AiMedley
About AiMedley

A workbench for the other AI minds.

AiMedley is one home for DeepSeek, Qwen, Kimi, and Llama — four open-source-adjacent models with very different personalities. Chat with one, compare all four, research with the wisdom of crowds, build agents, analyze documents.

DeepSeekQwenKimiLlama
Why we built it

Four reasons AiMedley exists.

Open models deserve a stage

Closed labs each ship their own chat app. Open-source-adjacent models — DeepSeek, Qwen, Kimi, Llama — never had a polished home where you could chat, compare, and build with all of them. Until now.

Comparison is the new search

The right model depends on the prompt. AiMedley turns picking the right mind into a one-click side-by-side experiment — streaming live across all 4 columns at once.

Playful beats serious

Most AI apps look like spreadsheets. AiMedley feels like a toy you actually want to touch — neo-brutalist clay, color-coded personalities, and starter prompts tuned to each model's vibe.

Your data stays yours

Conversations, agents, and documents are scoped to your account. Your OpenRouter key is yours. We don't train on your prompts. We don't sell your data. We don't have ads.

The cast

Meet the four minds.

DeepSeek
Analytical & Deep

Math, code, and step-by-step reasoning. The one you want when you need the answer to actually be right.

Qwen
Bold & Quick

Fast, direct, and a little spicy. Great for brainstorming, naming, copy, and rapid back-and-forth.

Kimi
Creative & Fresh

Long context and lyrical prose. Best for storytelling, ideation, and content with character.

Llama
Friendly & Open

Warm, practical, and helpful. The friendly generalist — the one you'd ask for dinner ideas.

Inside the workbench

Six tools. One key. Zero lock-in.

Streaming chat

Typewriter-fast SSE token streaming, per-model temperature and length controls.

Side-by-side compare

Send one prompt → watch all 4 models type their answers live in parallel columns.

Custom agents

Save reusable personas with system prompts + default models. Pick from starter templates or build from scratch.

Research synthesis

Multi-model parallel briefing + a final synthesized digest. Optional web search with URL citations.

Document Q&A

Upload PDF / text / code (≤5MB) and ask any of the 4 models grounded questions.

Web search inline

Flip a toggle and any model gets fresh web results (via OpenRouter's Exa plugin) with citations.

Under the hood

How it works.

1
One key, four families.

AiMedley routes every request through OpenRouter — a single API that unifies DeepSeek, Qwen, Kimi (Moonshot) and Llama. You pay only for what you use, at near-cost.

2
Real streaming, including for compare.

Single chats and the multi-model compare view both use Server-Sent Events. For compare we fan-out 4 streams server-side and tag every token with its model so all 4 columns type live.

3
Auto-titled conversations.

After you send your first message we fire a cheap background call to summarize the chat into a 3-5 word title. No more sidebars full of 'Hey, how do I…'.

4
Web search built in.

Flip the green Globe toggle on Chat / Compare / Research and any model gets fresh web results via OpenRouter's Exa plugin, complete with citation URLs.

5
Per-model controls.

Temperature and max-tokens are per-model and saved in your browser. Want Qwen wild and DeepSeek precise? Two clicks, persists forever.

Stack

Built with boring, dependable pieces.

FastAPI
Async Python backend
MongoDB
Conversations & state
OpenRouter
All 4 model families
React 19
Frontend
Tailwind
Styling
shadcn/ui
Primitives
JWT + bcrypt
Auth
Exa
Web search

Ready to mix?

Spin up the workbench in 10 seconds. No card. No setup. Bring your own OpenRouter key (or borrow ours in demo mode).

Open the workbench