Configure the research team.
Choose how FETCHN agents work, then inspect the evidence and performance behind each result.
1. Open Agents
Sign in and open Agents. The left panel saves your research choices; the right panel shows recent results and separate forecast and trading measures. Your choices apply to new Markets analyses and scheduled Favorite Token scans. Previous results keep their original snapshot.
2. Choose the model tier
Standard uses the current production model. Advanced lets eligible holders select from configured model options. Your account needs a verified linked wallet with the required $FETCHN balance; access is checked again when the run starts. If access or the selected model configuration is unavailable, the run is blocked. A provider fallback can occur after an upstream model failure, so the result shows the requested model rather than claiming a guaranteed execution model.
The operator publishes the holder threshold separately when access is enabled. Choosing a model changes research, not your authority to trade.
3. Choose specialists and providers
Select at least two internal specialists and one data source. Specialists are research roles; the provider switches determine which configured feeds may be queried. A selected provider can still be unavailable for a given asset or run. Mandatory asset and reference-price checks remain in place.
After saving, open a completed analysis from Recent analysis to see the agent reasoning, risks and URLs that actually accompanied the result. A URL is evidence provenance, not independent proof of the claim.
4. Add your own research
Under External research, connect an existing FETCHN HTTPS agent or a read-only MCP Streamable HTTP research tool. Choose whether the MCP tool is a research agent or data provider, enter its HTTPS endpoint, exact tool name and optional bearer token, then test it. Select up to three verified connections in Analysis settings and save.
MCP currently accepts JSON responses. An external report becomes untrusted evidence for the internal reviewers. The external server receives the research question and asset context; it never receives wallet keys or order permissions. See MCP transport documentation for the protocol.
5. Read the record correctly
Resolved forecasts and average Brier score appear only when outcomes have been recorded; lower Brier is better. Realized trading PnL sums closed testnet strategy results in test USDG and is separate from forecast accuracy. Missing data displays a dash. Each recent analysis links to a page with the specialist breakdown and source trail.