Koshi vs xerj
Local search and memory for AI coding agents compared
Koshi and xerj both provide local tools that help AI agents retrieve project information and retain memory. Koshi focuses on durable project context and team-shared decisions, while xerj centers on code and file search with BM25, vector, and hybrid options.

Koshi: Give AI Coding Agents Local Memory and Retrieval
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.

xerj: Local-First Search for Code and AI Agents
XERJ is a Rust search engine and MCP server for AI agents, combining BM25, vector and hybrid search behind an Elasticsearch-compatible API. It indexes local code and other files for code search, RAG and agent memory, and runs on Linux, macOS and Windows.
| Koshi | xerj | |
|---|---|---|
| Language | C# | Rust |
| License | MIT | Apache-2.0 |
| Stars | 2 | 13.2k |
| Forks | 0 | 947 |
| Last analyzed | Oct 13, 2026 | Oct 23, 2026 |
Key differences
- Koshi is written in C# and licensed under MIT; xerj is written in Rust and licensed under Apache-2.0.
- Koshi provides retrieval, persistent memory, context packing, and team telemetry; xerj emphasizes searching code and mixed-format files through an Elasticsearch-compatible API.
- Koshi shares project memory through directories that teams can commit and sync with Git; xerj provides namespace-based memory endpoints and indexes folders automatically.
- Koshi indexing is not automatic by default and installation requires a manual client restart; xerj supports automatic folder indexing and incremental indexing of changed files.
- Koshi has had no commits in the last 90 days; xerj has high recent commit and release activity, though it is changing quickly.
- Koshi is aimed at teams that want shared decisions and persistent agent context; xerj also targets reference-code search, log and document search, and security review.
Choose Koshi if you…
- need durable project decisions and conventions to persist across agent sessions.
- want to share a project memory vault with teammates through Git.
- need tools for planning token budgets and packing project context into prompts.
Choose xerj if you…
- need BM25, vector, and hybrid search over code and mixed-format files.
- want automatic and incremental indexing, including source-code symbols and line locations.
- need an Elasticsearch-compatible API or a single binary for Linux, macOS, and Windows.
Compare the numbers side by side, or add more projects
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.