Transept AI

Product Introduction

1. Definition: Transept AI is a cloud-based, AI-powered translation workspace and Computer-Assisted Translation (CAT) tool. It is a Software-as-a-Service (SaaS) platform designed for long-form, nuanced content translation, combining large language models (LLMs) with a structured, memory-driven workflow.
2. Core Value Proposition: It exists to solve the problem of voice and terminology loss in AI translation. Unlike standard machine translation tools, Transept AI systematically remembers a user's past translation decisions—including terminology choices, tone preferences, and stylistic rulings—and applies this "decision-context memory" across all future documents, projects, and team members to ensure unparalleled consistency and quality.

Main Features

1. Decision-Context Memory (Translation Memory+): This is the core technical innovation. The system doesn't just store source-target segment pairs in a traditional Translation Memory (TM) database. It captures the full context of a decision: the approved translation, the rejected alternatives, and any user comments explaining the rationale. This enriched data is then fed back as context to the LLM (currently Google Gemini) on subsequent translation runs, guiding it to replicate human judgment, not just words. This creates a self-improving, project-aware translation engine.
2. Integrated Glossary & Styleguide System: Users can create and manage centralized terminology databases (glossaries) and tone-of-voice rules (styleguides). These are not static references but active agents. When toggled on for a project, the system injects glossary terms and styleguide rules directly into the LLM's prompt for each relevant segment, ensuring enforced consistency across hundreds of pages and multiple languages without manual intervention.
3. Agentic AI Literary Editor (Literess): Beyond a chatbot, Literess is an AI agent with persistent memory. It reads translations, flags issues like tone flattening or meaning drift, and can execute workflows (e.g., batch proofreading, document drafting) based on user preferences remembered across sessions. It acts as a quality assurance (QA) layer and productivity tool, understanding project-specific context like "this is a sci-fi trilogy" or "use British English."
4. Sentence-Level Directional Control: Users can apply custom instructions (e.g., "more formal," "keep the metaphor," "translate as a proverb") to individual sentences. The system combines this directive with the active glossary and styleguide to regenerate that specific segment while leaving the rest of the document untouched, offering granular, context-aware editing.
5. Unified Workspace for Teams: The platform provides a collaborative environment with shared resources (memory, glossaries, styleguides), threaded comments on every text block, and shareable review links. It supports real-time collaboration and eliminates version chaos from emailing DOCX files, centralizing the localization workflow for distributed teams.

Problems Solved

1. Pain Point: Inconsistent terminology and loss of authorial voice/tone across long documents and translation projects, especially when using generic AI translation APIs or different translators.
2. Target Audience:
   • Professional Translators & Localization Freelancers: Who need batch processing, client-specific styleguides, and QA tools to deliver consistent, high-volume work.
   • Content Creators & Authors: Novelists, bloggers, and marketers who must translate their backlog while preserving their unique narrative voice and brand tone.
   • Localization Teams & Agencies: Teams managing multi-language projects that require shared glossaries, collaborative review, and a single source of truth for terminology and style.
3. Use Cases:
   • Translating a novel or series of books where character names, invented terms, and narrative tone must remain perfectly consistent.
   • Localizing a software documentation suite or knowledge base where technical jargon and UI strings must be identical across all languages.
   • A marketing team translating campaign materials (website copy, emails, ads) while strictly adhering to brand voice guidelines.
   • A solo translator managing multiple clients, each with distinct terminology preferences and style guides, needing quick context switching.

Unique Advantages

1. Differentiation vs. Competitors:
   • vs. Generic AI Translators (DeepL, Google Translate): Transept offers project-aware memory, glossary/styleguide enforcement, and collaborative workspace features, which generic tools lack.
   • vs. Traditional CAT Tools (memoQ, SDL Trados): It uses modern LLMs for superior initial translation quality and nuance, paired with an intelligent, context-aware memory system that goes beyond simple fuzzy matching.
   • vs. Other AI-Powered Tools: Its "decision-context memory" and the Literess agent provide a deeper layer of consistency and automated QA that is unique in the market.
2. Key Innovation: The "decision-context memory" system. By training the LLM on the reasoning behind past translations, not just the outcomes, it achieves higher consistency with cheaper models. The company's internal benchmarks show a model costing 4x less can match a premium model's quality when grounded with sufficient decision-context data.

Frequently Asked Questions (FAQ)

1. How does Transept AI's memory differ from a standard Translation Memory? A standard TM is a database for fuzzy matching of similar text segments. Transept's "decision-context memory" records the full rationale behind a translation choice—including rejected alternatives and user comments—and uses this to inform the AI's future decisions, ensuring it replicates stylistic judgment, not just textual matches.
2. What AI models does Transept AI use for translation? The platform currently utilizes Google's Gemini models. It operates with a "Standard" mode (1 source word = 1 credit) for fast, accurate translations and a "Pro" mode (1 source word = 3 credits) that uses a more capable, nuanced model for literature and complex marketing copy.
3. Is my data used to train the AI models? No. According to the provided schema, Transept AI has a "No training on customer content" policy. Your documents, translations, and memory are not used to train or improve the underlying foundation models.
4. Can I use Transept AI with my existing translation memories (TMX/XLIFF)? Yes. The platform supports importing existing translation assets via TMX and XLIFF formats, allowing you to bootstrap your decision-context memory with past work, as well as export in these standard formats.
5. How does the pricing work for teams? Creating a team and inviting members is free. You add paid "seats" (e.g., Starter plan at €29/month). Each seat contributes its monthly word allowance (e.g., 100,000 words) to a shared team pool, billed on a single invoice. This allows for collaborative resource management.