LoginSign Up
← Back to all sparks
A

Apache ShenYu

INFRA · APIS
Velocity0.0

High-performance API gateway and microservice proxy

shenyu.apache.org ↗

An API gateway that has quietly grown an LLM plugin family — and then stopped shipping.

api-gatewayai-proxymcptoken-limitingnacosapache
◆Current state
Apache ShenYu's tracked releases run through the 2.7.0.x patch line and stop in November 2025. The consistent thread across those releases is a set of AI plugins accumulating alongside the traditional gateway ones: an AI proxy, an AI token limiter, an AI request transformer, and an MCP server plugin that shows up in the most recent entry through timeout and request-size fixes. The rest of each release is bug fixing and test coverage, with configuration synchronization through Nacos accounting for a recurring share of the defects.
◆Where it's heading
ShenYu is positioning as a gateway for model traffic as well as service traffic, and the pattern is telling — the AI plugins appear first as features and then, one release later, as bug reports, which is what real usage looks like. Rate limiting by token rather than by request is the specific piece that matters, since it is the unit LLM providers actually bill on. Against that, the release notes are undifferentiated pull-request lists and nothing has shipped in the tracked feed for over eight months.
◆Prediction
Expect further hardening of the MCP server plugin, which is the newest and least settled of the AI additions. The gap since 2.7.0.3 should be checked against the project's actual activity before drawing conclusions from it.

◆Recent moves

  1. 8mo ago

    MCP server plugin fixes and Nacos sync cache corrections

    A patch release dominated by fixes to the MCP server plugin — request configuration length limits, tool timeouts and error handling — plus several Nacos data synchronization bugs including a selector cache that was not being cleared on removal. The MCP fixes indicate the plugin is being exercised in real deployments rather than sitting unused.

    View source ↗
  2. 11mo ago

    AI request transformer plugin joins the proxy and token limiter

    Adds an AI request transformer plugin and fixes the existing AI proxy and token limiter plugins, alongside a data buffer leak fix. This is the release where the AI plugin family stopped being a single proxy and became a set of composable pieces.

    View source ↗
  3. 1y ago

    JWT SPI extension and Dubbo method configuration land

    A patch release weighted toward refactoring and test coverage across the sync-data, registry and register-common modules, with a JWT SPI extension for compatibility with older JWT tooling and Dubbo method-level configuration. Consolidation work on the traditional gateway path rather than new capability.

    View source ↗