Dataline: $10.6M to Give AI Agents Data They Can Trust

On October 31, 2026, a startup called Dataline announced it has raised $10.6 million to become what its founders call "the routing layer for AI data" (the OpenRouter of data feeds), built for the moment when AI agents stop asking questions and start spending money. The pitch: one integration that gives your agent clean, cross-verified data from roughly 50 live sources, with an AI-generated confidence score attached to every response, so the agent can refuse bad data before it acts.

Here is the honest version first, because the press release buries the lead. The round actually closed in April 2026, according to an interview co-founder Nuo Shi gave The Block earlier in October. The October 31 release is the public announcement of an already-signed seed round (structured as a SAFE with token warrants, at an undisclosed valuation). The headline stats (19.4 million onchain transactions, a 96.4% agent task success rate) are company-reported, and a public token sale planned for later this year has no published terms. None of that kills the story. But if you are an agent builder deciding whether to wire this into your stack, you should know which numbers come from the company and which come from anyone else.

So what is Dataline, what does a quarter a call actually buy, and how does it compare to the data pipes you already use? I read the announcement, the independent reporting and the live pricing page to find out.

What Dataline actually is

Dataline is a decentralized network for agent-payable data, live at dataline.xyz. The core product is a single API your AI agent calls instead of stitching together a dozen data integrations itself. Describe what you need in plain language, and the router picks the sources, normalizes everything into one schema, aggregates the results and hands back clean JSON with a confidence score.

The launch coverage spans onchain and offchain sources. Dataline says the network covers around 50 data sources, and its first "AI Data Launch" partner group includes Hyperliquid, Polymarket, Kalshi, Base, Robinhood Chain, BNB Chain, Sui and TON. The site's own stats page claims 19.4 million onchain transactions and a 96.4% agent task success rate. Those figures are the company's, not an auditor's, so treat them as traction signals rather than verified throughput.

The developer surface is designed for how agents get built right now. Dataline ships an MCP-native integration: one add-server command plugs it into Claude, Cursor, Windsurf or Eliza, with JSON-Schema-clean tools and per-call pricing visible in each response. The homepage claims a P50 of about 87ms for cached intent calls and roughly 250ms for fresh multi-source aggregation. Again: the company's numbers, with no independent benchmark yet. But the shape of the product is what matters: it is a runtime layer your agent hits at decision time, not a dashboard you read afterward.

The confidence score is the whole bet

The feature Dataline leads with is the confidence score. Every API response carries one: a versioned score from 0 to 100 that weighs multi-source consistency, data freshness, coverage and source reliability. The idea is that your agent does not just receive data, it receives a judgment about whether to trust that data before acting.

Shi's own example, given to The Block, is worth quoting because it explains the problem better than any marketing copy: "a stablecoin priced at $0.97 could be a bargain or the start of a depeg, and a simple price feed / API can't tell the difference. Dataline pairs the pricing data with context on peg stability and cross-venue outliers, so bots, smart contracts, and people know which one they're looking at."

That is the clearest statement of what "AI-generated confidence score" means. It is the company's own scoring system: no public methodology, no third-party audit, no published accuracy track record for the scores themselves. The router does rebalance in real time when a source drops out, and the score reflects the degraded consensus so your agent can refuse, retry or proceed with caution. The mechanism is thoughtful; the proof will come when real trading agents run it against real money and the scores get stress-tested in public.

What $0.25 a call actually buys

This is the part most launch coverage skips, so I read Dataline's live pricing page myself. The pricing is unusually simple for crypto-adjacent infrastructure:

The foreign-currency figures are approximations at November 1, 2026 market rates (USD to GBP 0.7469, USD to CAD 1.3975), and your card provider's actual rate will differ. The pricing math worth doing: 500 free calls at $0.25 each is $125 of free usage, which is generous for testing. But Pro works out to about $0.06 per request inside the allowance ($299 divided by 5,000) versus $0.25 pay-as-you-go, so the subscription only makes sense if you consistently clear roughly 1,200 paid requests a month. Below that, stay on x402.

