Dataline Raises $10.6M to Build an Agent-Native Data Marketplace on x402 Micropayments

A two-sided marketplace designed specifically for AI agents to pay for data without API keys or billing accounts has raised $10.6 million in seed funding, positioning itself as infrastructure for an emerging economy where machines transact with machines at the speed of HTTP requests. Dataline, a BVI-registered startup that rebranded from Tearline in May 2026, publicly disclosed the round on October 21, 2026, with Lattice and Samsung-backed Dispersion Capital co-leading and additional participation from Symbolic Capital, Coinbase Ventures, Anagram, and Robinhood CEO Vlad Tenev. The raise marks the first significant capital deployment into what the company calls "agent-native data infrastructure" — a category distinct from both traditional oracle networks and AI training-data markets.

What Dataline Actually Does: A Routing Layer With Two Customers

Dataline operates as an aggregation and routing layer rather than a first-party data source. On one side, data publishers post structured feeds — price data, on-chain analytics, prediction-market probabilities, financial feeds — using normalized schemas that Dataline unifies across both on-chain and off-chain sources. On the other side, AI agents call a single API endpoint, receive a response, and pay per call using USDC stablecoins over the x402 micropayment standard.

The company says it currently indexes roughly 50 data sources and has processed 19.4 million on-chain transactions, with what it describes as a 96.4% execution success rate. Both figures are company-reported and have not been independently verified. Launch partners include Hyperliquid, Polymarket, Kalshi, Base, Robinhood Chain, BNB Chain, Sui, and TON — active production platforms in DeFi and prediction markets, which lend credibility to the use case even if the transaction figures remain unaudited. Builders on Base receive free API access.

The x402 Payment Standard: Accounts-Free Commerce for Agents

The payment mechanism behind Dataline's per-call economics is x402, an open standard originally proposed in the HTTP specification in 1991 as a "Payment Required" response code but never implemented. Coinbase revived it in May 2025, contributed it to what became the Linux Foundation's x402 Foundation in April 2026, and the x402 Foundation reports that the protocol had processed more than 165 million transactions across compatible endpoints by the time of its formal launch.

The x402 flow works like this: an agent calls a Dataline endpoint; the server responds with HTTP 402 and payment instructions (a USDC amount and a destination wallet address); the agent's wallet signs the transaction on-chain and retries the request with a payment proof in the header; the server verifies on-chain and returns data. No account registration, no API key, no monthly billing cycle — the entire exchange is autonomous and cryptographically self-contained.

This matters for AI agents specifically because agentic systems running in planner-executor-verifier loops cannot manage human-facing billing relationships. A developer building an agent that queries real-time prediction-market data hundreds of times per session does not want to pre-negotiate API tier contracts for each data source. x402 collapses that friction to a machine-native transaction.

Confidence Scoring: A Data-Layer Quality Signal for Agent Decision Trees

Every Dataline API response includes an AI confidence score — a scalar between 0 and 1 computed from three components: data agreement (cross-source consistency), source reliability (historical publisher track record), and freshness (recency of the underlying data). The formula is deterministic rather than probabilistic, making it a checksum-style signal rather than a model-generated estimate. The scoring methodology is described by the company but has not been subject to independent audit.

For agent developers, this creates a programmable quality gate. An executor agent receiving a confidence score of 0.5 on a market price can implement threshold logic to trigger a fallback query, request human review, or abstain from action entirely. This is fundamentally different from model-level confidence (a probability distribution over next-token predictions) and different from smart-contract oracle validity checks (which verify a single authoritative price). Dataline's confidence score is specifically designed for the agent-execution threat model: "Is this data reliable enough to act on right now, at machine speed?"

An optional zero-knowledge attestation path is available for use cases requiring cryptographic proof of data provenance. That path is opt-in — meaning the majority of agent calls receive no on-chain proof of where the data came from — which is a meaningful limitation for high-stakes applications.

Why Existing Oracle Networks Don't Solve the Agent Problem

The established oracle landscape — Chainlink, Pyth Network, and The Graph — was designed for different adversarial models than autonomous agent execution.

