New chatWeb

How can I help?

Combine with

Preview

Settings

General

Interface language
Response style

Applies to future answers. Evidence mode asks for sources and uncertainty; it does not verify facts or enable web search.

Connect your computer

Pair your own computer once. The model runs there; this website stores your chat and sends permitted context to that computer.

  1. Prepare your local model

    Start Ollama with an installed local model and its cloud features disabled. Skip downloading if you already have a model.

    Download Aid companion · Mac/Linux

    Unzip and keep the folder in a permanent location. On Mac, open Start Aid Companion.command; on Linux, run it with bash. First use installs locked dependencies with uv. Pairing still needs your confirmation.

    Download Ollama

    The companion supports Mac and Linux command-line setup. Windows connection support is not available yet.

  2. Create a pairing code

    Click below. Copy the code; it expires in five minutes and can be used once. Do not share it.

  3. Connect from your computer

    Open Terminal on the computer with your model, not the Vultr SSH terminal. You can open the downloaded Start Aid Companion.command on Mac. Alternatively, from the extracted Aid Companion folder, run:

    uv run python scripts/run_local_companion.py connect --confirm-local-only

    Paste the code when asked, then press Enter. Nothing appears while pasting; this is normal. A single installed model is selected automatically.

  4. Confirm this computer

    Click Refresh below. Match the device ID with your Terminal and click Confirm. Then return to Terminal and press Enter to start the connection.

  5. Use your local model

    When the computer says Connected, click Use this device. Close this guide and select Local model in the chat model picker. Keep the computer awake, online and Ollama running.

Optional: start automatically on Mac

After pairing, stop the foreground connection with Ctrl+C. Run this once from the same Aid folder; the connection will start at Mac login and reconnect after temporary outages. Terminal can then close.

uv run python scripts/run_local_companion.py enable-login --confirm-local-only

To disable automatic startup, run:

uv run python scripts/run_local_companion.py disable-login

This connection is for model answers only. Folder access is not available here.

Using Aid

Choose who answers

Select one model for one answer, or several for independent answers. Improve answer checks their gaps and disagreements and writes one response; the originals stay available underneath.

Balance and charges

Balance & payments shows dollars remaining, spending and purchase history. Credits appear only after confirmed payment. If a request fails with uncertain usage, its reserved amount stays pending until reconciled.

Use your computer’s model

Download the Mac/Linux companion in Local connection, then pair and confirm your computer. Keep it awake, online and Ollama running. Local-model replies still use hosted chat storage. Hosted folder access and Windows connection are unavailable.

Sign-in and recovery

Google manages Google passwords. Email/password accounts use the configured sign-in provider’s reset-email flow. Check Account for recovery and signed-in sessions; opening the session list refreshes it automatically.

Context, memory and feedback

Review saved memories and notes in Privacy & memory. Their API-sharing choices control whether they may enter API context. Turning off chat context does not delete saved messages. Answer feedback stays private; it does not enable shared training or automatic production changes.

If an action fails, use the retry offered beside it. Never paste a password, API key or payment information into chat to troubleshoot.

Activity & feedback

Private to your workspace. Viewing this report makes no model calls.

Completed calls are not completed tasks. Ratings reflect usefulness, not factual accuracy. Automatic training and promotion are off.

Chat options

Project

Context preview

Selected memory and recent conversation. The server may trim history to its input limits. Web search sends only the current question.

Notes & search

Saved in your private server workspace. Searchable; API sharing is off by default.

Memory

Save a source note / 保存来源笔记

Record a source you personally checked. Your note is saved with this comparison; it does not change the answer or run an AI check. / 记录您亲自查阅的来源。笔记保存在本次比较中，不会修改回答或启动 AI 核验。