Hey everyone,

I run a small dropshipping e-commerce store and currently use a semi-automated Python script to generate listing copy and push drafts via marketplace API. However, it still requires manual inputs and URL passing, and for sure - looking for good products on the marketplace (as I use other sellers products for reselling).

I want to build an autonomous agent pipeline that handles this end-to-end or at least some of the process.

Hardware: NVIDIA GeForce RTX 4070 Ti (with 12GB VRAM) and 32GB DDR5

Agent architectures are brand new to me.

My current workflow is:

1. Manual identification of SKUs to dropship (just sufring through the marketplace offers one by one).
2. Manually copying product URLs from the offer - speficication and photos (if open to use by others, on popular product mostly) are uploaded later automatically from the marketplace side catalouge.
3. If I need - generating or enhancing product mockups and clean packshots locally using ComfyUI or AI Studio.
4. Running custom Python scripts to draft localized titles and marketplace-compliant, SEO-friendly descriptions, matching my own structured section/text blocks.
5. Using Python to push draft offers via the marketplace API.
6. Manual review in the Allegro seller dashboard before taking the listing live, adding other obligatory info.

The primary goal is a local agent that can navigate marketplace listings undetected by their strict anti-bot systems and automatically create item drafts (or at least cleanly dump the structured product data for my script to ingest).

Questions:

1. Can an RTX 4070 Ti reliably run agentic workflows with multi-step function calling? Or is local tool-calling still too brittle for production pipelines, making commercial APIs the obvious choice?
2. Which open-source agent framework or tool on GitHub fits?
3. What are the most reliable open-source tools or libraries to feed clean web data to agents?
4. Is a multi-agent setup worth the complexity here, or is a single model with structured tool-calling validation cleaner and less prone to hallucination?

Would love any guidance, repository links, or lessons learned from similar projects!