1/ Tearline is now Dataline. Same team, same product surface. The name change names the job: the data lifeline an autonomous agent runs on. If your agent has to act on crypto data, this thread is for you.
The next big Dataline milestone: announcing the Data Launch Partner Program, opening the initial cohort.
Building the data layer for the next era of AI. Internet and crypto should be unified, not walled gardens.
Your AI is evolving. Shouldn't your data be too?
Thread π
Dataline is already running in action:
19.4M+ on-chain transactions
96.4% execution success rate
BNB Chain Β· Sui Β· TON
Three production agents live on top: ChatPilot, GhostDriver, FlowAgent.
4/ When a venue disappears or a competitor takes share, the customer doesn't rewrite the agent.
The schema absorbs the change two layers down. Whether Polymarket comes back online or Kalshi extends its lead, the agent code stays.
more about using one Schema for both markets
https://t.co/mvlqtpMdKD
1/ Polymarket went dark in India last week.
Indian ISPs cut access after a government betting-platform directive. Kalshi could be next.
If your agent only reads Polymarket's API, it doesn't have a graceful fallback today. Here's what one schema across both venues looks like:
3/ get("senate-2026-OH")
Dataline returns one response carrying both venues. Source IDs, freshness per source, divergence_flag = true when Polymarket and Kalshi sit more than 3pts apart.
The agent picks the median, the range, or the gap by reading one field.
4/ Dataline returns every fee inline before the trade executes: pool quote, swap event log, price snapshot, freshness flag. The agent reads one response and decides on a number that doesn't move after the click.
Building a trading agent? Imagine it operating on fee data that isn't clean.
For an agent, the gap between the quoted price and the executed cost isn't friction. It's a wrong trade. Here's what the real number looks like:
3/ The displayed fee is incomplete. Slippage, bridge cost, routing markup, and gas all stack on top of the quoted percentage. If the agent reads only the quote, it ships decisions on a number that's already wrong by hundreds of bps.
9/ Token math: a 1KB shaped Dataline response is ~250 tokens. The equivalent raw GraphQL payload is ~1,800 tokens.
At scale that becomes a real bill. Your model also parses the shaped response more reliably.