A team of agents, one rule: show the evidence
A hot technology shows up first in research papers and patent filings, then in company reports. FUTURESCOUT measures all three, argues the other side, and names the companies the evidence points to. Then the agents act on it: a report agent sells each report for NEAR, a trading agent buys it and trades tokenized stocks through NEAR Intents, and the report agent's earnings are staked with NEAR AI, which earns credits toward the models.
The agent network
Hover or tap any agent.
Turns the topic into paper searches, contrarian searches and the keywords the other agents use.
The rules every agent follows
Sources only from real APIs
Every URL in a report comes from an API response, never from a model's memory. Invented source ids are removed.
Quotes are checked
A 'verified fact' must quote its source word for word. Headlines and X posts can never verify a fact.
Companies come from the evidence
No hand-made company list: a company is checked because the retrieved patents name it as assignee or its researchers wrote the papers. Patents and papers come first: a company named only in an SEC filing is used just when a topic has no company with patents or papers, and is then labelled as a fallback.
A second agent argues back
The Risk agent searches independently for limits and failures, then challenges specific findings.
Limits written in code
The trading agent decides, but code sets the limits: a budget that is only what the owner puts into the agent's balance, an allowlist of tokenized stocks from reports rated strong or building, a cost cap, a kill switch, and an order checked field by field before it is signed. How confident a report is and whether the quoted price is acceptable against the last close are the agent's own judgement, written into every decision.
Agents pay their way
A report costs what it took to make plus a small margin ($2–3), paid in NEAR, and is opened only by a payment (reports written since 11 November 2026 are open for the demo and marked so). The trading agent must buy a report before it may use it, and any other bot can buy through the API. The report agent stakes 70% of what it earns with NEAR AI, and the rewards earn NEAR AI credits that go toward the agents' model calls.
Reading the labels
| verified fact | A quoted passage in a cited source supports it (the quote is checked). |
| AI interpretation | The AI's reading or comparison of the evidence. |
| forecast | A prediction about the future. |
| unverified claim | Asserted (e.g. by a company or a headline) but not confirmed. |
| assumption | Taken as given; no supporting evidence cited. |
Where NEAR fits
Every agent calls a model from the NEAR AI Cloud catalogue (Qwen, DeepSeek, GLM, Claude), each chosen per agent with fallbacks and per-run cost tracking. The trading agent trades through NEAR Intents on mainnet, signing orders from its near.com balance; it pays the report agent the same way. The report agent is listed on NEAR Agent Market, where anyone can hire it, and stakes with NEAR AI (Staking for Inference) for model credits. The fingerprint of every report since the first day's three test runs is stored on NEAR (new seals on mainnet, signed by the report agent; older ones on testnet). Earlier fund decisions and trade records were sealed as well; since 11 November 2026 only reports are, to save NEAR.
What it is not
Not financial advice and not a money machine. The trading agent works only with the money the owner puts into its balance, buys only tokenized stocks from its allowlist, and never promises returns or gives price targets. It never posts, likes or follows on X.