The task
How can an agent understand an organization whose activities, dependencies, and claims are spread across many sources?
My solution
I built an evidence workspace that resolves an organization, follows its relationships, retrieves source material, and produces a brief with a traceable record of the investigation.
A small set of MCP tools separates entity lookup, graph traversal, source retrieval, and reporting. Each consequential statement can be traced to an evidence ID and a source location. Images and sampled video frames provide another view of the same evidence.
How it works
Python · MCP · local models · source-linked reports- 01
Resolve
Match the organization before investigating. Ambiguous names produce a clarification request.
- 02
Trace
Follow relevant relationships and read the evidence behind each connection.
- 03
Explain
Separate source claims from observations and inferences in a cited analytic brief.
Try it
Interactive browser demonstrationInteractive browser demo
Follow a finding back to its evidence
Choose a question, then open its sources. Compare dated claims and see why the conclusion keeps its qualifications.
Loading the bundled evidence…
Methods / evidence / reproduction
Read the research.
The documented Riverwatch investigation uses 15 tool calls and 9 sources. Its review assesses one authored synthetic case; the record preserves the report, citations, and tool sequence.
The paper connects the research question to the implementation, worked examples, and evidence. The project directory includes the runnable software, documentation, and tests.