One discrepancy worth flagging: the homepage FAQ mentions a "Hobby tier" at $19 a month with 50,000 calls and MCP access, but the pricing page as I read it on November 1 lists only Free, Pay-as-you-go, Pro and Elite, with no Hobby tier anywhere. The FAQ copy may be stale, or the tier may be rolling out. I would not budget around a plan that does not appear on the rate card.

The payment story is the other half of the pricing. Dataline uses the x402 protocol, the same pay-per-call standard behind the agent payment rails I covered in my Mastercard Agent Pay breakdown, which lets agents pay for individual API requests directly without a subscription or a credit card on file. That is a real fit for agent workloads: an agent that makes 40 data calls a day should not need a seat license.

The marketplace: bounties, stakes and slashing

Beyond the curated feeds, Dataline runs a two-sided marketplace for data itself. Anyone can publish a feed and price it per call. Anyone can post a bounty for data they need (the press release compares it to a job board for data), and any fulfiller in the network can claim it, supply the data and get paid. Payments settle through x402, sellers' credentials never leave the gateway, and an optional zero-knowledge attestation tier lets providers attach a verifiable proof bundle to each response.

Quality is enforced economically rather than by any central authority. Fulfillers stake tokens to submit responses: good data earns the bounty, bad data gets the stake slashed. If a customer rejects data, The Block's reporting says an AI review happens first, followed by a network vote if needed. Providers whose data fails lose their tokens, which get added to the reward pool for another provider.

This is Dataline's most novel part, and the one with the least public track record. Stake-slashing data markets work well in whitepapers and unevenly in production, because the failure cases (data that is plausible but stale, or correct at one venue and wrong at another) are where "an AI review" is hardest to trust. Dataline has published no dispute data, and there is no independent evaluation of the staking economy. Watch this space, but do not build your agent's critical path on it until there is evidence.

The token sale and the fine print

A few facts from the reporting that the press release soft-pedals:

• The round was signed in April 2026 as a SAFE with token warrants, at an undisclosed valuation. No investor received a board seat, observer role or advisory position. Shi told The Block the round maintained a high level of founder control.
• Dataline plans to launch its native token by the end of 2026 and hold a public sale on Base through Uniswap's Continuous Clearing Auction. In the October 21 interview, Shi said sale details were expected "next week" (pointing to early November), but as of November 1 no terms, raise target or token allocation have been published. Shi also said the company wants it to be the largest Uniswap CCA held on Base, which is the company's goal, not a fact about the market.
• The company has not started charging fees yet. It plans to make money by taking a cut of data payments, and its foundation plans to earn money by providing its own data, per The Block.
• The team is mainly based in Europe and North America; headcount was not disclosed. Dataline was founded in 2025 after winning the BNB Chain AI hackathon (it was formerly called Tearline and renamed in May 2026), and Shi says the new funding goes toward more data sources, better verification, and expansion into robotics and AI training data.

How it stacks up: Chainlink, Pyth, The Graph and the rest

Shi named Chainlink, Pyth, The Graph and Grass as the existing data platforms when asked who the competition is, but said Dataline was built natively for AI agents, not for smart contracts or dashboards. The site's own FAQ makes the positioning more explicit: Chainlink and Pyth are oracle networks designed to push price data onchain for smart contracts; The Graph, Dune and Alchemy are query layers over indexed historical state, strong for analytics dashboards but slow for live agent decisions; a direct CoinGecko or exchange SDK call works for one source, but becomes a multi-week integration per source.

Dataline's claim is the middle layer between your agent and 22+ live sources: live aggregation across exchanges, chains and prediction markets, one schema, a confidence score, and execution in one round-trip. That framing is the company's, and it is broadly fair as positioning, but it remains positioning, not a benchmark. There is no published head-to-head latency or accuracy comparison against any of those alternatives. If you are evaluating, the right test is boring and concrete: point your agent at Dataline and at your current stack for the same queries over a week, and compare the confidence scores against outcomes you can verify yourself.

