dsh-tool-bandit-search: Thompson Sampling-Based Adaptive Search Strategy Plugin
2026-09-14
DeepSeek Harness plugin
DeepSeek
DSH plugin
dsh-tool-bandit-search
Thompson Sampling
Search strategy optimization
dsh-tool-bandit-search is a DeepSeek Harness plugin that adaptively selects search strategies using the Thompson sampling algorithm. It replaces the built-in `web_search` with a `search` tool that offers "quick" and "thorough" modes. The system computes rewards based on the number and speed of results, dynamically adjusting strategy selection. Installation requires running the specified command and restarting the Web UI. Note that this plugin only optimizes the strategy for individual invocations, stores state in memory, and is based on the developer preview version.
Read More