Odysseus is a fully local personal AI workspace from PewDiePie https://github.com/pewdiepie-archdaemon/odysseus

Since there are many agent related projects nowadays so it can be confusing how this fits into the ecosystem, so the simplest way to explain is by comparing with existing projects.

Odysseus vs. OpenClaw + Hermes

Three self-hosted, local-first, MIT-licensed AI tools — but two different species.

• Odysseus is an interface you sit down at.
• OpenClaw and Hermes are an agent that runs in the background and reaches you. (OpenClaw and Hermes share the same architecture, so they're treated as one category here.)

Odysseus is synchronous and you-driven -- a private Codex/Claude UI with a suite of apps. Hermes/OpenClaw are asynchronous and self-directed -- a persistent daemon that runs on a schedule, acts on its own, and pings you on Telegram/Discord/etc.

	Odysseus	OpenClaw + Hermes
What it is	Self-hosted Codex/Claude-style web workspace	Persistent local agent daemon, reached via CLI + chat apps
Mental model	A dashboard you open and work in	A background worker that runs continuously and pings you
Interaction	Synchronous, you-initiated — open a tab, work a session	Async-capable — a built-in cron scheduler runs while you sleep, then messages you
Primary surface	Browser UI (+ PWA on mobile)	CLI/TUI + 16–24+ messaging platforms (Telegram, Discord, WhatsApp, iMessage…)
Where breadth comes from	Bundled apps — chat, doc editor, email triage, calendar, notes/tasks, deep research, blind model compare, hardware-aware model installer	Reach + autonomy — channels, self-written skills, MCP tools; apps are emergent, not bundled
Architecture	Multi-service stack (FastAPI + ChromaDB + SearXNG + ntfy), Docker-first	A single long-lived process that is the entire system
Agent core	Wraps opencode — the agent is one component	The agent loop is the product
Memory	ChromaDB vector store + SKILL.md	File/markdown memory + skills auto-written from experience ("learns you over time")
Manages the model backend?	Yes — Cookbook scans hardware, recommends, downloads, and serves via vLLM/llama.cpp	No — point them at an endpoint; pure consumers
Maturity	New, v1.0, ~12.6k stars	Larger, more established (OpenClaw by Peter Steinberger; Hermes by Nous Research)

Odysseus vs. Codex + Claude Desktop

Basically Odysseus competes with Codex and Claude Desktop, NOT OpenClaw and Hermes.

The main difference is it works completely locally, using local AI models AND local search engine via SearXNG (This launcher automatically installs and launches SearXNG 1-click launcher during installation as well)

DeepResearch

One of the most impressive features of Odysseus is DeepResearch. It creates a beautiful research WITHOUT relying on a single 3rd party API:

• No OpenAI/Claude type LLM API usage. Uses local LLMs
• No 3rd party search engine APIs. Uses self-hosted SearXNG (included in this launcher)

And the result is a beautiful report:

I recommend everyone to try it out:

1. Download a strong local model via Ollama/LMStudio/etc. such as GPT-OSS or the latest Qwen
2. Run DeepResearch

Everything takes place with ZERO reliance on ANY API service. We all know this has been possible but once you actually experience it, it feels special.

Also posted on https://x.com/cocktailpeanut/status/2061954378718642686?s=20