Who should try it — and the common mistakes

The 500 free calls with no card required make the trial decision easy: if you are building an agent that consumes market data (a trading agent, a prediction-market bot, a research agent that needs live prices), you can test the response shape in under a minute. The free access for Base and Robinhood Chain builders is a real incentive if you are already deploying there.

The mistakes to avoid here are the same ones every agent-data story shares. First, do not confuse the confidence score with a guarantee; it is one company's heuristic, and your agent still needs its own refusal logic for low scores. Second, if you touch the "Act" layer, where the site offers non-custodial execution for swaps, perps, bets and votes (your agent signs, your wallet broadcasts), remember that an agent spending your money with your signature puts your money at risk, whatever the confidence score says. It is the same agent-permissions problem I covered in the 1Password for Agents piece: the credential plumbing being clean does not make the agent's decisions safe. Third, on-chain settlement means every failed call still costs gas somewhere in the stack, so run the pay-as-you-go math before you scale past the free tier.

Dataline is one of the more substantive entries in the agent-payments wave of late 2026: a real product, a published rate card, and a mechanism (confidence-scored data) that addresses a real failure mode for trading agents. The open questions (whether the staking economy holds up, whether the token sale lands, and whether those confidence scores deserve your trust) are what the next few months will answer. The 500 free calls are a cheap way to start forming your own opinion.

FAQ

What is Dataline?

Dataline is a decentralized data network for AI agents, live at dataline.xyz: one API aggregating roughly 50 live data sources (crypto exchanges, prediction markets and chains), attaching an AI-generated confidence score to every response so agents can judge data reliability before acting. It raised $10.6 million in a seed round announced October 31, 2026.

How much does Dataline cost?

Pay-as-you-go costs $0.25 (about £0.19 / C$0.35) per successful request via x402, with no monthly commitment. Pro is $299 a month for 5,000 requests, and Elite is $499 a month for unlimited requests under a fair-use policy. Foreign-currency figures are approximations at November 1, 2026 market rates.

Is there a free tier?

Yes. New users get 500 free calls with no credit card required, and builders on Base and Robinhood Chain get free Dataline access under the AI Data Launch program. A $19-a-month Hobby tier mentioned in the homepage FAQ does not appear on the live pricing page, so I would not count on it.

What is the AI confidence score?

It is a versioned 0–100 score attached to every API response, weighing multi-source consistency, data freshness, coverage and source reliability. The intent is that an agent can refuse or flag low-confidence data before acting on it. The scoring system is Dataline's own mechanism, with no independent audit or public accuracy track record yet.

When is the Dataline token launching?

The company plans to launch its token by the end of 2026, with a public sale on Base through Uniswap's Continuous Clearing Auction. As of November 1, 2026, no sale terms, raise target or token allocation have been published. Sale details were expected in early November, per the co-founder's October interview with The Block.

How is Dataline different from Chainlink?

Chainlink and Pyth are oracle networks that push price data onchain for smart contracts; Dataline is built for AI agents, returning normalized data plus a confidence score through one API, with pay-per-call x402 billing and an MCP integration. That positioning is the company's own, and no independent benchmark compares the alternatives head to head.

Dataline looks like one of the more serious infrastructure bets in the AI-agent economy: a working product, a readable price list and a mechanism aimed at a real failure mode. The 500 free calls are a quick way to test whether the confidence scores deserve your trust, and whether a quarter a call beats the stack you already run.

This is general information, not financial or procurement advice. Token sale terms, pricing and availability can change; verify them on the vendor's own pages before committing.

Sources

• Dataline official site — dataline.xyz (read live, November 1, 2026)
• Dataline pricing page — dataline.xyz/pricing (read live, November 1, 2026)
• Dataline press release via GlobeNewswire, October 31, 2026
• The Block: "Dataline raises $10.6 million for decentralized data network for AI agents, plans ICO," October 21, 2026
• WEEX / Foresight News: "Crypto Company Backed by Robinhood Founder Aims to Create Data Layer for the AI Era," October 22, 2026 (updated October 30)
• x402.org — the x402 payment protocol
• FX rates: USD to GBP 0.7469, USD to CAD 1.3975 — November 1, 2026 market rates, verified at publication; GBP/CAD figures are approximations