oscarbotteri

oscarbotteri / osquius

Public

An open-source agent harness for the terminal — long-term memory, subagents, slash commands, and multi-provider model routing through the Vercel AI Gateway. Built to help you understand how agent harnesses actually work.

10
0
100% credibility
Found Jun 01, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Osquius is an open-source terminal-based AI coding assistant that serves as a learning tool to understand how agent harnesses work, featuring file editing, command execution, long-term memory, subagent orchestration, and multi-model support through a clean, modular architecture.

How It Works

1
🔍 Discovering a learning project

You find Osquius while exploring how AI coding assistants like Claude Code are actually built under the hood.

2
📦 Installing and building

You download the project and run a simple build command to prepare everything on your computer.

3
🔗 Connecting your AI service

You provide your AI service credentials so the assistant can think and reason about your code.

4
🚀 Launching your coding assistant

With everything ready, you start the agent and see a beautiful terminal interface appear with your current project loaded.

5
💬 Chatting about your codebase

You ask questions about your code, and the assistant reads files, searches for patterns, and explains what it finds.

6
Choosing how it works
💬
Ask-first mode

The assistant shows you what it plans to do and waits for your approval before making changes.

⚡
Automatic mode

The assistant works autonomously, making decisions and executing tasks without stopping.

📋
Planning mode

The assistant builds a step-by-step plan and follows it, keeping you informed along the way.

🔎
Research mode

The assistant reads and analyzes code but cannot make any changes or run commands.

7
🧠 Watching it remember

The assistant stores important facts about your project in long-term memory, so it learns your conventions over time.

✨ Understanding how it all works

You've explored a transparent, hackable agent harness and now understand the inner workings of AI coding tools.

Sign up to see the full architecture

6 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is osquius?

Osquius is a terminal-based AI coding agent written in TypeScript that gives you a transparent, hackable alternative to tools like Claude Code. It runs in your terminal, reads your codebase, edits files, executes shell commands, and can spawn specialized subagents for focused tasks. The project centers on long-term memory using SQLite with vector embeddings, so it remembers important facts across sessions. You interact via slash commands like `/mode`, `/model`, `/memory`, and `/debug`, and it routes requests through Vercel AI Gateway to support multiple model providers from a single endpoint.

Why is it gaining traction?

The hook here is transparency. Osquius is explicitly designed as a learning project -- the entire reasoning loop, tool layer, memory system, and UI live in plain TypeScript packages you can read and extend. It ships with four operating modes: `chat` for conversational assistance, `plan` for step-by-step approval workflows, `auto` for fully autonomous execution, and `research` for read-only code analysis. The skills system loads your project's `CLAUDE.md` out of the box, making it a zero-config drop-in if you're already using Claude Code. A real-time debug panel shows exactly which tools fire, token consumption, and error traces -- invaluable for understanding agent behavior.

Who should use this?

This is for developers who want to understand how AI coding agents actually work under the hood, or those who prefer a highly customizable, experimental harness over polished commercial tools. Researchers exploring agent architectures will find the modular design educational. If you want a self-hosted alternative to GitHub Copilot with full visibility into the decision-making process, this fits. It's less suitable for teams needing production-ready stability.

Verdict

At 10 stars and marked experimental, osquius is firmly in the "interesting prototype" category. The 1.0% credibility score reflects its early stage -- documentation is present but feature completeness varies, and you should expect breaking changes. That said, the architecture is sound, the codebase is readable, and the learning value is genuine. Fork it to learn, experiment with agent internals, or build your own harness on top of its components. Do not adopt this for critical production work yet.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.