quarqlabs

quarqlabs / agent-oss

Public

A recursive evidence-gated cognitive runtime for memory-native AI agents, combining hybrid retrieval, temporal reasoning, async learning, and plug-and-play tools.

60
4
89% credibility
Found May 27, 2026 at 60 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Quarq Agent is a memory-first AI assistant built by QuarqLabs that distinguishes itself by actually remembering everything about you over time. Unlike typical chatbots that treat every conversation as isolated, Quarq builds persistent memories across three categories: facts about you (your name, job, preferences), events and conversations you've had, and your preferred communication style. The system stores everything locally on your computer using a local vector database, making your data private and accessible. It includes optional integrations with email, calendar, and PDF generation tools, and comes with built-in benchmarks showing high accuracy on long-term memory recall tasks. The project is open-source under the Apache 2.0 license and is designed for anyone who wants an AI assistant that truly understands them over time.

How It Works

1
🔍 Discover Quarq Agent

You hear about an AI assistant that actually remembers everything about you, unlike regular chatbots that forget after each chat.

2
⚙️ Set up your assistant

You create a simple configuration file with your API connection and give your assistant a name. Everything runs locally on your computer.

3
💬 Start chatting naturally

You talk to your assistant just like you would to a helpful colleague. You ask questions, share information, and discuss topics.

4
🧠 Your assistant learns about you

Behind the scenes, your assistant quietly stores what you tell it—your name, preferences, past conversations, and how you like to be addressed.

5
🔄 Your assistant gets smarter over time

The more you chat, the better your assistant understands you. It builds a mental model of who you are and what matters to you.

6
Optional: Connect your tools
📧
Email integration

Your assistant can search your inbox, read emails, and send messages on your behalf

📅
Calendar integration

Your assistant can check your schedule and create events for you

📄
PDF creation

Your assistant can generate PDF documents and reports for you

✨ A memory that lasts

You have an assistant that genuinely knows you—your history, your preferences, your projects. It answers questions about your past accurately and helps you without needing to re-explain yourself.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 60 to 60 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 agent-oss?

Quarq Agent is a memory-native AI agent runtime built in Python that combines vector search, temporal reasoning, and async learning into a single cognitive loop. Unlike typical RAG systems that treat memory as simple vector retrieval, it separates facts, events, and behavioral rules into distinct layers with their own retrieval strategies. The system uses hybrid search combining embeddings with keyword matching, applies query expansion techniques, and can fall back to secondary retrieval passes when initial results are insufficient. It ships with built-in tools for email, calendar, and PDF generation, plus a FastAPI server and terminal chat interface.

Why is it gaining traction?

The project targets a real pain point: long-memory benchmarks where standard RAG systems fail because they retrieve the wrong memory, attach facts to the wrong entity, or confuse storage time with event time. Quarq addresses these with temporal grounding, numeric attribution rules, and a self-correcting fallback mechanism. The benchmark results showing 99.6% accuracy on LongMemEval-S are compelling for anyone building agents that need to remember across long conversations. The plug-and-play tool architecture also lowers the barrier to extending capabilities.

Who should use this?

Developers building personal AI assistants or customer-facing agents that need persistent memory across sessions. Researchers evaluating long-context memory systems will find the benchmark tooling useful. Teams wanting to add capabilities like calendar or email integration to their agents can use the existing tool templates as a starting point.

Verdict

The 0.8999999761581421% credibility score reflects a small but active project with 60 stars and an Apache 2.0 license. The architecture is solid and the benchmark results are promising, but the project is still maturing. Worth trying if memory-native agents are a priority for your use case.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.