OllieChat free & kind · by WeirdMM
Settings
Privacy
Discord
Canvas

HTML Ollie writes will render here, live.

Settings

Appearance, media generation, and read-aloud voice.

Voice tuning
Speaker boostBoost similarity to the original voice (slightly slower).
Auto-read repliesRead each of Ollie's replies aloud automatically.
Read thinking aloudInclude the model's reasoning when reading a reply aloud.

Roleplays

Pick a character to start, or create your own.

Create roleplay

System prompt

Ollie's standing instructions. Applies to this chat and any new ones.

Tools & search

Ollie can call tools mid-answer. Web is built in (DuckDuckGo). Custom tools run as JavaScript in your browser — write a function body that reads args and returns a value.

About OllieChat

Hey — I'm WeirdMM, and OllieChat is my little gift to you. 💛

I put the good models behind one simple chat and made it free for everyone — no sign-up, no paywall, no catch. Just powerful AI you can actually use, shared out of kindness. Use it, enjoy it, build with it.

  • Models Eight, from quick to flagship — Fable, Sonnet, Opus, GLM, Kimi and more.
  • Think Watch Ollie reason before it answers — dial it Low→Max in the composer.
  • Canvas HTML pages render live on the right as Ollie writes them — preview & edit in place.
  • Web Live DuckDuckGo search, on demand.
  • Tools Teach Ollie your own functions.
  • Yours Chats live only in your browser. Nothing's tracked.
Files & images: Attach text/PDF docs and Ollie reads their contents as text. Attach text/PDF docs and Ollie reads their contents as text.
💜 Free forever — but if OllieChat helps you, tip the tree:
ltc1qs0vphf4fz2ya5wd84zqxs88kslclg95lh0sz58
Litecoin (LTC)

Made with love by WeirdMM. Logo by biggest_birdie (Discord). · Privacy

✨ Introducing Multi-Agent

Put more than one model on the job — together.

Turn on Multi-agent in the model menu and pick 2–4 models. They collaborate on every message: the lead drafts an answer, the others review and correct it, and the lead writes the final answer — so you get the sharpest result, checked by more than one mind.

  • Lead Your first pick drafts and finalizes.
  • Team The rest catch mistakes and add ideas.
  • Example Add Opus 4.8 + Fable 5 and watch them work it out.

Install to iOS

Add OllieChat to your Home Screen so it opens like an app. This only works in Safari on iPhone or iPad (Chrome/Firefox on iOS can’t add home icons).

  1. 1
    Open this site in Safari If you’re in another browser, copy the URL and paste it into Safari.
  2. 2
    Tap the Share button The square with an arrow pointing up — usually at the bottom of Safari (or top on iPad).
  3. 3
    Scroll and tap “Add to Home Screen” You may need to swipe the bottom row of actions to find it. Tap Add to confirm.
  4. 4
    Launch from your Home Screen Look for the OllieChat icon. It opens full-screen without Safari’s address bar.

⌘ OllieCLI Beta

OllieChat, now in your terminal.

Chat with the good models straight from the command line — the same free, keyless access as the site, no sign-up. Install it globally with npm:

npm install -g @weirdmm/olliecli@latest
  • Free Same models as the site — no key, no account.
  • Fast Ask a question without leaving your terminal.
  • Beta Fresh out the oven — expect rough edges; feedback welcome on Discord.
View on npm

Benchmark

Pick the models to compare, then run the benchmark. Every model answers the same set of freshly-generated, hard problems — graded on the spot against machine-computed answers. The best score in each row is highlighted.

Methodology: each category generates fresh, deliberately hard problems every run (large multiplication, modular exponentiation, string reversal, long-string letter counting, sequences, divisor counting) with exact answers computed in your browser — so the scores are real and un-gameable. Reasoning is off and the difficulty is tuned to the edge where strong models start to slip. Frontier models still cluster high on clean computational tasks; the clearest gaps show between a fast/small model and a flagship. Small samples, so expect a few points of run-to-run variance. “—” means every problem in that category errored.