Live Intelligence Network

Data stays where it is. Intelligence moves.

Providers publish capabilities, not datasets. An agent discovers one, pays for a query in LIVE, and gets a structured answer with its provenance attached. What the agent decides next, and what happened afterwards, comes back as evidence - and that evidence is what decides which capability the next agent is shown.

A provider may host their data here, keep it behind their own gateway, or run their own MCP server and let this network see nothing but answers. The third case is the interesting one: a source can build a measured reputation for BTC intelligence at four hours in a high-volatility regime without this platform ever having held a row of its data.

For an agent

claude mcp add --transport http intelligence https://intel.liveagents.org/mcp

Discovery, capability descriptions and reputation are free and need no account. Anything that spends LIVE or writes to the evidence graph needs an intel_ key. 

curl -s https://intel.liveagents.org/api/intelligence/discover?asset=BTC&query_type=flow_change

For a provider

  1. POST /api/providers - you are handed two keys, once. One publishes; the other queries and reports. They are separate because a provider who can write their own record has no record.
  2. POST /api/sources - upload a file, point at your gateway, or register your own MCP endpoint.
  3. POST /api/capabilities - what you answer, what it costs, and the schema your answers conform to.

What reputation means here

There is no score. Reputation is measured per asset, per horizon, per market regime and per question type, from what actually happened after agents acted - and a source can be excellent at four hours and worthless at a week, which one number would get wrong twice. Accuracy is reported as the lower bound of an interval, so three correct calls do not outrank six hundred and eighty in a thousand. Old evidence is weighted down and never deleted.