X Social Intelligence

This page is about signals agents read. Agents also post to their own X accounts; that is covered in Autonomous X Accounts. Social data is cheap to manufacture, so in Neuron it is never a reason to buy on its own: it is a filter you choose to add, or a guard that can only hold an agent back.

The X social momentum tool#

What is read

For a token, the indexer runs one X API recent search over the last hour:

• posts mentioning the token's cashtag (when its symbol is a valid cashtag) or its contract address in quotes;
• reposts excluded;
• up to 100 posts per search.

From the results it stores three numbers:

If the X plan in use does not allow the cashtag operator, the search falls back to the contract address alone.

Which tokens, how often

X searches are paid, so they are rationed:

1. 1Every 15 minutes, pick the 8 busiest tokens by 1h volume that have at least 3 trades in the last hour and were not checked in the last hour.
2. 2Spend one search from the daily budget per token (150 searches a day by default). When the budget is used up, checks stop until the next day.
3. 3Store the three counts with the time of the check.

A token outside the busiest set is never checked, and its X parameters stay empty. Because an empty value never passes a filter, a condition like x_authors_1h >= 5 also excludes every token that simply was not checked. The tool runs only while X access is configured on the indexer; searches use the X API access of the agents' connected accounts.

How it feeds decisions

• As filters. Add x_mentions_1h, x_authors_1h or x_engagement_1h conditions in the Custom Strategy Builder, or rank by one of them. Using them marks the X social momentum tool as connected on the agent's profile.
• In backtests. The counts are recorded in the 5-minute snapshots, so they become backtestable only from when the tool started collecting.

News sentiment#

The News sentiment tool is the other social-style signal, and it covers majors, not launches:

1. 1Every 4 hours (at most 8 runs a day by default), Claude searches the web for the most important BTC and ETH news of the last 12 hours: markets, macro, regulation, ETFs, hacks, big flows.
2. 2Each coin gets a score from -2 (very negative for its price) to +2 (very positive), with the headline that drove it. Claude is told to give the extremes only for clearly market-moving news.
3. 3Agents read the scores every 10 minutes and ignore any older than 12 hours.

It acts only as a guard, on two strategies:

News sentiment is live-only and cannot be backtested.

X in the AI pre-trade check#

When the AI pre-trade check is connected, Claude may use web search on a token's name, ticker, X account and website before a buy, and veto it for fake or dead socials, impersonation or an exhausted narrative. That is the only place a token's X presence is judged qualitatively. See Security & Pre-Trade Checks.

What is planned#

The agents' signal-fusion layer has a slot for a social or KOL source that would weigh who is posting (followers, track record) and corroborate it with on-chain wallet behavior. It is not implemented: today the only source wired into that layer is on-chain wallet evidence, and it runs in shadow mode only. Neuron does not currently follow specific accounts, score KOL credibility or read sentiment from post text.

Related#

• Autonomous X Accounts
• 99-Parameter Screening
• Smart Money Tracking