# How Caveman Handles Different Languages: Portuguese, Spanish, and Beyond > Discover how Caveman supports multiple languages like Portuguese and Spanish. Learn how it preserves your dominant language using style compression for a seamless user experience. - Repository: [Julius Brussee/caveman](https://github.com/JuliusBrussee/caveman) - Tags: deep-dive - Published: 2026-07-06 --- **Caveman preserves the user's dominant language by applying style compression while leaving the original language untouched, as defined in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md) and implemented in [`src/hooks/caveman-activate.js`](https://github.com/JuliusBrussee/caveman/blob/main/src/hooks/caveman-activate.js).** Caveman is an open-source text compression tool designed to reduce token count while maintaining clarity across multilingual inputs. Unlike systems that default to English output, Caveman respects the user's original language—whether Portuguese, Spanish, or others—ensuring that **how Caveman handles different languages** remains transparent and configurable through its skill-based architecture. ## Language Preservation Rules The foundation of Caveman's multilingual support lies in its skill definition. According to [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md) at line 23, the rules explicitly mandate: > “Preserve user’s dominant language. User write Portuguese → reply Portuguese caveman. User write Spanish → reply Spanish caveman. Compress the style, not the language. No forced English openings or status phrases. ALWAYS keep technical terms, code, API names, CLI commands, commit‑type keywords (feat/fix/…), and exact error strings verbatim — unless user explicitly ask for translation.” This directive ensures that the system distinguishes between **stylistic elements** (articles, filler words, hedging) and **linguistic content** (grammar, vocabulary, sentence structure). By targeting only the former, Caveman achieves compression without translation. ## Implementation in the Activation Hook The practical execution of language preservation occurs in [`src/hooks/caveman-activate.js`](https://github.com/JuliusBrussee/caveman/blob/main/src/hooks/caveman-activate.js) at line 131. This hook activates the mode during each response cycle through the following process: 1. **Language Detection**: The system analyzes the user’s input using simple heuristics, such as detecting non-ASCII words or language-specific stopwords 2. **Flag Setting**: It sets a `language` flag corresponding to the detected dominant language 3. **Compression Application**: The flag is passed to the response generator, which applies **caveman-style compression**—dropping articles and filler—while preserving the original language structure The hook ensures that the compression engine operates on the token stream only after language detection completes, preventing any cross-linguistic interference. ## Technical Architecture Caveman's core engine does not perform language-specific tokenization. Instead, it works on the token stream after the language detection step, identifying universal patterns of verbosity common to formal writing. Because the compression rules affect only stylistic elements and never touch code blocks, API names, or technical terminology, the output token count is reduced without altering the language’s grammar or vocabulary. ## Language Handling Examples ### Portuguese Input Processing When receiving Portuguese text, the system maintains Portuguese output while removing conversational filler: **Prompt:** ```text Estou tentando conectar ao PostgreSQL mas recebo o erro “password authentication failed for user”. ``` **Caveman Response (Full Intensity):** ```text Erro “password authentication failed for user”. Verifique senha, usuário, host, porta. Use `psql` -U usuario -h host -p 5432. ``` The response remains in Portuguese; only filler and pleasantries have been stripped. ### Spanish Input Processing Spanish inputs similarly retain their linguistic identity through the compression process: **Prompt:** ```text Mi aplicación se bloquea cuando llamo a la API REST y recibo un 500. ``` **Caveman Response (Lite Intensity):** ```text Bloqueo al llamar API REST, recibe 500. Revise logs, traceback, punto de fallo. Asegure que endpoint exista y que cuerpo sea válido. ``` The output stays in Spanish while achieving the terse caveman style. ### Explicit Translation Override The preservation rule includes a specific exception for translation requests. When users explicitly ask for translation, the system disables the preservation flag: **Prompt:** ```text Estoy usando Node.js y quiero que me traduzcas al inglés antes de comprimir. ``` **Caveman Response (Full Intensity):** ```text Error “module not found” in Node.js. Check import path, file extension, and that the module is installed. ``` Because the user requested English translation, Caveman converts the language before applying compression, demonstrating the flexibility of the `language` flag logic. ## Key Files for Multilingual Support Several files work together to ensure Caveman respects the user's original language: - **[`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md)**: Defines the preservation rule at line 23 - **[`src/hooks/caveman-activate.js`](https://github.com/JuliusBrussee/caveman/blob/main/src/hooks/caveman-activate.js)**: Implements the activation logic and `language` flag at line 131 - **[`src/rules/caveman-activate.md`](https://github.com/JuliusBrussee/caveman/blob/main/src/rules/caveman-activate.md)**: Documents the human-readable activation logic - **[`src/plugins/opencode/commands/caveman.md`](https://github.com/JuliusBrussee/caveman/blob/main/src/plugins/opencode/commands/caveman.md)**: Provides command documentation for invoking the mode ## Summary - Caveman preserves the user's dominant language rather than defaulting to English, as defined in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md) - The `language` flag set in [`src/hooks/caveman-activate.js`](https://github.com/JuliusBrussee/caveman/blob/main/src/hooks/caveman-activate.js) controls whether compression preserves or translates the input - Technical terms, code blocks, API names, and error strings remain verbatim regardless of the surrounding language - Language detection relies on simple heuristics before the compression engine processes the token stream - Users can override preservation by explicitly requesting translation in their prompt ## Frequently Asked Questions ### Does Caveman support all languages or only specific ones? Caveman supports any language the underlying model can process. The system uses simple heuristics in [`src/hooks/caveman-activate.js`](https://github.com/JuliusBrussee/caveman/blob/main/src/hooks/caveman-activate.js) to detect input language without requiring language-specific tokenization, making it compatible with Portuguese, Spanish, French, and others, provided the model recognizes the vocabulary. ### How does Caveman handle mixed-language inputs? The activation hook identifies the dominant language through heuristics like non-ASCII word patterns and stopword detection. While the specific logic for mixed inputs isn't explicitly segmented in the source, the system processes the token stream after language detection, suggesting it compresses based on the predominant detected language while protecting technical terms in their original form. ### Can I force Caveman to output in English even if I write in Spanish? Yes. The preservation rule in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md) includes an explicit exception: "unless user explicitly ask for translation." By requesting translation in your prompt, you disable the language preservation mechanism, allowing the system to translate to English before applying caveman-style compression. ### Does language preservation affect how technical terms are handled? No. The preservation rule explicitly mandates that technical terms, code, API names, CLI commands, commit-type keywords (feat/fix/etc.), and exact error strings remain verbatim regardless of language. This protection applies universally across all supported languages, ensuring that `psql`, `npm install`, or "password authentication failed" appear identically in Portuguese, Spanish, or English contexts.