pi-ds-web-search

Silent-by-default DeepSeek server-side web search for Pi: overrides the built-in deepseek provider (v4-flash via Responses API with web_search injected, v4-pro unchanged) plus an on-demand deepseek_search tool.

Packages

Package details

extension

Install pi-ds-web-search from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:pi-ds-web-search
Package
pi-ds-web-search
Version
0.1.0
Published
Aug 6, 2026
Downloads
176/mo · 24/wk
Author
waitforit_785
License
MIT
Types
extension
Size
65 KB
Dependencies
0 dependencies · 2 peers
Pi manifest JSON
{
  "extensions": [
    "./extensions"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

pi-ds-web-search

DeepSeek's built-in server-side web search for Pi — silent by default.

中文版 README

What it solves

Pi's built-in deepseek provider uses Chat Completions, which has no server-side search. DeepSeek's Responses API ships a native web_search tool (search → read pages → summarize in one request), but Pi doesn't expose it. This plugin bridges that gap: install it, and your DeepSeek model can search the web with no extra keys or setup.

Features

  • Silent by default — overrides the built-in deepseek provider: deepseek-v4-flash switches to the Responses API with web_search injected into every request; deepseek-v4-pro stays on Chat Completions (Responses does not support Pro yet).
  • Works with any model — the deepseek_search tool calls DeepSeek's API directly, so it also works when your main model is from another provider (GLM, OpenAI, Claude, ...).
  • Interactive setup — /ds-ws config: API key (with a reminder when missing), model selection, auto-search toggle.
  • Cost control — auto-search is on by default; toggle off anytime with /ds-ws off.

Install

Requirements: Pi 0.83+, Node 22+.

pi install npm:pi-ds-web-search        # after npm publish
# or
pi install git:github.com/Lan-zk/pi-ds-web-search
# or local
pi install E:\File\self\github\pi-ds-web-search

Then /reload. No configuration needed if you already logged in to DeepSeek (pi /login); otherwise run /ds-ws config.

Usage

Command Description
/ds-ws Show status
/ds-ws config Interactive setup: API key, search model, auto-search toggle
/ds-ws on / /ds-ws off Toggle automatic web search

Automatic mode (default): the model decides when to search. On demand: ask the agent to call deepseek_search({ query, reasoning_effort?, max_output_tokens? }).

How it works

  1. pi.registerProvider("deepseek", ...) overrides the built-in provider: deepseek-v4-flash uses the Responses API (openai-responses); auth is inherited from Pi (stored credential → DEEPSEEK_API_KEY).
  2. before_provider_request injects { type: "web_search" } into every Responses payload for deepseek-v4-flash. DeepSeek executes the search server-side and returns the final answer in the same response.
  3. The deepseek_search tool posts to https://api.deepseek.com/responses directly — independent of your main model — and returns the answer plus sources and token usage.

Cost

Each server-side search is a paid API request; search context can add ~100k+ input tokens per query. Use /ds-ws off when you don't need it.

References

License

MIT