# Koshi: Give AI Coding Agents Local Memory and Retrieval This repository profile is provided by osrepos.com, an open source repository discovery platform. Source: osrepos.com Repository profile: https://osrepos.com/repo/jsharma1105-koshi Generated for open source discovery and AI-assisted research. Koshi is a local-first MCP server for giving AI coding agents durable project memory, code retrieval, and shared team decisions. It suits teams that want context to persist across sessions and machines without relying on cloud services or API keys. GitHub: https://github.com/jsharma1105/Koshi OSRepos URL: https://osrepos.com/repo/jsharma1105-koshi ## Summary Koshi is a local-first MCP server for giving AI coding agents durable project memory, code retrieval, and shared team decisions. It suits teams that want context to persist across sessions and machines without relying on cloud services or API keys. ## Topics - csharp - mcp - mcp-server - agent-memory - context-engineering - bm25-retrieval ## Repository Information Last analyzed by OSRepos: Tue Oct 13 2026 08:26:26 GMT+0100 (Western European Summer Time) Detail views: 0 GitHub clicks: 2 ## Safety Notice OSRepos shares public repositories for knowledge and discovery only. Review source code, dependencies, licenses, and security implications before running or installing anything. ## Content ## Overview Koshi is a .NET MCP server that connects AI coding agents to project-aware retrieval, durable memory, and team-shared decisions. It aims to solve the problem of agents losing useful context between chats or working in isolation from decisions captured by teammates. It stores data locally, with shared memory directories that teams can commit and sync through Git. The project is currently quiet: it has had no commits in the last 90 days, and most contributions come from one person. ## Key Features - Offers MCP tools for retrieval, memory, context management, and team telemetry. - Indexes project code and documentation for agent search, including BM25 retrieval. - Stores facts, decisions, and patterns for recall across agent sessions. - Supports sharing a project memory vault through Git. - Provides tools to plan token budgets and pack prompt context. - Includes separate sub-agent personas for retrieval, memory, context packing, interaction quality, and orchestration. - Supports multiple MCP clients using the same on-disk format. ## Use Cases - A development team can record architectural decisions once and make them available to agents on teammates’ machines through a shared Git vault. - A developer working across multiple chats can use persistent memory to recover project decisions and conventions instead of re-explaining them. - An agent can search an indexed codebase and documentation when answering project-specific questions. - A team can use context-packing tools when it needs to fit relevant project information into an agent’s prompt window. ## Our Take Koshi offers a focused, local-first approach to persistent coding-agent context, but its quiet development and highly concentrated contribution base call for a cautious evaluation. **Good fit if:** - You use MCP-compatible coding agents and want local memory and retrieval without cloud dependencies or API keys. - Your team is comfortable committing a shared memory vault to Git. - You can evaluate a .NET 10 tool with tests and CI, while accounting for its recent lack of commits and limited contributor base. **Look elsewhere if:** - You need evidence of ongoing maintenance or rapid issue handling; there have been no recent commits or issue closures. - Your environment cannot run .NET 10 or you require a managed cloud service. - You need a project with contributions spread across a broader maintainer community. ## Project Health | Signal | Value | |---|---| | Status | **Quiet**: no commits in the last 90 days | | Last commit | 2026-08-21 (2 months ago) | | Commits, last 90 days | 0 | | Releases, last 12 months | 7 (latest v0.9.1, 2026-06-03) | | Contributors | 2 (top contributor: 96% of commits) | | Issues closed, last 90 days | 0 | | Pull requests merged, last 90 days | 0 | | Project age | 5 months | Checked on 2026-10-13 with the GitHub API. ## Project Facts - Language: C# - License: MIT - Stars: 2 - Forks: 0 - Topics: agent, bm25, context-engineering, csharp, dotnet, mcp, memory, model-context-protocol - Archived: no ## What You Need Detected in the repository: - C# (main language on GitHub) - A test suite and automated checks on GitHub Actions ## Getting Started On Linux or macOS, install with: ```sh curl -fsSL https://raw.githubusercontent.com/jsharma1105/Koshi/main/scripts/install.sh | sh ``` See the [README](https://github.com/jsharma1105/Koshi) and [installation guide](https://github.com/jsharma1105/Koshi/blob/main/docs/install.md) for Windows, other installation methods, and client setup. Restart your MCP client after installation. ## License in Plain Words **MIT** (permissive). - Commercial use: yes - Modify and redistribute: yes - You must keep: the copyright and license notice - Share your changes: no A summary, not legal advice: the LICENSE file is what applies. ## Alternatives - [plur](https://osrepos.com/repo/plur-ai-plur): PLUR stores shared agent memory in readable YAML files, while Koshi adds code retrieval and team decisions through a local MCP server. - [holaOS](https://osrepos.com/repo/holaboss-ai-holaos): holaOS embeds shared local memory in an Electron workspace for agents and apps, while Koshi focuses on project memory and code retrieval via MCP. | Project | Language | License | Stars | Status | |---|---|---|---|---| | **Koshi** | C# | MIT | 2 | Quiet | | [plur](https://osrepos.com/repo/plur-ai-plur) | TypeScript | Apache-2.0 | 298 | Active | | [holaOS](https://osrepos.com/repo/holaboss-ai-holaos) | TypeScript | – | 11.4k | Active | ## Considerations Koshi is early in its lifecycle and marked quiet by its recent activity: there have been no commits in the last 90 days, despite several releases in the past year. Two contributors are listed, with nearly all contributions from one person, and no issues were closed in the last 90 days. Teams should test the installer, indexing workflow, and shared-vault setup in their own client environment before relying on them. Indexing is not automatic by default, and client restart is a manual step after installation.