, merge with real conflict handling, roll back to any earlier version, and find and link duplicate source strings.

Branch strings

Merge / rollback / duplicates

Chat translation

Live chat messages translated in real time between the two parties of a connection — each party reads the other’s messages in their own language. Brand and product terms are protected, and sensitive data (SSN, phone, email) is tokenised, never translated. Translation happens only inside an existing connection’s thread.

Translate a message

Website proxy

Reverse-proxy website localization: the proxy serves your translated site in the visitor’s language (browser language or location) with no code changes to your website. One setup step is needed first: point your domain at the proxy host — the step is shown on each site below. Content rendered by JavaScript on the visitor’s page is not translated by the proxy.

Your proxy sites

Sign in, then press “Load proxy sites”.

View a translated page

Multilingual SEO

Translated sites get real multilingual SEO: hreflang tags, translated metadata and localized URLs — read from the proxy’s live cached pages, never assumed. Localize a URL slug first, then check the report.

Community

Community translation: suggest translations for a string, vote for the best, and the winning suggestion is the one used. Glossary terms can be upvoted or downvoted too. Uses your Client ID from New job (saved in this tab).

Open a string for the community

Community strings

Glossary votes

App Store

Installable apps and integrations. Only apps whose code works are listed here — an app appears only when it runs. Uses your Client ID from New job (saved in this tab).

Installed apps

Run Public URL import

Fetch a public .docx, .pdf or .txt file from a URL into a project. The app must be installed above first.

AI Assistant

Ask about your projects, strings, translation memory, glossary, reports, gates and credits — answers come from live backend data, shown as tables and charts. The Assistant can also act: create a project, upload a file (.txt, .csv, .docx, .pdf, .xlsx — text is extracted and the file can start translation in a project), upload text, submit a project for review, edit a source string, create an open task, or manage workflows (list/pause/resume; cancelling asks you to Allow or Deny first — nothing changes until you allow it). Automations run on their schedule, event or webhook when the dispatcher runs — or press Run dispatcher below.

Ask

Upload a file

Upload a real file (.txt, .csv, .docx, .pdf or .xlsx, up to 25 MB). The text is extracted and stored as an attachment; give a Project ID to also enqueue it for translation in that project.

Act

Actions waiting for your decision

Automations

An automation stores a trigger (schedule, event or webhook) and one action. Due automations run when the dispatcher runs — on its schedule on the server, or when you press Run dispatcher. Destructive actions (editing a source string, cancelling a workflow) cannot be put in an automation.

Marketing

Long-form marketing content, in reading order, segment by segment. Upload a document (.txt, .docx, .pdf, .md or .html) — the workspace applies your translation memory and glossary, and only high-risk copy goes to reviewers. Uses your Client ID from New job (saved in this tab).

Documents

Departments

Self-service translation for your own internal departments, with governed access: each department sets whether MT/AI is allowed, a maximum request size, an approval threshold and the target languages it may use. A department member sees only their own department. Uses your Client ID from New job (saved in this tab).

Create department

Departments

How it works

The whole system is automated. People are involved only when the Chief AI says a job needs them.

1

You submit a job

Pick a source and a target language — the pair resolves itself — paste your text or upload a file (.txt, .docx or .pdf), and give the job a title.

2

AI translates

Translator and Reviewer AI agents translate and check every unit, in both directions of the pair. The roster gate (S1) first confirms a named native linguist covers the pair.

3

AI checks quality

Each unit gets an LQA quality score. The bar is 90 out of 100, and confidence must be at least 0.70. Low scores, retries and flags are recorded per unit.

4

Chief AI decides

The Chief AI approver looks at the whole job and decides: approve it directly, or send it to a human.

5

Human Gate — only when needed

Only when the Chief AI says "pass this through the Human Gate" does the job go to that pair's named native linguist. Otherwise it is approved directly and delivered.

6

Delivered

Approved jobs go to export and delivery. If the Human Gate rejects, the job is re-run with the linguist's notes.

What the numbers on this platform mean

Every number is read live from the backend at the moment you load the page. "—" means the backend did not answer or the item needs sign-in. This platform creates real jobs on the live backend. It does not publish, send or pay for anything.

Term discussions

Discuss a term, a segment or a whole document. Assign it to someone, follow it, star it or pin it, and set its status. An LQA error can be disputed here by opening a discussion on it.

Dispute an LQA error

Open a discussion on an LQA record to dispute it — the discussion is anchored on that record and its project.

Segment queries

Raise a question on one segment of a job, for the project manager and the linguists on that job. Only the job's parties can see or answer a query. This is separate from connection chat — it lives on the segment, inside the job.

Assist

Describe what you need in plain English — translate a file for a campaign, create content, manage terminology, or ask for an analytics report. Assist plans the workflow, you review and approve it, then it runs and reports progress. Every number comes from the live backend.

File limit: 2 GB per file and 2 GB in total. Assist cannot create accounts or connector integrations.

Your Assist sessions

Past prompts, their plans and their results — live from the backend.

Sign in to see your sessions.

Bulk Actions

Pick the assets, describe the update in plain English, review the preview plan, then approve — Assist runs it in the background across up to 5,000 items. Destructive actions need an explicit confirmation first.

Your bulk actions

Past bulk updates, their status and progress — live from the backend.

Sign in to see your bulk actions.