Parallel Web Systems Lets AI Agents Query Crunchbase and PubMed Directly
Parallel now lets agents pull from specialized third-party sources like Crunchbase, Polymarket, and PubMed through three flexible access models.
- Parallel launched Data Connectors, giving agents access to specialized third-party data inside Task and Responses APIs.
- Three access models: default Index Partners, Pay-Per-Use invocation, and Bring Your Own License via MCP servers.
- Launch partners include Crunchbase, Apollo, Polymarket, SimilarWeb, Harmonic, Middesk, Allium, and Carbon Arc.
- Free biomedical connectors include PubMed, ClinicalTrials.gov, ChEMBL, bioRxiv, medRxiv, CMS Coverage, and NPI Registry.
- Index Partners earn revenue through a Shapley-based attribution model tied to answer contribution.
- Available in Responses (Medium effort and above) and Task API playgrounds with per-task toggles.
Parallel adds licensed data sources to its agent research APIs
Parallel Web Systems has launched Data Connectors, which let agents using its Task and Responses APIs query third-party data alongside the open web. Launch sources include Crunchbase, Apollo, Polymarket, PubMed, and ClinicalTrials.gov.
The connectors address a limitation of general web search: many sales, finance, and life sciences workflows depend on structured databases that require subscriptions, credentials, or specialized query tools. Parallel now routes those sources through the same research layer used for web results.
One research layer, three access models
| Model | How it works | Billing |
|---|---|---|
| Index Partners | Partner content is available automatically through the Parallel Index. Developers submit a normal research request, and Parallel retrieves relevant partner data. | Parallel pays providers through a Shapley-based attribution model. |
| Pay-Per-Use | Developers select a provider for a request under advanced_settings.data_sources. Parallel queries it alongside its standard web research. |
Customers pay for successful invocations. |
| Bring Your Own License | Customers connect an existing subscription by passing the provider’s remote Model Context Protocol server and credentials under mcp_servers. MCP is a standard interface for exposing tools and data to agents. |
The customer keeps its existing provider contract. |
The Index payout model uses Shapley values, a cooperative game theory method that estimates each participant’s marginal contribution to a result. Parallel applies that method to determine how much value a provider’s content added and allocates compensation accordingly.
The first connector catalog
The initial catalog concentrates on company intelligence, financial research, sales enrichment, retail analytics, and life sciences.
| Access model | Provider | Data |
|---|---|---|
| Index Partner | Particle | AI-native news summaries |
| Index Partner | Pensa Systems | Retail shelf intelligence |
| Pay-Per-Use | Baselayer | Business identity and risk intelligence |
| Pay-Per-Use | Carbon Arc | Alternative data for investors |
| Bring Your Own License | Allium | Blockchain data and analytics |
| Bring Your Own License | Apollo | Contact and company data |
| Bring Your Own License | Crunchbase | Private company and funding data |
| Bring Your Own License | Faraday | Consumer predictions |
| Bring Your Own License | Harmonic | Startup and investor intelligence |
| Bring Your Own License | Middesk | Business identity and know-your-business data |
| Bring Your Own License | Polymarket | Real-time prediction market data |
| Bring Your Own License | Similarweb | Website traffic metrics |
Seven life sciences sources are free to invoke through the Pay-Per-Use mechanism: bioRxiv, ClinicalTrials.gov, ChEMBL, CMS Coverage, medRxiv, NPI Registry, and PubMed. Together, they cover preprints, published literature, clinical studies, drug discovery, health coverage, and provider records.
Configuration stays close to the request
- Choose providers in the connector directory.
- Add credentials or remote MCP endpoints when a provider requires them.
- Pass explicit Pay-Per-Use sources through
advanced_settings.data_sources. - Pass licensed MCP integrations through
mcp_servers. - Disable any connected source for an individual task or response when it is unnecessary.
Connected sources are available in the Task API playground and in the Responses playground at Medium effort or higher. Parallel’s developer documentation covers the request fields and provider-specific setup.
Source routing moves into the research call
Parallel’s design moves part of the orchestration layer from application code into its research system. A life sciences request can draw from PubMed, ChEMBL, and public websites, while a sales enrichment request can use existing Apollo and Crunchbase subscriptions. Developers still choose which paid or licensed sources are available, but Parallel handles retrieval within the broader research task.
The practical benefit is less custom routing code for authentication, source selection, and web research. Teams evaluating the system will still need to measure citation quality, latency, provider-specific costs, credential handling, and the visibility they receive into source selection.
The provider economics also remain an open test. The Shapley model gives Parallel a framework for compensating sources according to their contribution, while its durability will depend on whether providers consider the resulting payouts sufficient as the catalog expands.
Comments
