Codebase: GTA Sims Game
Mantic Think<br>Your models. Your key.<br>Nothing stored.
Keep me signed in on this device<br>Connect
Bring your own key — sent directly to Ollama through our proxy, never saved on our servers. Checked: stored in this browser until you sign out. Unchecked: kept only for this tab and cleared when you close it.
Get an Ollama API key →
or
Use local Ollama
Chats with the Ollama running on your machine — requests never touch our servers. Start it with CORS enabled:<br>OLLAMA_ORIGINS=https://manticthink.com ollama serve<br>Needs Chrome, Edge, or Firefox — Safari blocks localhost from https sites.
Streaming replies
Charts & diagrams
Runs code in-browser
GitHub context
iOS beta on TestFlight
Questions, answered
What is SƎER?
A free, private web UI for Ollama models. Chat with Ollama Cloud models using your own API key, or connect to the Ollama running on your own machine — with streaming replies, charts, in-browser code execution, and GitHub context.
Do I need an account?<br>No account, no signup. Paste your own Ollama API key — or use local mode with the Ollama on your computer, which needs no key at all.
Where are my conversations stored?<br>Only in your browser. Our server never stores your messages or your key — and in local mode, requests never reach our servers at all. The one exception: if you tap Share on an AI Debate, that debate is saved on our servers so the short link works for others — only what you explicitly share. See the privacy policy.
Can I chat with models on my own computer?<br>Yes — choose “Use local Ollama” above and your browser connects straight to your local instance. A zero-install web UI for local LLMs.
Which models can I use?<br>Any model your Ollama key or local instance offers — gpt-oss, Qwen, DeepSeek, Llama, and more. You curate your own list in the model picker.
Created by Cole<br>Support<br>Privacy
+ New chat
+ New project
+ New codebase
Sign out<br>Created by Cole
Download the iOS beta →
Two AI coders built this on Mantic Think — free, private, bring-your-own-key. Build your own codebase by chat.
Build your own<br>Import a copy<br>Download ZIP
‹ Codebases
Stop
Export ▾
Download ZIP<br>Copy active file<br>Share link<br>Push to GitHub
Files+
No file openCopy
New codebase
Name
Builder model — writes & edits files
Reviewer model — critiques each change
Enable reviewer (two-model build)
Builder ⇄ Reviewer rounds
Files live in your browser. Export to ZIP, a share link, or push to GitHub anytime.
Create codebase<br>Delete<br>Cancel
Push to GitHub
GitHub token — needs Contents: write ; creating a repo needs the classic repo scope
Connect
Repository
Create as a new private repo if it doesn't exist
Branch
Commit message
Force push (overwrite remote on conflict)
Push<br>Disconnect<br>Cancel
Models
Your presets
+ New preset
Your models
Add by name
Add
Ollama catalog
New preset
Basics
Name
Base model
Personality / system prompt
Generation optional — blank uses defaults
Temperature (0–2)
Top P (0–1)
Top K (0–200)
Max tokens
Save preset<br>Cancel
Generation settings
System prompt
Sent as the first system message on every request.
Sampling
Advanced
Max tokens
Seed
Stop sequences
Rendering
Let models draw charts & diagrams<br>Adds a hint so models can emit mermaid diagrams and chart blocks, rendered inline. Markdown tables always render.
Tools function calling
Let the model call tools
Connect
Only tool-capable models (e.g. gpt-oss, qwen3-coder) use these. Built-in tools run code, do math, and fetch URLs in a sandbox. You can also connect your own remote MCP servers — they're reached directly from your browser (the server must allow cross-origin requests from manticthink.com; tokens stay in this browser).
Reset to defaults<br>Manage models →
Reasoning scaffolds
+ New
Your scaffolds
Start from a template
New scaffold
Basics
Name
Summary
Role
Perspective
Tone
Reasoning steps
One step per line · up to 12 · 220 characters each.
Output guidance optional
Helpful elements to cover
Avoid by default
Safety hints optional
Disclaimers
Prohibited actions
Compiled preview
Save scaffold<br>Cancel
Saved code
GitHub
Connect a token
Connect<br>Stored only in this browser, sent directly to GitHub — never to our servers. Create a<br>fine-grained token with read-only Contents on the repos you want.
Connected as Disconnect
Repository
Open
0 selected<br>Attach to chat
New project
Name
Instructions
Injected as context on every message in this project's chats.
Context files
+ Upload files
or paste text instead
Add pasted file
Text/code files only · keep them small (stored in your browser). Large files increase prompt size.
Save project<br>Delete<br>Cancel
AI Debate
vs
Debate (for vs against)<br>Discussion (two perspectives)<br>Round table (2–4 models)
1 round each<br>2 rounds each<br>3 rounds each<br>4 rounds each<br>5 rounds each<br>6 rounds each
Synthesis<br>Perspectives<br>Start debate<br>Stop<br>★ Save<br>Share<br>Saved
Two AI models debated this on Mantic Think — free, private,...