In the news
T-Search: An Open Agentic Retriever and Playground for Hard Multi-Step Search
arXiv cs.AI · Published · 3 min read
In 30 seconds
- What happened
- T-Search, an open-weight agentic retriever for multi-step search over fixed corpora, achieves 56.0 Recall@10 on seven benchmarks including a new Russian dataset.
- Why it matters
- Engineers building retrieval systems or RAG pipelines should care when they need multi-round search with interpretable evidence ranking without retraining for different generators.
- Watch out
- T-Search searches only fixed corpora and leaves answer generation to downstream models, so it cannot handle dynamic data sources or end-to-end question answering alone.
- agent
- agentic
- rag
- retriever
- fine-tun
The patterns behind this
- Tool Retrieval (Tool RAG)
- Query Transformation Retrieval
- Agentic Context Engineering (Evolving Playbook)
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.