AI Agent Tooling and Integration

4/3/2026

This post analyzes the memory architecture of Claude Code, identifying limitations such as a 200-line index cap, grep-only retrieval, lack of reasoning capture, complexity stacking, and memory lock-in. It then introduces 'memsearch', a persistent memory layer for AI coding agents that utilizes Milvus for hybrid semantic and full-text search. Memsearch stores memory as dated Markdown files, with Milvus providing a rebuildable vector index for efficient recall. It employs semantic search, BM25, and RRF for robust memory retrieval, addressing the limitations of Claude Code's native memory system.

4/1/2026

This post critically analyzes the limitations of the Model Context Protocol (MCP) for AI agent tool integration, specifically addressing context window bloat, passive architecture, and the inability of MCP servers to leverage the agent's LLM. It contrasts MCP with a CLI and Agent Skills paradigm, showcasing Zilliz's implementation with Zilliz CLI, Milvus Skills, and Zilliz Skills for managing Milvus and Zilliz Cloud. A key technical contribution is the memsearch memory layer, which uses a multi-stage process involving Milvus for initial retrieval, the agent's LLM for relevance evaluation, and a final drill-down, thereby reducing noise and improving context quality. The post provides a framework for choosing between MCP, CLI, and Agent Skills based on technical requirements.

3/2/2026

This post introduces the integration of Deep Agents, a LangChain framework for building AI agents with structured workflows, with Milvus for long-term memory. It explains how Milvus, as a vector database, enables agents to store and retrieve embeddings of conversations and tool results, overcoming context window limitations and high token costs. The integration uses Milvus's compute-storage separation, horizontal scaling, and high-concurrency query capabilities to provide persistent memory. The post details the technical setup using `CompositeBackend` with `StateBackend` for temporary data and `StoreBackend` with Milvus for persistent memories, along with code examples for dependency installation, memory backend setup, and agent creation.

2/13/2026

Introduces the memsearch ccplugin, a lightweight plugin for Claude Code that integrates with the memsearch library to provide persistent, searchable memory. It details the plugin's architecture, which uses shell hooks and the memsearch CLI to manage memory via plain Markdown files. The post highlights automatic context injection, background session summarization using Claude Haiku, and the avoidance of MCP overhead, contrasting it with heavier solutions like claude-mem.

2/13/2026

Introduced memsearch, a standalone, plug-and-play long-term memory library that extracts and open-sources the memory system from OpenClaw. Memsearch provides agents with persistent, transparent, and human-editable memory by storing all memory as plain Markdown files on the local filesystem and using Milvus for vector indexing. Key features include readable Markdown for debugging, Git-backed memory for versioning and collaboration, effortless migration due to plaintext storage, and shared Markdown files for co-authoring memory between humans and agents. The underlying workflows include Watch (automatic re-indexing on file save), Index (smart chunking, deduplication, and version-aware embeddings), Search (retrieval), and Compact (summarization).

1/23/2026

Introduces Anthropic Skills as a successor to MCP for AI agent tool integration. Details the structure and benefits of Skills, and provides a hands-on tutorial for building a custom Milvus Skill that automates RAG system setup, including collection creation, schema definition, indexing, and data ingestion via natural language commands.

11/19/2025

This post introduces Anthropic Skills as a new paradigm for AI agent tool integration, contrasting it with MCP. It explains how Skills offer modularity and reduced token overhead through progressive disclosure. The post provides a practical guide on integrating Milvus with Anthropic Skills by configuring the `claude-context` MCP plugin, demonstrating how to build custom Skills for semantic code search and architecture analysis within the Milvus codebase.

10/30/2025

This post demonstrates the integration of Milvus with Langflow, a visual RAG workflow builder. It provides a step-by-step guide on deploying Milvus, setting up Langflow, and configuring a RAG template to use Milvus as the vector database. The post details how to prepare test data, ingest it into Milvus via Langflow, and run a full RAG workflow, highlighting the code-light approach to building RAG systems.

8/22/2025

Introduces the Milvus SDK Code Helper, a system that connects AI coding assistants to the latest Milvus documentation via an MCP server. Details the setup process for integrating this helper into various IDEs and development environments (Cursor, Claude Desktop, Claude Code, Windsurf, VS Code, Cherry Studio, Cline, Augment, Gemini CLI, Roo Code). Highlights the benefits of using up-to-date code generation and provides examples of improved code quality compared to outdated AI-generated snippets.

7/29/2025

This post details the integration of Code Context, an open-source MCP-compatible plugin, with Qwen Code and Qwen3-Coder to enable semantic code search for AI coding assistants. It outlines the prerequisites, environment setup, and configuration steps for integrating Zilliz Cloud as the vector database. The tutorial demonstrates how to use Code Context's `index-codebase` and `search-code` tools to enhance AI coding copilots with the ability to understand existing codebases, find relevant code snippets, and assist in tasks like code renaming.

6/13/2025

Introduces Milvus MCP, a system combining Model Context Protocol (MCP) and Retrieval-Augmented Generation (RAG) to ensure AI-generated code for Milvus is always up-to-date. It addresses the issue of AI assistants producing outdated code by leveraging a Milvus database of vectorized documentation. The post details the architecture, transport mechanisms (stdio and SSE), and showcases three specific tools: pymilvus-code-generator, orm-client-code-convertor, and language-translator. It also compares Milvus MCP with other tools like Context7, DeepWiki, and Cursor Agent Mode, highlighting Milvus MCP's specialization and direct integration with official Milvus documentation.