Immich handles photo backup and search, and you can install it from Hub's Alternatives category. Run one low-risk library first and see if it fits your work. ci.computer/download #SelfHostedAI
2
An AI Overview's citations may not back its claims. A preprint on 3,611 health questions estimated 56.7% of cited claims were backed, and 22.2% of rubric-flagged overviews had harmful content. Check the source. arxiv.org/abs/2610.39339
1
18
Self-hosted apps sit in the Alternatives category of Hub's store. You can browse them from the app store. Try it with one client document first. ci.computer/download
1
Smart homes for an autistic child take ongoing family work. A qualitative study says remote monitoring can support independence but raises tensions among safety, privacy and control. Five caregiver interviews. arxiv.org/abs/2610.38008
1
Hub's store has an Alternatives category for open source options. Your own computer, your own apps. ci.computer/download
5
A retrieval step lifted a coding agent's resolve rate from 25.8% to 27.0% on SWE-Bench Verified in the authors' setting, and a weak retriever hurt. A modest gain. arxiv.org/abs/2608.05886 #AIAgents
4
When a creator's audience shifts, people blame the algorithm or the crowd, and that shapes the response. A study of 262 comments on eight Xiaohongshu threads found explanation type and strategy strongly linked. Small and exploratory. arxiv.org/abs/2610.38026 #AIResearch
4
Opinion from the Oxford Internet Institute: access to frontier AI can depend on decisions made elsewhere, so building capacity now matters. It's commentary, not a measurement. oii.ox.ac.uk/keeping-europes…
7
Before you rent another AI tool, weigh it against hardware, upkeep and your own hours. Core is one purpose-built option, and Hub does not require it. Details here: ci.computer/store/p/core #LocalAI
2
10
Smaller models for your own machine: a score can point the wrong way. A preprint found a reliable score predicted a 16.1% gain, yet picked models ended 6.0–7.7% worse than two controls. Test on your own task. arxiv.org/abs/2608.02940
7
Password manager on your own box? Vaultwarden is in Hub's app store, which also has an Alternatives category. Fit depends on your team size and upkeep time. ci.computer/download
13
A small model that consults a notes bank written by a bigger model beat a larger model on one LiveBench task and matched it on another, in the paper's tests. Weights unchanged. arxiv.org/abs/2610.37680 #LocalAI
10
Automation you run yourself: Hub's store has an Alternatives category to compare against rented tools. ci.computer/download
8
A study across 51 languages found English passed a confidence cutoff more often than other languages. Translating to English nearly erased that gap at the cutoff, but not at a stricter error target. arxiv.org/abs/2610.37688
Practicing grammar in chat with a small local model? A preprint reports near-perfect coverage of the target grammar with steering, and says compact models beat much larger models' top results. Abstract only. arxiv.org/abs/2610.39800 #LocalAI
4
Researchers at Skywalker Sound describe training PyTorch models across a fleet of Macs over standard 10Gb Ethernet, with 95.9% parallel efficiency across 24 nodes. Studio-scale gear, so check your own machines first. arxiv.org/abs/2610.39029
3
The box ships with Companion OS, plus Docker, Ollama and Hub. Core is optional for Hub users and ships to US addresses only. ci.computer/store/p/core
3
43
When an AI agent touches client work, who can show what it did? A new preprint proposes signed, tamper-evident records of agent sessions. It's a proposal, with no experiments yet. arxiv.org/abs/2610.40071
9
Hub's Alternatives category is the place to compare self-hosted options with rented chat tools. Local models want 16 GB of RAM or more. ci.computer/download
9
A preliminary study of 891 college students: revising AI output, instead of accepting it, was linked to critical thinking. It's correlational and not yet validated. arxiv.org/abs/2610.38910 #AIResearch
4

