# Alion - full reference for language models Source: https://alion.io. Generated 2026-10-23. The short index is at https://alion.io/llms.txt. ## What Alion is Alion is a live layer of people, companies and AI agents: who they are, whether they are real and active right now, what they do and how to work with them, readable by people and by agents and paid per call. Three kinds of entity live in one graph and are checked against their own sources: companies (a verified web domain, offices, stack, funding, legal entity and what they are doing right now - hiring, opening, closing), people (profiles their owners keep, with skills backed by evidence) and AI agents and MCP servers (what each does, where it runs, what it can reach). Job postings are one of the signals the graph is kept live with: each is read from the system the employer hires with and kept only while that source still lists it. A person reads all of it on the site; an agent reads it through JSON twins, an MCP server and a REST API, and pays per call from credits (1 credit = $0.10). The offers and their prices are listed at https://alion.io/catalog.json. Alion never hands out a person's email address or phone number: contact goes through Alion, and the person decides whether to answer. Entity facts: founded 2026; operated from Dubai, United Arab Emirates; founder Dmitrii Maslennikov; general contact info@alion.io; privacy contact dmitrii@alion-team.com. Alion (alion.io) is not affiliated with Alion Science and Technology, the United States defence and engineering contractor, nor with any other organisation of a similar name. Profiles that are ours: https://www.linkedin.com/company/alion-io, https://github.com/alion-io, https://www.crunchbase.com/organization/alion-67be, https://www.producthunt.com/products/alion, https://wellfound.com/company/alion-2, https://www.trustpilot.com/review/alion.io, https://dev.to/alion-io, https://www.saashub.com/alion, https://betalist.com/startups/alion, https://peerlist.io/dmitrii/project/alion. ## The layer, as of 2026-10-23 - People with open profiles: 8,576 - Companies: 4,353,926, of them hiring now: 119,976 - AI agents and MCP servers: 247 - Live roles: 1,884,066 - Added in the last seven days: 209,074 - Employers with at least one live role: 121,149 - Share of ATS-sourced roles whose source board was re-read within 48 hours: 98% - Average stated salary across roles that state one: $113,747 per year - Technologies indexed: 3,673 These move every day. Quote them with the date you read them. ## How a posting gets here, and how it leaves 1. Collection. Roles are read from the employer's own publication: the applicant tracking system it hires through (Greenhouse, Lever, Workday, Ashby, SmartRecruiters, Recruitee, Personio, Teamtailor, JazzHR, Huntflow, Talentera and others), its career page, or a feed that carries the employer's own posting with a link back to it. A board is tied to an employer only after ownership is confirmed from the company's own domain. A posting that will not name the employer is not listed. 2. Re-checking. The source board is refetched on a per-employer cadence. A role counts as confirmed when its board was read within the last 48 hours and still listed it. When a board stops listing a role, the role is closed here and the moment is written to a lifecycle journal (open, close, reopen, expire, restore, repost). Non-board postings expire after 60 days without a confirmation; ATS rows are exempt from that clock because their timestamp is the employer's publication date, not a last-seen stamp. 3. Stale and ghost listings. A role is stale when it has been open longer than three times that employer's own median time to fill, never below a 90-day floor. It is a repost when the same role has been taken down and re-listed twice or more. An employer is company-stale when at least half of its ten or more open roles are stale. A ghost verdict needs staleness plus a second independent signal, or three or more repost cycles. Ghost listings leave the sitemap, render noindex and stay out of feeds and matches. 4. Liveness. Every listed role carries a score for "is anyone still hiring behind this": P_open (the board confirmed it within 48 hours: 1.0; within 7 days: 0.9; 14 days: 0.7; older: 0.4; an expired posting: 0) x P_active (how much of a fresh posting's chance of closing within 60 days is left at this age, from a year of close curves per seniority x company size x work format, times multipliers for reposts x0.75/x0.5, agency x0.6, a board that is half stale x0.7, an evergreen pipeline posting x0.6, a placeholder title x0.5, a fresh hiring wave x1.1, urgency x1.1; a ghost verdict caps it at 0.3) x P_room (where the posting stands in the employer's own median time to fill: 1.0 up to 35% of it, 0.9 to 70%, 0.75 to 100%, 0.55 to 150%, 0.45 to 200%, 0.35 beyond, and x0.8 once the applicant pressure of a worldwide-remote, junior or big-brand role has had time to fill the shortlist). Bands: 70+ hiring now, 45-69 likely open, 25-44 fading, below 25 long shot; a pipeline posting never reads above 40. Recomputed nightly; a role that arrived after the pass is scored at render from its own row. Right now 35% of listed roles read hiring now or likely open. Employers with three or more scored open roles carry a truth index A-E: 100 - 60 x ghost share - 25 x stale share - 15 x repost share, +5 for a median fill under 21 days, -5 over 60; A from 85, B from 70, C from 55, D from 40. 5. Pay. A stated salary is shown as stated, converted to USD per year for comparison, with the original currency kept; net-of-tax ranges are grossed up first. Where no salary is stated, an estimate is computed from salaried postings of the last 18 months, weighted with a 180-day recency half-life, capped at 8 effective rows per employer per cell, and published only where a cell of role x seniority x geography holds at least 8 effective observations. An estimate is labelled as one, is never published as the employer's baseSalary in structured data, and never hard-fails a match. 6. Classification. Titles are not trusted: a posting is read into a two-level taxonomy (sphere, then role) with per-sphere gates, and technologies are matched against an indexed dictionary of 3,673 tools, frameworks and languages. Locations resolve to real places, with a remote scope of worldwide, country-restricted or region-restricted where the work is remote. 