01 · CONNECT
Add providers
Hosted APIs or local runtimes. Models are discovered dynamically per provider.
Multi-provider chat
One chat for every model.
A multi-provider chat application with automatic fallback, real-time streaming, prompt templates and token analytics. Switch providers without switching tools.
MIT13+ providersSSE streamingPostgreSQL
$ git clone https://github.com/iwebbo/LLM.io
$ cd LLM.io && docker compose up -d
# add a provider key, pick a model, chat
$ _
How it works
Connect once, chat anywhere.
02 · CHAT
Streamed, with fallback
Responses arrive over SSE with auto-reconnect, and fall back to another provider when one fails.
03 · TUNE
Templates and temperature
Per-conversation temperature, reusable prompt templates with placeholders, and a deep reasoning mode.
04 · REVIEW
History and analytics
Full-text search over stored conversations, token totals by day, provider and model, and export.
Rate limiting and multi-origin CORS are built in, so the same backend serves web, mobile and desktop clients.
Features
Built for people who use many models.
PROVIDERS
13+ providers, automatic fallback
Switch between hosted and local models without changing your workflow.
STREAMING
SSE with auto-reconnect
Real-time responses that survive a flaky connection.
TEMPLATES
Prompt templates
Create, share and reuse prompts with {{placeholders}}.
ANALYTICS
Token analytics
A dashboard with daily, provider and model totals.
HISTORY
Searchable conversations
PostgreSQL storage with full-text search, exportable as Markdown, JSON or HTML.
REASONING
Reasoning-model aware
Chain-of-thought auto-detection and a deep reasoning mode with a high max_tokens budget.
Every capabilityexpand
Current
- Multi-provider — automatic fallback
- SSE streaming — auto-reconnect
- Temperature — 0.0–2.0 per conversation
- Prompt templates — create, share, placeholders
- Token analytics — daily, provider and model totals
- History — PostgreSQL with full-text search
- Export — Markdown, JSON, HTML
Advanced
- Chain of thought — auto-detection for reasoning models
- Deep reasoning mode — high max_tokens for long reasoning
- Model auto-discovery — fetched per provider
- Health checks — connection tests with latency
- Rate limiting — 60 req/min by default, configurable
- Multi-origin CORS — web, mobile and desktop at once
Providers
- OpenAI
- Anthropic Claude
- Google Gemini
- OpenRouter — 200+ models
- xAI Grok
- Groq
- Hugging Face
- Ollama · LM Studio
- vLLM · LMDeploy
- Oobabooga
Built withPythonReactPostgreSQLDockerLinux
Star or fork on GitHub ↗