EleosGuides, comparisons and research
Every page here is written from something we can show — a filmed run, a measurement, or another product's own documentation with the date it was checked.
Start here
- The AI plans you already pay for, in one app on your PC — Claude, ChatGPT, Grok, Cursor or a model on your own PC — a different one per chat.
- Write the instructions your AI runs under — Replace the instructions the model runs under, not just add a note on top.
- Local AI on Windows, offline and on your own machine — An open model on your own PC: offline, no AI subscription, and what a small one can do.
Guides
- When an AI "uses your computer", this is the loop — Screenshot, decision, click — and every way that loop goes wrong.
- "Sort my Desktop into folders." What actually happened — One sentence at a messy Desktop: 15 files moved, and the two files it deliberately left alone.
- Where Claude Code keeps your chats — Your sessions are plain-text files on your own disk, deleted after 30 days by default.
- Claude Code on Windows, and the errors you will hit — Install routes, the five errors people actually hit, and where your sessions are stored.
- Scheduled AI tasks on Windows — What decides whether a scheduled job actually runs: sleep, overlap, and where the work happens.
Compared with other tools
- Four ways to reach your PC's AI from your phone — SSH, Anthropic's remote control, a tunnel of your own, or an app relay — what each costs you.
- A Claude Desktop alternative for Windows — What Anthropic ships on Windows today, from their own docs, and where an independent app differs.
- LM Studio alternatives on Windows, compared — Ten local AI apps on Windows compared against their own licences and docs.
Research
- Ten dead ends between download and first answer — Ten things that stopped a first-time user on clean Windows installs, and what each fix was.
- What actually fits in 8 GB of VRAM — Working sets, the agent context floor, and which small models can call a tool at all.
- What a system prompt is worth, measured — 18 trapped tasks, three prompt layers, two models, ranked blind.
How these pages are written
- Facts about other products come from their own documentation, linked, with the date checked, because this field changes monthly.
- Our own numbers come with the method and the sample, and a section saying what they cannot tell you.
- Limits of our own product are stated on the page that would otherwise oversell it.
- Pages are re-read and re-dated rather than left to rot.
If something here is out of date or wrong, tell us: hello@eleosagent.com.