Unverified50% confidenceFactExact time
Retrieve-then-Plan (RtP) is currently the dominant approach in the industry's tool discovery field; its process is: first embed-encode the user query, retrieve semantically similar candidate tools via approximate nearest neighbor search, then inject them into the prompt for the LLM to perform planning.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/15/2026
First Seen
Valid until: 1/13/2027
Sources
Related Entities
Related Claims
UnverifiedAn increasing number of information extraction tasks in the current NLP field are shifting from discriminative methods to generative methods.64% similarUnverified语义缓存（Semantic Cache）通过计算新查询与历史查询的向量相似度，超过阈值时直接返回缓存答案，GPTCache等开源工具实现了这一机制，通常与TTL机制结合64% similarUnverifiedDevelopers who want to understand such projects in depth are advised to first become familiar with the latency and jitter models of RTP/RTCP, then understand the signaling state machine of SIP or WebRTC, and finally return to Phreak.ing to observe how it re-expresses these concepts in a more lightweight way.62% similarUnverifiedTaking LLMs as an example, context includes the system prompt, user query, externally retrieved information, historical interaction records, and additional system outputs.60% similarUnverifiedThe recommended hybrid workflow is to use large models (such as ChatGPT) for planning and solution review, and local models for specific coding execution, so as to balance data privacy and cost control60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1040078API
curl https://kongchang.com/api/v1/knowledge/claims/1040078MCP
get_claim(id=1040078)