{"slug": "building-an-ai-task-generator-for-vikunja-that-doesn-t-trust-itself", "title": "Building an AI task generator for Vikunja that doesn't trust itself", "summary": "A developer built Trof, an open-source AI task generator for the self-hosted project management tool Vikunja that requires human approval before any tasks are added or modified. The system uses a two-service architecture with a \"propose-before-apply\" boundary, ensuring the AI never directly writes to Vikunja but instead generates proposals that users must explicitly confirm. Trof is provider-agnostic, supporting Anthropic, OpenAI, Google, DeepSeek, or fully local models via Ollama, and can also edit existing projects by proposing diffs of tasks to add, edit, or remove.", "body_md": "I self-host [Vikunja](https://vikunja.io/) to organize my projects, and I kept hitting the same chore: turning a vague idea (\"set up CI for a Kotlin service\") into a clean list of individual tasks, by hand, every time.\n\nLLMs are obviously good at this kind of decomposition. But I didn't want an AI that silently writes a pile of tasks into my tracker and hopes for the best. I wanted it to *propose*, and then get out of my way so I can approve.\n\nSo I built **Trof** — an AI task generator for Vikunja with a human-in-the-loop step baked in. It's open source and self-hosted. This post is about how it works and a few decisions I made along the way.\n\nWhy \"Trof\"?It's named after the Trophy active protection system on the Merkava tank, which intercepts incoming threats before they reach the hull. Trof does the same for your task board: nothing lands in Vikunja until you've reviewed it.\n\nIt also works for editing existing projects, not just creating new ones — you describe the change, and it proposes a diff of tasks to add, edit, or remove.\n\nI split it into two services instead of one:\n\nThe agent workflow runs on [Koog](https://github.com/JetBrains/koog), JetBrains' agent framework for Kotlin. The frontend is React + Vite. Everything ships as Docker Compose behind nginx with TLS.\n\nIt's provider-agnostic — Anthropic, OpenAI, Google, DeepSeek, or fully local via Ollama, in which case nothing leaves your\n\nmachine. Your Vikunja token and API keys live in\n\nThe interesting part wasn't \"call an LLM and pars-before-apply** boundary.\n\nIt would have been easier to let the agent call t's basically what an MCP server does). But thenthe AI mutates your real data on every run, and you're left cleaning up after a confident-but-wrong decomposition.\n\nInstead, the AI never touches Vikunja. It produces a *proposal*; the gateway holds it; you edit it; and only an explicit confirm\n\nturns it into real API calls. The AI proposes, yo manages my actual projects, that tradeoff wasworth the extra plumbing.\n\nThis one I'm genuinely unsure about. When the description is vague, the agent pauses and asks a follow-up question instead of\n\nguessing. It produces noticeably better task breaound-trip and some friction.\n\nIf you've built conversational/agentic tools: do t, or generating a best-effort result and lettingthe user correct it? I keep going back and forth.\n\nIt's early and rough in places, but it works end to end. Repo with setup instructions:\n\n👉 [https://github.com/Yooshyasha/Trof](https://github.com/Yooshyasha/Trof)\n\nFeedback welcome — especially on the review flow and whether the clarifying-questions step is worth the friction.", "url": "https://wpnews.pro/news/building-an-ai-task-generator-for-vikunja-that-doesn-t-trust-itself", "canonical_source": "https://dev.to/yooshyasha/building-an-ai-task-generator-for-vikunja-that-doesnt-trust-itself-ie7", "published_at": "2026-05-31 10:44:15+00:00", "updated_at": "2026-05-31 11:12:45.730099+00:00", "lang": "en", "topics": ["ai-agents", "large-language-models", "ai-tools", "ai-products", "generative-ai"], "entities": ["Vikunja", "Trof", "Koog", "JetBrains", "Anthropic", "OpenAI", "Google", "DeepSeek"], "alternates": {"html": "https://wpnews.pro/news/building-an-ai-task-generator-for-vikunja-that-doesn-t-trust-itself", "markdown": "https://wpnews.pro/news/building-an-ai-task-generator-for-vikunja-that-doesn-t-trust-itself.md", "text": "https://wpnews.pro/news/building-an-ai-task-generator-for-vikunja-that-doesn-t-trust-itself.txt", "jsonld": "https://wpnews.pro/news/building-an-ai-task-generator-for-vikunja-that-doesn-t-trust-itself.jsonld"}}