In development

bAIbel TMS is in development. It is not on sale, and nothing here is an offer.

bAIbelTMS
/tms/everything-else

Everything else

Everything below is the part nobody puts on a homepage.

It is also the part you will be using on a Tuesday afternoon in eight months, so here it is in full.

Your memories and terminology

TMX arrives in a dialect, and leaves in one.

memoQ and Trados write TMX differently, and a system that reads both into one shape and writes one shape back has quietly changed your assets. Each dialect has its own conversion to and from the canonical form, written as an exact pair, with a version the client and server pin together so the two cannot drift.

Import what you have. Export what the other party needs.

Context-aware matching, and it says how much context it found.

A full match is raised by one point for each neighbouring segment that also matches and is confirmed in the same memory, walked in both directions and stopping at the first difference. The amount of the boost is recorded on the match, so you can see why one hundred per cent from this memory is better than one hundred per cent from that one.

Two perfect matches break the tie on the codes.

When two candidates both reach full marks and context cannot separate them, the one whose inline formatting matches yours wins. That is the one you would have chosen.

Search is a real index, and the strategy is yours to choose.

Lucene underneath, with the matching strategy resolved per request, per memory, or per deployment. A memory can have its own default. A request can override it.

It will not deduplicate on a guess.

Confirming a segment can fold it into an existing entry rather than adding a near-duplicate — but only where the request carries the document it came from. Without that provenance the server does not dedup, because a dedup that guesses is one you discover six months later.

Terminology in and out, in the formats you already have.

TBX. MultiTerm. memoQ CSV. All three both ways.

And you can take a snapshot before you do anything alarming.

Memories and termbases both.

Quality

Seven checks built in.

  • Empty segments
  • Inconsistent translations of the same source
  • Numbers that do not survive
  • Punctuation
  • Inline-code integrity
  • Terminology the termbase says should have been used
  • Whitespace

And your own rules beside them.

Custom checks, ordered, with your own message on each. A rule that goes wrong is stopped after two seconds rather than taking the check with it.

Inline-code integrity without an AI model in the path.

Tag checking is deterministic. No model, no call, no cost, no confidentiality question to answer about it.

Projects, people and work

Projects, tasks, and who is carrying what.

Tasks by target language and type, assigned to people, with an open workload you can see.

The candidate list shows you who is qualified.

When you assign a task, the list is ordered by who can actually do it: language capability for that kind of work first, then who is least loaded. Alongside each name, whether their qualifications stand up against the regimes this client engages — a data processing agreement, a business associate agreement, standard contractual clauses, a completed training course.

Qualifications are a registry, not a fixed list.

A UK addendum, a Swiss one, a client's own security schedule, a sector code — add the instrument and record who holds it, without waiting for a release. And an instrument covers only the regimes it was explicitly registered as covering: registering one as covering nothing is a usable state, and it means nothing.

Somebody who has left stops appearing. Somebody who mistyped their password does not.

A disabled or erased account leaves the candidate list. Three fumbled passwords does not remove a colleague from the list for five minutes.

Two people cannot edit the same segment at once.

A segment is held while somebody is in it, released when they leave, and reacquired by the same person if their connection drops.

Knowing what happened

Three reports, and you can take them away as CSV.

Productivity. Memory usage. Project progress.

And a processing record that knows who received the document.

Filed where only you can reach it, collected by someone who signed in, or taken away by another company's software. Or nothing, deliberately, when the run's product was its findings.