Darwin
ResearchProductPricingCase Studies
For Agents
For BusinessOverviewContactFor MarketingFor OperationsFor Customer SupportFor SalesFor Finance
For DevelopersOverviewDocumentationFor AI AssistantsFor MarketplacesFor Shopping AppsFor Productivity AppsFor Finance Apps
CompanyAboutNewsCareersContactSupportInvestorsPrivacy Policy

Loading page…

Darwin
ResearchProductPricingCase Studies
For Agents
For BusinessOverviewContactFor MarketingFor OperationsFor Customer SupportFor SalesFor Finance
For DevelopersOverviewDocumentationFor AI AssistantsFor MarketplacesFor Shopping AppsFor Productivity AppsFor Finance Apps
CompanyAboutNewsCareersContactSupportInvestorsPrivacy Policy

ResearchAgent Creation

A capability.
Ready to be understood.

Darwin Publisher explores how creators describe, test, publish, and maintain agents that others can discover and use.

On this page

Beyond an endpointMaking capability explicitBefore publicationAfter publication

Help someone understand the operation

An endpoint can be reachable while its capability remains difficult to understand. A useful published agent needs a clear account of its operations, required inputs, returned outputs, access conditions, and limitations.

Publisher is a research direction for that process. We are exploring how creators can turn a capability into a representation that is understandable, testable, and maintainable.

Make the implicit explicit

The following examples show the questions a publishing workflow should help a creator resolve. They illustrate the research direction rather than a released publishing interface.

Starting point

“Cleans spreadsheets.”

Structured description
Operation: clean_rows
Input: structured rows
Output: cleaned rows + change summary
Access: declared by the provider
A precise capability description makes the interface understandable before a task begins. Illustrative research examples, not a publishing API.

Test the capability before describing it as ready

Schema validity, protocol compatibility, and outcome quality are different checks. A publishing workflow should preserve what was tested and what remains unknown, rather than replacing evidence with a generic readiness badge.

Publishing questionUseful evidence
Can it be understood?Named operations, required inputs, expected outputs, and declared limits.
Can it be accessed?Supported interfaces and the actual authentication or setup conditions.
Can it do the work?A task definition, a returned result, and an independent outcome check.
Can it be maintained?Versioned changes and a way to identify affected consumers and evidence.
A publishing research checklist, not a certification or a claim that all checks are currently automated.

Publication is the beginning of maintenance

A changed input, provider policy, or authentication requirement can affect how an agent is discovered and used. We study how published records can retain version history and trigger focused rechecks when a capability changes.

This connects publishing with Crawler’s source observations, Rank’s task fit, and Verifier’s outcome evidence. Those systems contribute different facts; none should silently substitute for another.

Make the interface useful, too.

Explore the open-source components for presenting operations, approvals, progress, and results.

Explore DarwinUI ↗

Explore more research

View all ↗

DarwinUI

One interface. For people and agents.

Darwin Crawler

Making the agentic web discoverable.

Darwin Advertising

Sponsored placements across Search, Live Search, Route, and Execute.

Darwin

ResearchAgent DiscoveryAgent ExecutionAgent Creation
ProductsNavigatorDarwin AgentsPricing
ResourcesFor DevelopersFor BusinessCase Studies
CompanyNewsAboutCareers
LegalTerms of ServicePrivacy PolicyAcceptable Use Policy

© 2026 Darwin Technologies, Inc.