ETOS LLM Studio Docs

English

简体中文

English

简体中文

Theme

Hidden Gems ​

For people who think "I know this app does a lot, but a lot of stuff just isn't going to discover itself."

The most important one up front: in ETOS, "hidden tricks" often hide in gestures, not in deep menus.

Gestures ​

1. Always Try "Long-press + Swipe" ​

When you see a row, card, message, list item, or record — don't just tap. The reliable exploration order is:

  1. Tap once to see the normal flow
  2. Long-press to check the context menu
  3. Swipe left or right to look for quick actions

This rule is near-universal in ETOS, particularly in:

  • Session list (long-press → Move / Rename / Delete)
  • Providers & models list
  • Memory and session summaries (swipe left to delete, swipe right to archive)
  • Worldbook entries
  • Daily Pulse feedback history
  • File and log lists
  • Chat messages (long-press is the biggest Easter egg surface)

If you ever think "surely this thing has more than one tap action" — long-press, then swipe.

2. Long-press in Chat = Hidden Feature Treasure Chest ​

Each message's long-press menu has 8–10 actions, including:

  • Create Prompt Branch (prompt only, or copy message history too)
  • Export Up to This Message (with context) — truncate and export
  • Token Info + Thinking Duration — metadata for reasoning models
  • Edit Message (yes, including the AI's reply)
  • Quote — quote this message in your next input

Details in Start Your First Chat.

3. The Session List Search Field Is Full-Text, Not Title-Only ​

The search field at the top of the session list (placeholder "Search session titles or messages") does full-text content search, not just titles.

Tapping a result jumps to that specific historical message — way faster than scrolling.

Configuration ​

4. Get One Model Working End-to-End Before Splitting Roles ​

ETOS lets you pick separate models for chat / TTS / Daily Pulse / STT / embedding. But first time around, get one stable model working end-to-end, then split. Saves 90% of the debugging time.

5. More Tools ≠ Better ​

First-timers see MCP / local tools / Shortcuts / Skills and enable all of them at once. Result:

  • Half the model's context is eaten by tool descriptions
  • When something breaks, you can't tell which tool is to blame
  • Approval bubbles keep popping up

A more practical approach:

  • Daily chat sessions: only essential tools
  • Workflow sessions: enable a dedicated tool set
  • Decide approval policy before allowing auto-execution

6. Memory vs Worldbook — Simple Decision Rule ​

This piece of info …Goes to
Gets used repeatedly, like a knowledge blockMemory
Only useful when a specific keyword / scene appearsWorldbook

Maintenance cost drops a lot.

7. Multi-Key with English Commas ​

The Provider's API Key field supports rotation:

sk-aaaaaaaaaa,sk-bbbbbbbbbb,sk-cccccccccc

Each request rotates; if one key trips 429, the next is used. Free / trial keys stitched together last a while.

8. {api_key} Is a Placeholder, Not a Literal ​

Both Provider's "Header Overrides" and MCP's header overrides support placeholders like {api_key} / {token}. They get substituted with the current key:

Authorization=Bearer {api_key}
X-Custom-Auth=xxx-{api_key}-yyy

Use for services with non-standard auth headers.

Usage Habits ​

9. Daily Pulse's Real Key Is "Feedback" ​

If you just look at cards without giving feedback, it evolves slowly.

The four buttons — Like / Downvote / Hide / Save as Conversation — improve quality more than model-hopping. Save as Conversation is the strongest positive signal.

10. "Tomorrow's Curation" Is For Next Time, Not This Time ​

Daily Pulse has two text inputs:

  • Current Focus — long-term direction, applied every generation
  • What I Want Tomorrow — applies only to the next generation

Things you think of at night for "tomorrow morning's view" go into "Tomorrow."

11. Debugging Tools Save UI Hopping ​

When you suspect a problem is about files, config, or requests, jump to Settings → Extended Features → App Logs / Advanced Diagnostics / Feedback Helper.

They let you see data directly instead of guessing inside the UI.

12. Import/Export Is Migration, Not Just Backup ​

Beyond backup, ETOS bundles work for:

  • Switching devices
  • Cross-account / cross-environment migration
  • Save a stable version before big changes — instant rollback if anything goes wrong

Apple Watch ​

13. Let the iPhone Prep for the Watch ​

The comfortable two-device flow:

  1. iPhone: configure models, tools, display preferences
  2. Sync: bring the environment to the Watch
  3. Watch: focus on chat, alerts, quick follow-up

Don't try to do complex setup on the Watch — screen too small, efficiency terrible.

Final Recommendations ​

14. Build Your Own "Usage Layers" ​

A simple layering helps:

Use caseConfig
Daily chatMost stable model + minimum tools + auto-approve memory write
Specialized tinkeringAdvanced params + all tools + MCP / Skills when needed
Fragments of time (Watch)Receive reminders + short-turn chat + no config

No matter how many features pile up, you don't end up with a soup.

15. When You Can't Find Something, Ask "Is This Governance or Execution?" ​

ETOS's main screen has only "Chat + Settings" tabs:

  • Execution (chatting, attachments, model switching) → Chat
  • Governance (providers, tools, memory, worldbook, sync, Daily Pulse config) → Settings

Following this rule, hit rate is 90%+.

Next ​