PosterChan AI - Self-hosted AI Powerhouse

Meet Poster-chan AI: One Self-Hosted Box, Your Entire AI Stack
  
 If you've ever wanted the convenience of a modern AI assistant without handing your conversations, keys, and data to someone else's cloud,
 Poster-chan AI is built for exactly that. It's a single self-hosted FastAPI application that bundles a surprising amount of capability behind one
 chat box — and it runs on your hardware, on your rules.

 Bring your own model

 At its core, Poster-chan AI is a streaming chat assistant with an OpenAI-compatible /v1/ API. Point any compatible client at
 your-box:3051/v1 and you've got a private, function-calling-capable model server. It works with cloud models (any OpenAI-compatible API)
 or fully local ones via Ollama or llama.cpp — across CPU, NVIDIA CUDA, AMD HIP/ROCm, and even Intel Arc SYCL. You can round-robin load-balance
 across several backends, and agentic coding tools like opencode can drive your local models directly.

 Far more than chat

 The chat box is really a command center. Poster-chan AI can:

 - 🎨 Generate images (ComfyUI or native SDXL diffusers), music (text-to-song via ACE-Step), and video (in-process diffusers — Wan2.1 / LTX /
 CogVideoX)
 - 🗣 Speak and listen with TTS and STT
 - 🧠 Know your code — RAG over your git repos and folders, code-aware chunking, exposed to your IDE over MCP
 - 📸 Take full-page website screenshots, summarize and download YouTube/X videos from their transcripts
 - 🗂 Handle email, news, finance/budgeting, to-dos, torrents, and per-user file storage
 - 🛠 Run media tools — compress, clip, and convert files
 - 🎴 Build interactive flashcard quizzes from a PDF, slide deck, or Word doc

 It's also a bot platform

 This is where Poster-chan AI really stretches. From a single admin tab you can run autonomous bots on Telegram, Matrix, Misskey, and  
 Pleroma/Mastodon — reply bots, nitter relays, welcome/report/hashtag daemons, and more. You can post to social platforms from any chat, relay
 your fediverse notifications to Telegram or Matrix DMs, and even mirror a whole fediverse timeline into a Matrix room with threaded
 conversations, avatars, and inline media — where members can react, boost, and reply each under their own linked account.

 There's also remote node management: an agentic tool-calling mode that can run OS commands across your SSH-reachable fleet and stream each step
 live.

 Self-hosted, honest, hackable

 Everything is SQLite-backed, single-admin and multi-user, with no telemetry. Setup is turnkey — a single Docker image builds for CPU, CUDA, ROCm,
 or Intel Arc, or you can use the interactive installer on bare metal. The codebase keeps things clean (thin routers, logic in services), and
 it's GPLv3 free software.
  
 ▎ Point any OpenAI-compatible tool at it and you've got a private model server. Open the web UI and you've got a full assistant. Link a bot and  
 ▎ it's in your pocket.

 Get it

 The project is on GitHub: https://github.com/loblawbob873-svg/posterchanai

 Clone it, run ./install.sh (or docker build), open localhost:3051, and log in with admin / admin. Your own ridiculously capable AI
 assistant — private by default.