windwhiterain/dsh-model-proxy ↗★ 0

dsh-model-proxy

Route selected models through an outbound proxy in DeepSeek Harness, falling back to a direct connection while that proxy is unavailable. 适合需要对特定模型或域名配置代理、并在失效时自动回落直连的用户。

Package
dsh-model-proxy
Compatibility
Unverified
Version
0.1.0
License
MIT
Last updated
Sep 25, 2026

Other repositories with this package name

Install

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:windwhiterain/dsh-model-proxy

Configuration

- id: model-proxy
  config:
    enabled: true
    proxy: http://127.0.0.1:10793
    origins:
      - opencode.ai
    models:
      - grok-*
      - gpt-*-luna
    fallbackToDirect: true
    probeTimeoutMs: 300
    probeCacheMs: 2000
    peekBytes: 16384
FieldMeaning
enabledWhen false the plugin installs nothing and every request keeps its route.
proxyEndpoint the matching models use. http: or https:, with optional credentials.
originsHosts whose models may be routed. Requests to every other host are untouched.
modelsModel patterns; * matches any run of characters, every other character matches itself.
fallbackToDirectWhile the proxy refuses connections, matching requests use the previous route instead of failing.
probeTimeoutMsTCP connect timeout for the reachability probe.
probeCacheMsHow long one probe verdict is reused, so a burst of requests probes once.
peekBytesHow many body bytes may be inspected before giving up on finding a model name.

Edit the row in /profiles/ /cordis.patch.yml (hot-reloaded), or override the same id from a settings layer. A configuration value that cannot be used fails the activation instead of silently routing traffic to the wrong place.