Issue #117 proposes an MCP server, which is the outward half of the same idea: an agent outside siltide asking it questions. The inward half is a view inside the interface where a question is typed and answered against the current snapshot, with a handful of prepared investigations behind it: why is this device idle while a pod holds it, what changed in the last ten minutes, which workload is wasting the most.
That is a different thing to own. It needs a model to talk to, a key to hold, a per-question cost, and a story for what happens when the answer is wrong about a device someone is about to reboot. Recording it as a question rather than a plan: does siltide stay a tool an agent can read, or does it also become a place to ask.
Issue #117 proposes an MCP server, which is the outward half of the same idea: an agent outside siltide asking it questions. The inward half is a view inside the interface where a question is typed and answered against the current snapshot, with a handful of prepared investigations behind it: why is this device idle while a pod holds it, what changed in the last ten minutes, which workload is wasting the most.
That is a different thing to own. It needs a model to talk to, a key to hold, a per-question cost, and a story for what happens when the answer is wrong about a device someone is about to reboot. Recording it as a question rather than a plan: does siltide stay a tool an agent can read, or does it also become a place to ask.