<!-- Canonical: https://www.agentlist.io/articles/every-ai-agent-one-endpoint -->

ARTICLE 01 · API · agentlist.io · 2026-09-25 · 6 min read

# Every AI agent on one endpoint

The catalog is data, not just pages — one JSON response carries every system, its spec fields, and its last change.

Every page on this site renders from the same normalized catalog — and the catalog itself is public. One GET against `/api/v1/systems` returns all 112 systems, 89 vendors' names resolved, each record carrying the spec fields the site normalizes: category, interfaces, hosting, pricing, status. No auth, CORS open, cached at the edge for a day.

```
curl -s https://www.agentlist.io/api/v1/systems | jq '.meta.counts'
# → { "systems": 96, "vendors": 77, "categories": 14 }
```

## What a record looks like

Each entry in `systems[]` is one catalog record — the same shape whether it's a VS Code fork or a phone-answering agent:

**slug / name / title**: Stable identifier, seed name, display name — the slug is the spec-page URL.

**category**: One of the fifteen catalog categories (ai-ide, digital-employee, agent-communication, agent-services, …).

**vendor**: The organization behind the system, resolved from its vendor id.

**interfaces**: How you drive it: editor, cli, api, chat, browser, voice.

**hosting**: local, cloud, or both.

**pricing**: List price as verified on the last catalog pass — hedge words ("verify") mark stale figures.

**status**: active, active-caution, or deprecated — caution flags stale repos, removed tiers, pivots.

**aliases**: Other names the system is known by (renames, codenames, upstream project names).

**summary / sources**: The one-paragraph editorial summary plus the URLs it was checked against.

**lastChanged**: The system's most recent changelog entry, or null if it's never moved.

**url**: Canonical spec-sheet URL on agentlist.io — link here, don't copy data.

## Useful slices

The endpoint is deliberately flat — filter client-side. A few slices that matter:

```
# every digital employee
curl -s https://www.agentlist.io/api/v1/systems \
  | jq '[.systems[] | select(.category=="digital-employee") | .title]'

# open-source agents with GitHub repos
curl -s https://www.agentlist.io/api/v1/systems \
  | jq '[.systems[] | select(.sources[]? | test("github.com")) | {title, sources}]'

# systems flagged for caution (stale repos, removed tiers)
curl -s https://www.agentlist.io/api/v1/systems \
  | jq '[.systems[] | select(.status=="active-caution") | .title]'
```

In an app, it's one fetch — the payload is a few hundred KB and changes at catalog-cadence, not request cadence:

```
const { systems } = await fetch("https://www.agentlist.io/api/v1/systems").then(r => r.json())
const comms = systems.filter(s => s.category === "agent-communication")
```

## Tracking what changed

Two feeds sit next to the endpoint for diff-tracking. `/api/v1/changelog` returns the structured changelog (adds, repricings, cautions, retirements), and [/feed.xml](https://www.agentlist.io/feed.xml) is the same stream as RSS for reader subscriptions. Each system record also carries `lastChanged` — its latest changelog entry — so a sync pass can skip records that haven't moved.

Verified-as-of is the first field in `meta`: `generatedAt` is the seed's verification date (currently 2026-09-25), not the deploy timestamp.

## Etiquette

Cache it — the response is built to be stored, and `s-maxage=86400` means the edge serves a day-old copy anyway. Don't scrape the HTML pages for the same data; the JSON is canonical and cheaper for everyone. When you publish against it, link the `url` field — spec pages carry the caveat text and changelog context a raw record can't.

---

Source: [Every AI agent on one endpoint | agentlist.io](https://www.agentlist.io/articles/every-ai-agent-one-endpoint). This is the public page rendered as Markdown; interactive controls require the website.