_done

_done is a collection of APIs that handle real-world tasks AI agents cannot perform on their own. When a language model needs to check DNS records, take a screenshot of a webpage, scan a port, or make an actual HTTP request, it hits a wall. These are operations that require deterministic execution, not probabilistic generation. _done provides 26 live endpoints that fill those gaps, letting agents complete tasks that involve interacting with external systems, verifying data, or producing outputs that a model simply cannot fake. If you are building agents that need to touch the real world, _done is infrastructure for that boundary.

The API catalog spans several categories. Data and lookup services include DNS and WHOIS lookups, domain age checks, ASN lookups, ENS resolution for Ethereum Name Service addresses, and DMARC record verification. These are the kinds of queries where you need an actual response from an external system, not a model's best guess based on training data. A model can tell you what a DNS record probably looks like, but it cannot actually query a nameserver. _done does the query and returns the real answer. Validation tools cover BIMI checking, CORS header verification, HTTP status and header inspection, and domain availability using the RDAP protocol. Each endpoint returns real data from real servers, which matters when accuracy is not optional.

Technical operations include a curl service that makes real HTTP requests from the cloud and returns status codes, headers, and body content. This is useful when an agent needs to verify that a URL is live, check what a server is returning, or fetch content that is not in its training data. TCP port scanning lets agents check whether specific ports are open on a target host, which is relevant for security checks or infrastructure monitoring. OCR text extraction pulls text from images, which is another operation models cannot perform directly without external tooling. Full-page screenshots render webpages completely and return the image, capturing JavaScript-driven content that would not appear in a simple HTML fetch. If you need to show what a page looks like rather than just describe it, this is how.

Content processing endpoints handle JSON validation, repair, and manipulation. If an agent produces malformed JSON or needs to transform a structure, these endpoints provide deterministic operations rather than model-generated fixes that might introduce new errors. Sitemap extraction and SEO data extraction help agents gather structured data from websites without parsing HTML themselves. There is even a leetspeak translation endpoint for cases where you need to decode or encode that format reliably. Specialized calculators cover mortgage amortization with period-by-period breakdowns and loan scenario modeling, giving agents access to precise financial math rather than approximations.

E-commerce and news endpoints include Shopify AI rank checking and Hacker News data fetching. Utilities round out the set with QR code generation and cryptographic hashing with HMAC support. The overall pattern is operations where you need a real result, not an approximation, and where the operation cannot be performed inside the model itself. These are the blind spots that trip up agent architectures, and _done provides endpoints that cover the most common ones.

What makes _done unusual is the payment model. There are no accounts, no API keys, and no logins. Instead, agents receive OpenAPI specifications, encounter an HTTP 402 Payment Required response, sign a USDC transaction on Base or Solana, and receive results once the payment verifies. Each operation costs one cent with response times under two seconds. This micropayment approach removes the friction of managing credentials and usage tracking while letting agents pay for exactly what they use. It assumes agents can handle cryptocurrency transactions, which is a specific architectural choice that fits some stacks better than others.

The target audience is developers building AI agents that need to interact with the real world. If your agent architecture calls for external verification, network diagnostics, content extraction, or any operation that requires deterministic execution, _done provides endpoints that return reliable results rather than model guesses. The design assumes agents will call these APIs programmatically, though humans can invoke them directly for testing or one-off tasks. The tradeoff is the cryptocurrency payment requirement, which fits naturally for teams already operating in that ecosystem but adds a dependency for everyone else. If your agents need to do things models cannot do alone, _done is built for that gap and priced at a level where experimenting is cheap.