Transept

Transept is an AI translation workspace built for people who care about how their words land in another language, not just whether they parse. Instead of framing translation as a one-shot machine job, it treats a project as an ongoing document with shared glossaries, styleguides, and a decision-context memory that records the wording chosen, the alternatives rejected, and the comments that explained the call. Files come in as DOCX, PDF, Markdown, TXT, or via Notion and Google Drive connectors; translated output can be exported back to those formats plus Google Docs and localization-industry TMX/XLIFF. The editor is side-by-side with sentence-level regeneration, custom rewrite directions, QA checks, and collaborative comment threads. Two quality modes are exposed as word budgets: Standard spends roughly one word per source word, while Pro spends about three, letting teams trade cost against depth of reasoning. A layer called Literess acts as an agentic literary editor that flags tone inconsistencies across a manuscript and can execute multi-step editing workflows, with up to 20 remembered preferences per account. Under the hood, Transept says it benchmarks Claude, GPT, Gemini, and open models rather than committing to a single family, so the product's value is really the wrapper: the glossary/styleguide enforcement, the decision memory, the team-shared pools, and the localization-friendly file plumbing. Typical workflows include translating a book or long-form manuscript into 8-12 languages while keeping character voice consistent, running client-specific styleguides across recurring marketing briefs, and localizing product content where terminology accuracy matters more than raw throughput. Free viewers are unlimited on paid plans, so reviewers, authors, and clients can sit inside the same project without extra seat cost.