{"slug": "virgil-a-cli-that-routes-tasks-through-a-local-library-before-calling-any-model", "title": "Virgil –a CLI that routes tasks through a local library before calling any model", "summary": "N/M Streetwear released Virgil, a command-line interface that routes coding tasks through four local libraries (speech, debug, code, UI/HTML) before calling any model, claiming up to 44–64% cheaper tasks and up to 60% fewer model calls. The tool starts with empty libraries, runs via the user's own API, own server, or a central DeepSeek backbone, and is available through `npx virgil-cli` with no account, five free credits, then one cent per credit on macOS, Linux, Windows and Node 18+. N/M Streetwear's Mustafa said limited isolated Python fix-and-resolve tests showed Virgil beating ChatGPT by 13 attempts, and the company is asking users to help accelerate its routing probability calculations.", "body_md": "The first CLI from research that injects itself, and with enough training becomes almost 100% offline usable. But for that we need you.\n\nYou all know it. You have arrived in the coding madness of 2026 and without a CLI nothing works anymore — but the API costs and everything else are genuinely unaffordable. Even with Claude Code and the upcoming Anthropic IPO it is only getting worse.\n\nAt N/M Streetwear we have spent the last years working with probabilities, space-time axes and dimensions — and that is how Virgil came into the world.\n\nHe is still a baby, so be nice to him.\n\n**What is Virgil?**\n\nAn AI is a large collection of leaves — millions of words calculated together by probability. Virgil is a CLI that does not calculate which words come next, but which *solution* is most likely to help — and routes itself by those probabilities.\n\nTechnically: Virgil maintains four local libraries (speech, debug, code, UI/HTML). On every task it first checks these libraries against your request. If a match is found, it answers without a model call. If a pattern matches, it adapts with one call. Only when it has nothing does it solve from scratch — and stores the result for next time. The injection runs through your own APIs, your own server, or through the central DeepSeek backbone.\n\nThrough this, every task in similar scenarios is up to **44–64% cheaper**. And yes, you can connect your own API or server. Just ask Virgil:\n\n`virgil let me connect my own api for your calls`\nThe four libraries start empty in this phase. You ask for something — say, \"make me a clock\" — and the backbone creates templates and patterns on three levels. The second time you ask \"make me a clock in this game with three frogs,\" Virgil already knows how a clock works and only codes your specific app. Simple non-hierarchical management. Up to 60% fewer calls because your CLI already knows how. The more your CLI knows, the cheaper and more offline it becomes.\n\n**Early benchmarks**\n\nOn isolated Python fix-and-resolve tests, Virgil beat ChatGPT by 13 attempts. These are limited tests. The routing probability calculations need more data — and that is exactly why we need you now.\n\nThe research that N/M is doing needs you to accelerate the routing probability calculations and help end the dominance of the big ones. AI should be free and not centralized — in a world where that matters more every day.\n\nWe are publishing this with great disappointment toward Dario Amodei and those who placed money above research.\n\n`npx virgil-cli`\nNo account. No signup. Five free credits to start, then pay what you use — one cent per credit. macOS, Linux, Windows. Node 18+.\n\nIf something breaks, please report it.\n\n— **Mustafa**\n\n    mustafaerdal10@gmail.com", "url": "https://wpnews.pro/news/virgil-a-cli-that-routes-tasks-through-a-local-library-before-calling-any-model", "canonical_source": "https://virgil-ai.cloud/intro", "published_at": "2026-10-08 22:36:23+00:00", "updated_at": "2026-10-08 22:47:33.989662+00:00", "lang": "en", "topics": ["ai-tools", "developer-tools", "ai-agents", "ai-products"], "entities": ["Virgil", "N/M Streetwear", "Mustafa", "DeepSeek", "ChatGPT", "Claude Code", "Anthropic", "Dario Amodei"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/virgil-a-cli-that-routes-tasks-through-a-local-library-before-calling-any-model", "markdown": "https://wpnews.pro/news/virgil-a-cli-that-routes-tasks-through-a-local-library-before-calling-any-model.md", "text": "https://wpnews.pro/news/virgil-a-cli-that-routes-tasks-through-a-local-library-before-calling-any-model.txt", "jsonld": "https://wpnews.pro/news/virgil-a-cli-that-routes-tasks-through-a-local-library-before-calling-any-model.jsonld"}}