363 papers on agents, sorted so you don't drown in arXiv

What it does

This is a hand-maintained awesome-list that tracks AI agent papers published from January 2026 onward. The maintainers trawl arXiv weekly, keep only what is directly relevant to building or understanding agent systems, and bucket everything into five sections: multi-agent coordination, memory & RAG, evaluation & observability, agent tooling, and security. At 363+ papers and counting, it is essentially a human-curated RSS feed with categories.

The interesting bit

The curation is the product. The README does not just dump links; each entry includes a one-sentence distillation of what the paper actually tries to do, which saves you from clicking through to yet another “LLM agents are important” abstract. The list also appears to be a soft marketing channel for the maintainers’ own agent framework, VoltAgent, though the papers themselves are framework-agnostic.

Key highlights

• 363+ papers as of the last commit, with per-section counts visible (Multi-Agent: 53, Memory & RAG: 57, Eval & Observability: 80, Agent Tooling: 95, AI Agent Security: 82)
• Updated weekly from arXiv with explicit filtering for agent-ecosystem relevance
• Each entry includes a plain-English summary and a direct arXiv link
• Covers concrete engineering concerns: tool use, long-horizon workflows, conformal prediction for noisy agent messages, monotonic scaling guarantees for multi-agent pools
• Explicitly excludes pre-2026 work, so it functions as a “what is happening right now” snapshot

Caveats

• The list is a README, not a searchable database; you scroll or use browser find
• Some arXiv badge links appear to have copy-paste errors (e.g., one paper links to the wrong ID)
• Curated by a team building a competing agent framework, so the selection lens may favor their worldview

Verdict

Worth bookmarking if you are actively building agent systems and need to stay current without subscribing to arXiv’s daily deluge. Skip it if you want deep synthesis or critical commentary; this is a well-organized index, not a review article.