Self-hosted · Ollama · PHP + MySQL
Platypus is a monochrome web app for local LLMs: streaming chat, projects, agents, skills, memory, Telegram, and full-text search. Host the UI on your server; run models on your machine with a small bridge.
Open source friendly stack. No subscription to Platypus itself. You bring Ollama and hosting.
Product tour
Everything below ships in the repo today: real pages, real database tables, real bridge to Ollama.
Server-sent events stream tokens into the thread. Models that emit chain-of-thought get a resizable thinking column. Markdown rendering, code highlighting, regenerate, reply, stop generation, and per-message reactions.
On shared hosting the PHP app never runs Ollama. A CLI bridge on your PC polls for jobs,
streams from 127.0.0.1:11434, and posts tokens back. Sidebar shows bridge
status and model count. Setup wizard walks you through connection.
Projects group chats with instructions and accent colors. Agents add persona, model
defaults, and system prompts. Skills are reusable prompts you invoke with
@SkillName autocomplete in the composer.
Save facts from conversations into global, project, or agent memory pools. Toggle memory per user. The inference builder injects the right context into each job so the model sees what you chose to remember, not the entire internet.
Link a bot and chat from Telegram using the same bridge queue. Search runs full-text over titles and message bodies. The sidebar lists projects, agents, skills, memory, archived chats, and live bridge status.
Honest comparison
Cloud assistants are excellent products. Platypus targets a different job: you control the stack, the database, and the model runtime. Trade-offs are real on both sides.
| Platypus | ChatGPT | Claude | |
|---|---|---|---|
| Who hosts the UI | You (PHP on your server or MAMP) | OpenAI | Anthropic |
| Who runs the model | You (Ollama on your PC via bridge) | OpenAI data centers | Anthropic data centers |
| Where chats are stored | Your MySQL database | OpenAI account (cloud) | Anthropic account (cloud) |
| Model choice | Any Ollama model you pull | GPT-4o, o-series, etc. | Claude 3.x / 4 family |
| Monthly fee to us | None (hosting + electricity only) | Free tier + Plus / Pro / Team | Free tier + Pro / Max / Team |
| Setup effort | Higher: PHP, MySQL, Ollama, bridge | Low: sign up and chat | Low: sign up and chat |
| Reply quality ceiling | Depends on your local model + hardware | Frontier models, tuned products | Frontier models, strong long context |
| Reasoning / thinking UI | Yes, dedicated thinking column | Yes (model-dependent) | Yes (extended thinking on supported models) |
| Custom agents & projects | First-class in app | GPTs, projects (product features) | Projects, styles (product features) |
| Telegram / API for your bot | Built-in Telegram + your MySQL | Separate products / APIs | API available; not this UI |
| Offline / air-gapped | Possible if site + Ollama stay local | No | No |
| Polish & mobile apps | Web UI; you maintain it | Excellent apps and UX | Excellent apps and UX |
Choose ChatGPT or Claude when you want frontier models, minimal ops, and product polish. Choose Platypus when privacy, data residency, model freedom, and owning the full stack matter more than turnkey convenience.
Create an account, connect Ollama with the bridge, and keep every chat in your own database.
