yunxiyang/dsh-web-search-litellm ↗★ 0
dsh-web-search-litellm
Web search provider for the DeepSeek Harness ctx.web seam via the LiteLLM proxy OpenAI Responses API: DeepSeek's native server-side web_search through your proxy. Fixes web_search 'Authentication Fails, Your api key is invalid' (DEEPSEEK_API_KEY that is actually a proxy key); no Anthropic protocol, no third-party search service, reuses LITELLM_API_KEY.
AI 분석
核心用途是解决因使用 LiteLLM 代理导致原生 DeepSeek 联网搜索鉴权失败的问题。适合企业内网或必须通过代理访问 API 的用户。
설치
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:yunxiyang/dsh-web-search-litellm何时使用 / When to use
Pick this provider when any of these is your situation:
web_searchfails withAuthentication Fails, Your api key: ****XXXX is invalid— usually becauseDEEPSEEK_API_KEYholds a LiteLLM proxy key, not a DeepSeek platform key.- All company traffic must go through LiteLLM (direct api.deepseek.com is blocked or forbidden).
- You prefer the OpenAI Responses protocol over the Anthropic
/messagesformat. - You want no free-tier / third-party search service (Tavily, Brave, Exa, …) — search stays on DeepSeek's official server side.
- You use
openai/deepseek-v4-flashoropenai/deepseek-v4-prothrough a LiteLLM proxy as your main model.
Configuration
Settings section web-search-litellm (harness Settings UI) or the bundle patch config:
| key | default | meaning |
|---|---|---|
baseURL | $LITELLM_SEARCH_BASE_URL → http://127.0.0.1:4000/v1 | LiteLLM proxy root; /responses is appended |
model | openai/deepseek-v4-flash | model id routed through the proxy (must support server-side web_search) |
apiKeyEnv | LITELLM_API_KEY | credential reference resolved at each search |
apiKey | — | optional literal key (secret role) |
maxTokens | 4096 | max_output_tokens for one search request |
timeoutMs | 60000 | per-request timeout; times out as a structured WEB_TIMEOUT error |
Known upstream limits (not configuration issues)
- DeepSeek's Responses API documents
includeas not supported, so structured result items are consumed server-side; sources therefore carryurlonly (no title/snippet). - Each search costs one DeepSeek model turn (official mechanism).