RedSearchAgent

RedSearchAgent / REDSearcher

Public

REDSearch: A scalable, cost-efficient framework for long-horizon search agents. Features complex task synthesis, optimized mid-training, post-training (SFT and Agentic RL)

35
1
89% credibility
Found Feb 19, 2026 at 18 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

A research project offering datasets and a forthcoming framework to train AI agents for efficient, long-duration web searches.

How It Works

1
🔍 Discover REDSearcher

You come across this project promising smarter search helpers that tackle long, complicated searches step by step.

2
🌐 Explore the project

Visit the colorful project page to see what's available, like ready-to-use training examples shared openly.

3
Pick your training data
📝
Text examples

Grab language-focused data to train on words and conversations.

🖼️
Image and text mix

Select multi-modal data blending pictures and words for richer learning.

4
💾 Download the data

Easily pull the collections into your workspace to start preparing your smart searcher.

5
🚀 Kick off training

Feed the data into your AI trainer to build advanced planning and searching abilities that feel magical.

6
⏳ Stay tuned for more

Watch for upcoming guides and tools to make training and testing even simpler.

🎉 Master long searches

Your search helper now handles tough, drawn-out tasks smoothly and affordably.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 18 to 35 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 REDSearcher?

REDSearcher, on red search github, is a scalable, cost-efficient framework for building long-horizon search agents that tackle complex tasks. It solves the pain of training agentic agents by providing Hugging Face datasets for SFT (10K text trajectories, 5K multi-modal) and RL (1K demos), plus optimized mid-training and post-training via SFT and agentic RL. Developers get ready-to-use data in ShareGPT format for ms-swift and Slime, with full training and eval code coming soon.

Why is it gaining traction?

It stands out with features like dual-constrained complex task synthesis, tool-augmented queries, and RL in simulated environments, cutting costs for long-horizon search over basic agent frameworks. The hook is instant access to high-quality datasets that plug into popular tools like ms-swift for SFT and Slime for RL, letting devs prototype agentic search fast without starting from scratch.

Who should use this?

AI engineers training search agents for web-scale queries or multi-step reasoning. Teams building agentic systems with proactive tool use, like RAG pipelines needing better planning and interaction. Researchers iterating on long-horizon RL who want cost-efficient scaling without custom data pipelines.

Verdict

Promising for agentic devs eyeing scalable search frameworks, but at 16 stars and 0.9% credibility score, it's pre-alpha—datasets shine now, but wait for training scripts before committing. Star it and check updates if long-horizon agents are your jam.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.