¡PewDiePie acaba de Reventar la IA! Lanzó Odysseus, IA GRATIS y SIN LÍMITES
The video introduces Odysseus (also called Obsidius), an open-source AI workspace platform launched by PewDiePie (Felix), which runs locally on your device with full privacy and no telemetry. The presenter walks through installation, connecting AI models both locally and via APIs, and demonstrates seven use cases including agents, deep research, task automation, image editing, and advanced customization. The platform is positioned as a free, unlimited, privacy-first alternative to Big Tech AI tools.
Summary
The video opens by framing PewDiePie (Felix, the Swedish YouTuber with over 100 million followers) as an unlikely AI innovator who has launched a platform called Odysseus (interchangeably referred to as Obsidius throughout the transcript). The presenter claims to have used it for three days since launch and describes it as the most complete open-source AI platform he has ever seen.
The installation process is covered in detail: the platform is installed via a single terminal command, with Docker being the recommended method for most users, though Apple Silicon users are advised to run it natively due to GPU compatibility issues with Docker. Once installed, users create a simple username and password account to access the interface.
A major feature highlighted is the 'cookbook' tool, which automatically scans the user's hardware and recommends compatible local AI models — eliminating the guesswork of whether a model will run on a given device. Models can also be connected via APIs from providers like OpenRouter, Groq, Anthropic, OpenAI, and Gemini. OpenRouter is specifically noted for offering over 300 models, many of which are free. The platform also supports reusing models already downloaded through tools like LM Studio or Ollama by connecting via local endpoint URLs.
The chat and agent capabilities are demonstrated, including the ability for the locally-running agent to access the user's file system (e.g., listing files in a downloads folder), execute terminal commands, preview generated HTML applications in real time, and compare multiple AI models side by side in a blind-test format. Users can also set custom prompt prefixes/suffixes, adjust token creativity, and create 'personas' — custom conversational agents similar to custom GPTs — which can even be grouped to interact with each other sequentially or in parallel.
The deep research feature is shown using the Tavily search API (which offers 1,000 free credits), allowing the platform to autonomously plan, branch, and execute multi-source research queries, generating formatted reports with AI-generated images in roughly 13 seconds for the demo query.
The 'Brain' section covers AI memory and skills management. The platform persistently stores memory and skills locally, and unlike major tech companies, only the user can access this memory. Users can manually add entries (e.g., 'Remember my name is Aleji') and the information persists across all connected models regardless of provider.
Task automation is demonstrated through a scheduling feature where users can configure recurring prompts (e.g., a daily AI news digest at a set time), with outputs delivered as in-app sessions, notifications, or emails. A notes and calendar system is also included for organizing reminders for both the user and AI agents.
The gallery and image editing section shows a lightweight Adobe Photoshop-like tool with AI-powered features including cropping, image expansion, inpainting, background removal (via installable add-ons), and erasing — though the presenter acknowledges the creator admits this feature is still maturing.
Finally, advanced settings cover full UI theme customization (including animated backgrounds like constellations, rain, and moving squares), incognito chat mode, email server integration (allowing the AI to read and auto-reply to emails while keeping everything local), MCP integrations for third-party tool connections, keyboard shortcuts, and a multi-user access system. The platform also works on Android and iPhone.
Key Insights
- The 'cookbook' feature automatically scans the user's hardware and recommends only the AI models that will run successfully on that device, eliminating the common problem of downloading models that turn out to be incompatible with one's hardware.
- The platform's memory system stores all user and agent data locally and exclusively accessible to the user — unlike major tech companies — and persists across all connected AI models regardless of whether they are local or API-based.
- The presenter found that both default free search engines were failing during the deep research feature and resolved this by integrating Tavily's API, which provides 1,000 free credits, allowing the platform to perform multi-source research in approximately 13 seconds.
- The persona and group chat system allows users to create custom AI agents similar to custom GPTs and configure them to interact with each other sequentially or in parallel — effectively simulating a multi-agent team within a single workspace.
- The creator of Odysseus himself acknowledges that the image editing tool has not yet reached the level of Adobe Photoshop, though it is actively being developed to compete with it using AI-powered features like inpainting and background removal.
Topics
Full transcript available for MurmurCast members
Sign Up to Access