Search and place data with your agent

Your agent uses search for two jobs: finding information, and finding where new information belongs before it adds it. Both run on the same engine as the search box: words and meaning together (a query like “companies that pay workers same day” finds staffing apps with instant pay even when no word matches), typo-tolerant, with abbreviations understood (HR, ATS, PEO, CRM, 3PL).

1. Search#

search returns ranked hits of every kind: organizations, industries and segments, primitives, constructs, glossary terms, and with a key your own workspaces, companies and components. Each hit has a stable id, its kind, a url, a score, and best: true on the one result the query clearly names. Organizations add ref, domain, coverage and trust (owner, confirmed or inferred); primitives add their key and part; industries their key and taxonomy path. matched names the public component that made an organization match.

curl "https://systematlas.org/api/v1/search?q=same+day+pay&limit=1"

{
  "query": "companies that pay workers same day",
  "total": 24,
  "understood": {
    "industry": null,
    "place": null,
    "kinds": [
      "company",
      "my_company"
    ],
    "semantic": true
  },
  "correction": null,
  "related": [
    "companies that pay workers same day industry:staffing"
  ],
  "hits": [
    {
      "kind": "company",
      "id": "01c7256e-8d85-4702-8580-084c7e15085a",
      "best": false,
      "title": "Toothio",
      "subtitle": "Staffing · US",
      "industry": "staffing",
      "url": "https://systematlas.org/registry/toothio-01c7256e",
      "score": 0.79,
      "matched": "Worker earnings and payout",
      "trust": "confirmed",
      "ref": "toothio-01c7256e",
      "coverage": 0.25,
      "claimed": false
    }
  ]
}

Narrow with filters in the query or as parameters: industry:staffing, region:US, layer:software, type:company,primitive, claimed:yes. The response also says what the engine understood (understood: an industry it recognised, a place such as “in the UK” that it ranks first, synonyms it used), a spelling correction when it made one, and related searches.

2. Autocomplete#

With mode: "suggest" the same tool completes a partial query instead: likely whole queries and the named things it starts, typo-tolerant from the first letters. Cheap enough to call while an agent is still deciding what to ask.

{
  "name": "search",
  "arguments": {
    "query": "payroll i",
    "mode": "suggest"
  }
}

3. Resolve before you add#

Before uploading an organization, a component or a primitive, ask resolve whether it already exists. Give what you know: an organization's name and website, a primitive's label and a sentence on what it does (and the industry), an industry's name or description. You get the top candidates, each with a confidence from 0 to 1 and the signals behind it (name, meaning, same domain).

curl "https://systematlas.org/api/v1/resolve?type=primitive&name=Instant+pay&description=Workers+cash+out+earnings+the+same+day&industry=staffing"

{
  "best": {
    "kind": "company",
    "id": "021b990f-50f0-4fad-af2f-917485c4260a",
    "name": "Stripe",
    "url": "https://systematlas.org/registry/stripe-021b990f",
    "ref": "stripe-021b990f",
    "path": [
      "Software & SaaS",
      "B2B SaaS"
    ],
    "domain": "stripe.com",
    "confidence": 0.98,
    "same": true,
    "signals": {
      "name": 0.64,
      "meaning": 0.59,
      "domain": true
    },
    "trust": "confirmed"
  },
  "candidates": [
    "the best candidate first, then up to `limit` more, e.g. Designstripe at confidence 0.25"
  ]
}

How to act on it:

• same: true (confidence 0.85 or more): attach to that node. Use its id, registry ref or primitive key in the upload (for example primitive in upload_components).
• 0.6 to 0.85: probably the same thing under another name. Check the candidate's page, or ask the person you work for.
• Below 0.6: it is new. Add it as a custom component, or propose a primitive with sources.

What search can see#

Without a key: public registry data only. With a key: also the workspaces your key's account belongs to, with the same tiers as in the app (private components only for owners and the consultant). The meaning index follows the same rule: public documents are built from public text only, and a workspace's own documents are matched only for people who can read them.