Red Octopus AI (Octo) is a terminal chatbot in the JARVIS style: an animated red octopus, a typing effect, and a memory that survives restarts. It chats in English, and it learns what you teach it.
Two brains
- Offline brain — small talk, time and date, calculator, jokes, and everything you teach it with /teach. No network, no key, works anywhere.
- Full AI mode — open questions go to a real language model, with your name, facts and recent conversation as context.
5 free APIs, no Gemini
Full-AI mode tries five free providers in order until one answers. Four need your own free key; the fifth needs nothing at all:
- Groq — GROQ_API_KEY — openai/gpt-oss-120b — No card: 30 req/min
- OpenRouter — OPENROUTER_API_KEY — meta-llama/llama-3.3-70b-instruct:free — Free :free models
- Mistral — MISTRAL_API_KEY — mistral-small-latest — Free experiment tier
- NVIDIA NIM — NVIDIA_API_KEY — meta/llama-3.3-70b-instruct — No card: 40 req/min
- Pollinations — No key needed — gpt-oss-20b — Free anonymous lane
Pollinations means full-AI mode works out of the box with zero setup. If every provider fails, Octo falls back to the offline brain and keeps working.
Memory that grows
Octo remembers your name, facts (/remember), answers you teach (/teach), notes, todos, and checkpoints (/checkpoint / /rewind). It also ships with a 300-entry offline knowledge bank, with a goal of 1,000,000 entries — check progress any time with /brain.
Try it
git clone https://github.com/aungzayyanaing/red-octopus-ai.git
cd red-octopus-ai
python octo.py # Windows PowerShell
python3 octo.py # macOS / Linux
One-shot mode: python octo.py "what time is it?"
Repo: https://github.com/aungzayyanaing/red-octopus-ai
Stdlib-only Python — no packages to install. Teach it something, close it, open it tomorrow, and ask again. It remembers.
Top comments (0)