# vLLM Brand Voice

> A high-performance, engineer-to-engineer voice that is technical, efficient, and community-oriented.

## Positioning
vLLM is a high-throughput and memory-efficient inference engine for serving open-source LLMs. It is built for developers and researchers who need to maximize hardware efficiency and slash inference costs through advanced optimization techniques like PagedAttention.

## Voice principles
*   **Efficient:** Use short, punchy sentences that mirror the speed of the software.
*   **Technical:** Lead with specific features and technical milestones rather than vague marketing promises.
*   **Action-Oriented:** Focus on what the user can do (deploy, run, maximize, slash) rather than just what the product is.
*   **Collaborative:** Maintain an open, community-driven tone that acknowledges contributors and shared resources.

## Tone by context
| Context | Tone |
|---|---|
| Marketing Hero | Bold and benefit-driven, focusing on performance metrics. |
| Technical Docs | Instructional, precise, and requirement-heavy. |
| Community/Support | Helpful, welcoming, and appreciative of contributions. |
| Project Updates | Informative and celebratory regarding new releases. |

## Lexicon
- **Use:** High-throughput, memory-efficient, slash costs, maximize, unified, pagedattention, community project, serving engine.
- **Avoid:** Not evident from captured copy (though the brand avoids flowery, non-technical adjectives).

## Messaging do's and don'ts
*   **Do:** Use strong verbs like "Slash," "Maximize," and "Deploy."
*   **Do:** Mention specific technical requirements (e.g., Python 3.10+, CUDA 13.x).
*   **Do:** Highlight the "open-source" and "community" nature of the project.
*   **Don't:** Use fluff or hype without backing it up with a technical feature (e.g., PagedAttention).
*   **Don't:** Use passive voice; tell the user exactly how to run the engine.
*   **Don't:** Over-complicate the value proposition; stick to "Easy, fast, and cost-efficient."

## Evidence
*   "Slash inference costs by maximizing hardware efficiency."
*   "The High-Throughput and Memory-Efficient inference and serving engine."
*   "One engine, endless possibilities."
*   "vLLM is a community project."
*   "Select your preferences and run the installation."