OS-Copilot / FRIDAY

by Shanghai AI Lab / MMLab (Wu et al.)

System Card

OrganizationShanghai AI Lab / MMLab (Wu et al.)

Released2024-02

Architectureagentic-workflow / Skill library + long-term memory + tool discovery

DetailsGeneralist OS agent with a Configuration Tracker using dense retrieval over long-term memory to recall tools, user profiles, and working directory state. Self-improves via accumulated skills across Excel, PowerPoint, web, code, and multimedia applications.

Parameters—

Domainagent-memorylifelong-learning

Open SourceYes

PaperView Paper

WebsiteVisit

gaia-benchmarkos-agentskill-libraryself-improve

Capability Profile

Benchmark Scores

6 of 14 benchmarks

Long-Context Retrieval

0/5

RULER

no data

NIAH

no data

LooGLE

no data

LongBench

no data

∞Bench

no data

Multi-Turn Recall

2/2

8086p

75.349p

Cross-Session Memory

1/1

83.393p

Multi-Hop QA

2/3

74.243p

MultiHop-RAG

no data

75.775p

Agent Task Memory

1/1

7226p

Personalization

0/1

PerLTQA

no data

Factuality / Grounding

0/1

RAGAS

no data

Sources:OS-Copilot / FRIDAY paper (arXiv:2402.07456); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)OS-Copilot / FRIDAY paper (arXiv:2402.07456); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)OS-Copilot / FRIDAY paper (arXiv:2402.07456); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)OS-Copilot / FRIDAY paper (arXiv:2402.07456); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)OS-Copilot / FRIDAY paper (arXiv:2402.07456); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)OS-Copilot / FRIDAY paper (arXiv:2402.07456); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)