Navrio-AI-powered workspace - AI-powered workspace for modern operators
by•
All-in-one workspace for freelancers, agencies, and small businesses: clients, projects, invoices, documents, AI employees, automations, and analytics.
Replies
Best
Maker
📌
🚀 Excited to launch Navrio on Product Hunt!
If you're building with AI, give it a try and let us know what you think. Every upvote, comment, and piece of feedback means a lot and helps us improve.
Thanks for your support! 🙌
Report
💎 Pixel perfection
How does the AI employees piece actually work in practice, are they more like pre-built templates you customize or something you train from scratch on your own workflows?
Report
Maker
@nilgn2902160993 In Navrio’s current implementation, AI employees are pre-built role templates you customize, not models you train from scratch.
Here’s how they actually work:
Each employee has a name, role, emoji, and a system_prompt (src/lib/ai-employees.ts).
The is_seed flag means some come pre-installed (e.g., a sales assistant, project manager, accountant persona).
When you chat with one, the backend sends that employee’s system_prompt to the model (google/gemini-3-flash-preview) as the personality/instructions (src/routes/api/ai-chat.ts).
Conversations are saved per thread in ai_threads / ai_messages, so context is preserved across a chat.
So in practice: you pick a role template, then the behavior is shaped by its system prompt and your ongoing conversation. There’s no fine-tuning or workflow-training step in the current code — it’s prompt-driven, not custom-model-driven.
Replies
🚀 Excited to launch Navrio on Product Hunt!
If you're building with AI, give it a try and let us know what you think. Every upvote, comment, and piece of feedback means a lot and helps us improve.
Thanks for your support! 🙌
How does the AI employees piece actually work in practice, are they more like pre-built templates you customize or something you train from scratch on your own workflows?
@nilgn2902160993 In Navrio’s current implementation, AI employees are pre-built role templates you customize, not models you train from scratch.
Here’s how they actually work:
Each employee has a name, role, emoji, and a system_prompt (src/lib/ai-employees.ts).
The is_seed flag means some come pre-installed (e.g., a sales assistant, project manager, accountant persona).
When you chat with one, the backend sends that employee’s system_prompt to the model (google/gemini-3-flash-preview) as the personality/instructions (src/routes/api/ai-chat.ts).
Conversations are saved per thread in ai_threads / ai_messages, so context is preserved across a chat.
So in practice: you pick a role template, then the behavior is shaped by its system prompt and your ongoing conversation. There’s no fine-tuning or workflow-training step in the current code — it’s prompt-driven, not custom-model-driven.