freeswitch

One command. Every free AI model. Zero cost.

Tired of juggling API keys, rewriting boilerplate, and guessing which free model actually works? freeswitch gives you one CLI to switch between, chat with, and benchmark free AI models — no vendor lock-in, no credit card.

pip install fswitch-cli
fswitch use nemotron-ultra
fswitch chat "write a REST API in Python"

Features

• Free by default — ships with 8+ free-tier models ready to go
• One command to switch — fswitch use <alias> and you're done
• Interactive REPL — multi-turn conversations with /switch between models mid-chat
• Autonomous agent — fswitch agent "build a todo app" — reads/writes files, runs tests, and self-corrects!
• Unified Provider Registry — supports OpenRouter, Google, Mistral, GitHub, Groq, and Native Anthropic (Claude)
• Beautiful UI — smooth streaming with rich and visual ⠋ Thinking... animations
• Benchmark mode — race all free models against each other on the same prompt
• Local models — works with Ollama for fully offline, private AI
• Tiny footprint — pure Python, just requests + rich, nothing heavy

Install

pip install fswitch-cli

Or from source:

git clone https://github.com/Biya-dev/freeswitch
cd freeswitch
pip install .

Setup (30 seconds)

1. Get a free API key from your preferred provider:

   • OpenRouter (Best for variety & free tiers)
   • Anthropic (For native Claude 3.5 Sonnet)
   • Google AI Studio (Best for huge context limits)
   • Mistral AI (Best for Codestral)
   • GitHub Models (Built into GitHub, uses personal access tokens)
   • Groq Cloud (Best for speed)

2. Tell freeswitch about it:

fswitch config --key sk-or-v1-your-key-here
# For other providers, use the --provider flag:
# fswitch config --provider anthropic --key sk-ant-...
# fswitch config --provider google --key AIzaSy...
# fswitch config --provider groq --key gsk_...

That's it. You're ready.

Usage

List available models

fswitch list

Switch model

fswitch use nemotron-ultra

Chat (single prompt)

fswitch chat "explain quicksort in 3 lines"
fswitch chat -m deepseek-r1 "solve this leetcode problem: two sum"

Interactive REPL (multi-turn)

fswitch repl

Inside the REPL:

• Type naturally to chat
• /switch deepseek-r1 to change model mid-conversation
• /models to list available models
• exit to quit

Autonomous Agent

fswitch agent "create a Python script that fetches weather data"
fswitch agent -m qwen3-coder "refactor this project to use async"

The agent can:

• Read and write files on your machine
• Run shell commands (with your approval)
• Complete multi-step coding tasks autonomously
• Self-Correct: Pass the --test flag, and if the tests fail, the AI will automatically fix the code until they pass!

Example of autonomous self-correction:

fswitch agent "create a calculator function" --test "pytest tests/test_calc.py"

Every action requires your confirmation before executing.

Benchmark all free models

fswitch benchmark
fswitch benchmark --prompt "write a binary search in Rust"

Races every free model on the same prompt and shows you which is fastest.

Supported Models

Alias	Provider	Model	Free	Best for
nemotron-ultra	OpenRouter	NVIDIA Nemotron 3 Ultra 550B	Yes	Reasoning, orchestration
nemotron-super	OpenRouter	NVIDIA Nemotron 3 Super 120B	Yes	General & agentic workflows
qwen3-coder	OpenRouter	Qwen3 Coder	Yes	Code generation, repo-scale context
gemma-4	OpenRouter	Google Gemma 4 31B	Yes	General multimodal instruction
laguna	OpenRouter	Poolside Laguna M.1	Yes	Coding-agent experiments
laguna-xs	OpenRouter	Poolside Laguna XS 2.1	Yes	Fast coding-agent tasks
deepseek-r1	OpenRouter	DeepSeek R1	Yes	Reasoning + math
auto-free	OpenRouter	Auto-Free Router	Yes	Auto-routes to best free model
gemini-2.5-flash	Google	Gemini 2.5 Flash	Yes	Massive 1M context, 1500 req/day
gemini-2.5-pro	Google	Gemini 2.5 Pro	Yes	High intelligence, 50 req/day
codestral	Mistral	Codestral	Yes	Native model for code generation
github-gpt4o	GitHub	GPT-4o	Yes	GitHub integrated access
github-llama3.3	GitHub	Llama 3.3 70B Instruct	Yes	Extremely capable open weights
groq-llama3.1-8b	Groq	Llama 3.1 8B	Yes	Ultra-fast LPU inference
groq-llama3.3-70b	Groq	Llama 3.3 70B	Yes	Fast and extremely smart
claude-3.5-sonnet	Anthropic	Claude 3.5 Sonnet	No	Complex agentic tasks (Native Endpoint)
ollama-llama3	Ollama	Llama 3 (local)	Yes	Fully offline

Local Models with Ollama

Install Ollama, pull a model, and use it completely offline:

ollama pull llama3
fswitch use ollama-llama3
fswitch chat "hello local model!"

Environment Variables

You can also set your API key via environment variable instead of fswitch config:

export OPENROUTER_API_KEY="sk-or-v1-..."

Contributing

PRs welcome! Here's how to help:

• Add a model: edit freeswitch/models.py — just add a new dict entry
• Add a provider: add a new function in freeswitch/client.py
• Report issues: open an issue

Please open an issue first for bigger changes.

License

MIT — do whatever you want with it.

---

If this saved you time, drop a star on the repo!