Workflow Agent
An AI agent that turns arXiv into a personal daily digest.
agentic
LLM
arXiv
tools
Agentic pipeline: search arXiv, filter seen papers, summarize, write a Markdown digest. Tool-calling loop with Groq, SQLite memory, and rate limiting from day one.
An AI agent that automates a daily research workflow.
Every day the agent:
- Searches arXiv for topics you care about.
- Filters out papers it has shown you before.
- Summarizes each new paper in 2–3 bullets.
- Writes a Markdown digest file.
You talk to the agent in plain English; it decides which tools to call.
Design choices
- arXiv API over scraping
- SQLite for seen-paper memory (lookup, not vectors)
- Groq with tool calling (
qwen/qwen3-8b-class models) - Run-scoped store, streaming summarizer, rate limiting from day one
Phase 1 (agent core) is complete. Later phases: evals, cost tracking, reliability, and deploy.
Stack: Python, Groq, SQLite, uv.
Source: Utkarsh736/workflow-agent.
Back to top