Note that this uses their Deepseek Search API and that it sends user identifiers on each request, so even if you run this locally, there's still some data leaking. Full telemetry is off by default though.
Notable that DeepSeek chose Cordis as the plugin core rather than building another MCP-only runtime — the paper's 'spatiotemporal composability' model is a genuinely different answer to the agent-middleware problem than OpenAI's function-calling or Anthropic's MCP. Also interesting: 11k forks within days of release means the plugin ecosystem is the real moat, not the harness itself. For anyone testing: the dsh-plugin topic on GitHub is the fastest way to see what the community is building.
Looks cool.
Note that this uses their Deepseek Search API and that it sends user identifiers on each request, so even if you run this locally, there's still some data leaking. Full telemetry is off by default though.
Notable that DeepSeek chose Cordis as the plugin core rather than building another MCP-only runtime — the paper's 'spatiotemporal composability' model is a genuinely different answer to the agent-middleware problem than OpenAI's function-calling or Anthropic's MCP. Also interesting: 11k forks within days of release means the plugin ecosystem is the real moat, not the harness itself. For anyone testing: the dsh-plugin topic on GitHub is the fastest way to see what the community is building.