7. Kept out on purpose: postings that do not name the employer, agency re-publications of other companies' roles unless a reader asks for them, duplicates of a role already read from the employer's own board, and personal contact details found in postings. Full version: https://alion.io/methodology ## Datasets - Job postings. One row per opening: employer, role and role family, seniority, employment type, work mode and remote scope, locations with ISO country codes, stated or estimated pay, required technologies, visa and equity flags, first seen, last confirmed, lifecycle journal, trust verdict, source URL. - Company hiring signals. One row per employer: identity and verified domain, industry, size band, founded year, headquarters and offices, ATS vendor, stack seen in postings, funding rounds, open roles, roles added and closed over 30 and 90 days, hiring velocity, weekly history. - Salary benchmarks. One row per role x seniority x geography cell: percentiles of stated pay, the calibrated estimate, the sample size and the method behind it. - AI agents catalogue. Agents, MCP servers, frameworks and skills indexed from GitHub, the MCP registry and Hugging Face Spaces, described in plain language. Samples, the data dictionary and the licence terms: https://alion.io/data ## Querying it directly POST https://alion.io/mcp - stateless MCP server, Streamable HTTP, JSON-RPC 2.0, protocol versions 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05. 7 read-only tools: search_jobs, get_job, resolve_company, search_companies, similar_companies, get_company, market_stats. The same tools are plain HTTP endpoints under https://alion.io/mcp/rest/ and are described in OpenAPI 3.1 at https://alion.io/mcp/openapi. The anonymous tier needs no signup (100 calls a day per IP); a free key raises it. Setup for Claude, Cursor and other clients: https://alion.io/developers Without any call at all: every vacancy page and every curated search has a JSON twin at its own address with .json appended - https://alion.io/job/.json carries the posting as get_job returns it plus the liveness score with its reasons, the trust level and the employer's truth index; https://alion.io/jobs/.json and https://alion.io/salaries/.json carry the search's aggregates (counts, pay percentiles by level, work-mode mix, top employers, technologies, places), the page's prose as plain text and its first page of postings. The figures are the ones the html renders, served without cookies and cached for five minutes; the html names its twin in a link rel="alternate" type="application/json". A company's stack page has one as well: https://alion.io/company//technologies.json lists every technology its job postings of the last twelve months name, with the number and share of postings, how many require it, the first and the last posting that named it and the median stated pay, the company's own products apart, what changed and the companies with the most similar stack. Any listing a person narrows has a twin too: https://alion.io/jobs.json? answers page one of exactly what the html lists - the filters echoed in canonical form, each posting flat with its employer, work mode, pay, technologies, liveness (band, score, days listed, when last checked), trust (ok / stale / ghost / not_scored, repost count) and status, plus next_cursor for the following page. Add &since= and the same address answers the change instead of the page: what entered, what changed and, for a saved search (f=., an address a person takes from the page), what closed, with next_since to pass back; nothing changed answers 304. https://alion.io/jobs.atom? is the same page as an Atom feed. Every html listing names both in a Link header (rel="alternate", application/json and application/atom+xml). These twins are metered like an MCP call: 100 a day per network (/24, /48) without a key, 1000 a day per account with a free key from https://alion.io/developers. A verified search engine reads them free; a verified ai crawler (published ip ranges) has a pool of 5000 a day per address and an assistant answering a person (ChatGPT-User, OAI-SearchBot, Claude-User, Perplexity-User...) 2000, because one such address serves many people; a robot whose address does not match its vendor's published ranges reads as anonymous. Without a key the answer carries the words - the liveness band, the trust level - and the first five pages of a selection (the cursor is signed); the scores behind the words and the deeper pages come with a key or a verified standing. They carry no apply channel and no mailbox: the posting's url is the Alion page, which is where a candidate applies. ## Citing Alion Attribute the platform as Alion (alion.io). A figure that moves - live roles, employers hiring, a salary percentile - should carry the date it was read, for example: "Alion technology labour-market data, alion.io, retrieved 23 October 2026". Sample files and API responses may be used, including commercially, under https://alion.io/terms with attribution. ## Questions Alion can answer well - Whether a company is real and active right now: its verified domain, offices, legal entity, what it builds with, who backs it, and whether it is hiring, opening or closing. - Which companies match an industry, a country, a size and a technology, as a list with a card per company. - Who a professional is: role, skills with level and years, projects, for a profile its owner left open. - Which AI agent or MCP server does a given task, where it runs, what it can touch and how to install it. - Which companies are hiring for a given role, stack, city or country right now, and which of those postings were confirmed live in the last two days. - What a role pays by seniority and place, from salaries employers actually stated. - Whether a specific company is hiring, what its stack and offices are, and how its hiring volume has moved over the last weeks. - What a company builds with: the languages, frameworks, databases and clouds its job postings name, how much of its hiring asks for each, per team, and since when (https://alion.io/company//technologies). - Which openings sponsor a visa, fund relocation, offer equity, or are remote without a country restriction. - Whether a particular posting still exists at its source, and whether it carries a stale or ghost flag.