An awesome list that actually checks its sources

What it does It’s a curated catalog of everything around computer-use agents: 1,673 reviewed papers and surveys, plus datasets, benchmarks, models, frameworks, and tools — 2,170 curated resources total. Coverage spans web, desktop, Android, iOS, and even HarmonyOS, organized by topic (grounding, planning, memory, RL, safety) and by platform. There’s a searchable website, but the whole catalog also ships as JSONL/SQLite and RAG-ready documents, with an llms.txt for language-model navigation.

The interesting bit Most awesome lists are link dumps with a shelf life. This one grades its own entries: curated records were checked against a first-party source snapshot, awaiting-review records have verified bibliographic metadata, and 2,062 discovered candidates sit in a queue awaiting resolution. It also tracks which code, datasets, and models are actually connected to which papers through primary evidence — the boring provenance work that makes the list trustworthy.

Key highlights

• 1,673 reviewed papers and surveys, grouped by topic and year, with original sources linked
• 86 datasets, 91 benchmarks, 98 model releases, 87 frameworks, 49 open agent implementations
• Machine-readable exports: JSONL, SQLite, RAG documents, and llms.txt for LLM consumption
• Local search via a dependency-free Python script (3.11+), no install required
• Also published as a Hugging Face dataset (TIGER-Lab/AwesomeCUA)

Caveats

• 451 records are still awaiting full review, and 2,062 candidates unresolved — the catalog is honest about being a work in progress
• The learning-resources section is thin: two courses/tutorials

Verdict Anyone doing GUI-agent research or building a computer-use agent should bookmark this — it’s the map of the field. If you just want a quickstart tutorial, look elsewhere (there are literally two).