Underscore launches 20+ APIs for AI agents and developers

The _done service is a collection of utility APIs for AI agents and developers, released under a product name _done. The suite includes over 20 endpoints spanning networking, security, data lookup, and text processing.

Networking and infrastructure tools include ASN Lookup (returns the network operator, country, and address block behind an IP address or ASN number); CORS Header Checker (sends a real request with an Origin header and reads back all six standard cross-origin permission headers); Curl HTTP Request (makes a real HTTP request from cloud servers and returns status code, headers, and body, supporting GET, POST, PUT, PATCH, DELETE, and HEAD, with automatic redirect following up to 3 levels); DNS & WHOIS Lookup (returns all 13 DNS record types plus full WHOIS registration data for up to 10 domains in one call); Domain Age Checker (returns registration date, expiration date, age, and time remaining for domain names); Domain Availability Checker (tells whether a domain has a registration record using the official RDAP protocol, returning 'registered', 'unregistered', or 'indeterminate'); ENS Resolver (converts an ENS name like vitalik.eth to an Ethereum address, or vice versa); Hacker News Data (fetches live Hacker News stories by category or scans all stories, jobs, and polls from a specific calendar date, with optional nested comments); HTTP Header Checker (fetches a web address and returns response headers, status code, and redirect path); and HTTP Status Checker (checks up to 10 web addresses and returns exact status numbers and redirect information). TCP Port Scanner sends a hostname or IP address and port range, and returns which ports are open and what services are running.

Security and cryptography APIs include Hashing, HMAC & Checksum Suite, which computes SHA-256, SHA-512, MD5, BLAKE2, CRC32 and other hash algorithms, signs data with HMAC, and verifies signatures using timing-safe comparison to prevent secret-leaking attacks. The article specifically notes that language models cannot perform hashing—they produce a plausible string that is simply wrong, making every webhook verification and integrity check that relies on a hallucinated hash silently broken.

Validation and data processing tools include Bulk DMARC Record Lookup (checks email security configuration for one or many domains in a single request, returning whether a valid record exists, the policy, and reporting address information); JSON Suite (validates JSON with exact error location, fixes near-JSON with typos, pretty-prints or minifies, extracts values using path expressions, compares two JSON documents, converts to and from YAML or CSV, produces byte-stable canonical form for hashing, validates against a schema, and provides size and structure statistics); Leetspeak Translator (converts regular text to leetspeak or vice versa); Mortgage Amortization (calculates monthly payments and complete amortization schedules given loan amount, interest rate, and repayment term, supporting extra payments and lump sums, with results pinned to identical cent-level accuracy across systems); OCR (sends a picture as a public web address or base64-encoded image and returns plain text recognized in it, with confidence scores, supporting English text only); and Bulk QR Code Generator (converts a list of texts, URLs, or strings into QR code images in a single call, returning each code as base64-encoded data).

CPI Report returns official United States Consumer Price Index inflation numbers from the government—the newest reading, a specific past month, or a range of months—with clear percentages for overall and core inflation plus plain-English summary. The article emphasizes that two correct systems running the same inputs produce identical results, and that the same input always gives the same output with no side effects. All APIs are accessed via HTTPS POST endpoints (e.g., https://asn-lookup.underscoredone.com/lookup, https://hash-hmac.underscoredone.com/hash).

The _done suite addresses a fundamental limitation in how AI systems operate: large language models excel at generating human-like text but are unreliable at deterministic, verifiable tasks. When an AI model is asked to compute a cryptographic hash or check whether a domain is registered, it cannot actually perform those operations—it can only produce text that resembles the correct answer. In production systems, this leads to silent failures, broken security checks, and corrupted data integrity. By packaging 20+ utility APIs under a single service, _done offers AI agents and developers a way to outsource these operations to reliable endpoints that always return the same result for the same input.

The APIs span three broad categories: network and domain infrastructure (ASN lookup, DNS/WHOIS, domain age, domain availability, ENS resolution, port scanning, HTTP headers); security and cryptography (CORS checking, hashing and HMAC verification); and data fetching (Hacker News scraping, OCR from images, HTTP requests). Several APIs explicitly acknowledge the hallucination problem—the JSON suite handles the common case where an AI model's JSON output is slightly malformed, and the hash-hmac tool uses timing-safe comparison to prevent secret-leaking attacks that might occur if an untrusted signature check were left to a language model.

The architecture appears designed for integration into AI agent workflows: each API accepts structured input (a domain name, an IP address, a JSON string) and returns deterministic output (status codes, record data, parsed text). Bulk operations reduce the number of requests needed to process many items at once. This lowers latency and cost for applications that rely on AI agents to orchestrate multiple data lookups or network checks as part of a larger task.

No comments yet. Be the first to share your thoughts!