Chainlink operates primarily as a push oracle: data is written on-chain on a subscription schedule, and smart contracts read it synchronously. With more than $33 billion in total value secured and 505-plus protocol integrations, Chainlink is the dominant DeFi oracle. But its architecture assumes a fixed set of price feeds pre-negotiated by developers; it does not support arbitrary agent-driven queries with per-call payment, and its confidence model is binary (valid / invalid) rather than a continuous quality score agents can reason over.

Pyth Network takes a pull approach, delivering high-frequency institutional trading data on demand with sub-second price updates. Pyth is purpose-built for latency-sensitive financial applications and has approximately $3 billion in total value secured. It is moving toward paid API subscriptions — a model that still requires account registration, making it impractical for headless agents.

The Graph indexes historical blockchain data via subgraphs and reports that 37% of new Token API users are AI agents — an early signal that agent-driven data consumption is growing. But The Graph's indexing model is backward-looking; it is a historical query layer, not a real-time inference-data router.

Grass, which built a decentralized network of 2 million residential bandwidth nodes to scrape the web for AI training data, solves a different problem entirely: acquiring raw content for pretraining or fine-tuning, not supplying structured, confidence-scored signals to agents executing live decisions. The markets do not overlap.

The specific gap Dataline occupies is agent-execution data: structured, schema-normalized, on-chain and off-chain unified, queryable on demand, payable per call, and annotated with a quality signal that agent code can consume programmatically.

Token Economics and the Protocol-Layer Ambition

The DLINE token, planned for a public sale in Q4 2026 via Uniswap's Continuous Clearing Auction on Base, signals that Dataline's founders see this as infrastructure with network-effect potential, not a SaaS API business. The funding round itself was structured as SAFE notes with token warrants rather than conventional equity — a structure that gives investors both a company stake and a claim on token appreciation.

Data publishers participating in Dataline's network are expected to stake DLINE tokens as a quality bond. If a publisher's feeds consistently fail cross-source agreement checks, a portion of their stake is slashed — destroyed or redistributed. This economic stake mechanism is designed to maintain publisher quality without centralized curation, creating a self-enforcing quality market.

The token sale creates regulatory exposure that the company has not publicly addressed. Public token sales to US persons remain in a legally gray zone in multiple jurisdictions, and the SAFE-plus-token-warrant structure has attracted SEC scrutiny in analogous situations. Dataline's BVI registration does not insulate US-based participants from US securities law.

Limitations: Coverage Gaps, Unverified Metrics, and Trust Calibration

Fifty data sources is a narrow footprint for enterprise agentic workflows, which may require hundreds of specialized feeds. The company has disclosed plans to expand coverage with the new capital, and the round includes specific investment toward robotics sensor data and AI training data feeds — both categories that would significantly broaden addressable use cases.

The confidence score's reliability depends entirely on Dataline's source-reliability assessments being accurate and consistently updated. An agent trusting a 0.9 score from a publisher whose reliability history is thin or incorrectly calibrated faces a false-precision problem: high confidence in bad data. No third-party audit of the confidence-scoring system has been published.

All performance metrics — 19.4 million transactions, 96.4% execution success, 300,000 community members, 2.5 million agent interactions — are company-reported without disclosed methodology. The interaction figures in particular are plausible for early-stage API testing but should be treated as directional rather than definitive.

What Comes Next

Dataline's use of funds points toward three near-term technical bets: expanding the data source catalog to broaden coverage for enterprise agents, upgrading verification infrastructure (likely including more ZK attestation paths), and entering robotics and physical-world sensor data — a market that does not yet have a mature data-routing layer and where confidence scoring on sensor-derived inputs could become critical.

The Q4 2026 DLINE token sale will test whether the broader market agrees that agent-native data infrastructure deserves a protocol-level token with staking incentives, or whether the value proposition is better served by a simpler API business. If the token launch succeeds, Dataline gains a self-funding quality-enforcement mechanism and a developer-aligned governance token. If it fails to attract meaningful staking, the economic slashing mechanism loses its teeth and the network's quality enforcement reverts to Dataline's own curation.

The deeper test is whether agentic systems, as they scale from experimental to production deployment across enterprise pipelines, will pull their data from a specialized agent-native layer or extend existing oracle and API infrastructure to serve the new demand. Chainlink's agent-compatible product roadmap and The Graph's growing AI-agent user base suggest the incumbents are not ceding the market. Dataline's window to establish the category before larger players move may be narrower than a $10.6 million seed round implies.