PaperAgent

Personal paper search and management agent for LLM reasoning systems optimization.

The first version focuses on:

• Searching DBLP and arXiv for papers about LLM inference, reasoning systems, serving, scheduling, caching, and optimization.
• Tracking target venues such as EuroSys, NeurIPS, OSDI, SOSP, MLSys, ASPLOS, SIGCOMM, NSDI, ICML, ICLR, and USENIX ATC.
• Keeping a local, portable research workspace that works on macOS, Ubuntu, and Windows.

Quick Start

python -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
python -m pip install -e ".[dev]"
paperagent search "LLM inference optimization" --source arxiv --limit 5

On Windows PowerShell:

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -U pip
python -m pip install -e ".[dev]"
paperagent search "LLM inference optimization" --source arxiv --limit 5

To customize topics or venues, copy config.example.toml to config.toml and edit it.

PaperAgent targets Python 3.11+ on macOS, Ubuntu, and Windows.

Project Layout

src/paperagent/
  agents/       Agent orchestration and future LLM workflows
  cli/          Command line entry points
  config/       YAML config loading and defaults
  connectors/  DBLP, arXiv, and future paper sources
  models/      Shared paper and search data models
  storage/     Local persistence, cache, and exports
  utils/        Cross-platform helpers
data/
  raw/          Original downloaded metadata
  processed/    Cleaned paper metadata
  exports/      Markdown, BibTeX, CSV, or reading lists
  cache/        Reusable source API responses
docs/           Design notes and setup guides
tests/          Unit tests

Next Milestones

1. Add SQLite storage for papers, authors, venues, tags, and reading status.
2. Add daily or weekly refresh commands for saved queries.
3. Add ranking rules for your interests: LLM reasoning, inference systems, KV cache, speculative decoding, serving, batching, scheduling, and distributed inference.
4. Add LLM-assisted summarization after the metadata pipeline is stable.