Get started

Query the index

Read the settlement record over REST, read-only SQL, or the pre-computed rollups, and know which to reach for.

Everything Roundhouse knows is readable — the same public dataset backs every tier. Reading the settlement index over the API takes a key; discovery does not: the capability layer (/v0/unified*) and the service directory (/v0/endpoints) stay open with nothing configured, and the website itself is browsable by anyone.

Pick a read path by how specific your question is.

Your questionUseKey
"Who sells X?"GET /v0/endpoints?q=X, GET /v0/unifiedNone
"What is happening right now?"GET /v0/flowsRequired
"Tell me about this wallet"GET /v0/agents/<w>, GET /v0/merchants/<w>Required
"Show me the payment graph"GET /v0/graphRequired
Anything with a group by in itPOST /v0/sql$0.005/query (x402) or paid-plan org key
Natural language, exploratoryAI playgroundQuery Units

A key is one request away: GET /v0/test/x402 mints a personal 30-day key for one cent over x402, and organisations create full-rate keys in the dashboard.

Fixed endpoints#

JSON over HTTP; send your key as authorization: Bearer rh_live_…. The full list with every parameter is the API reference; these are the ones you will reach for first.

http
GET /v0/flows?limit=50                    live flow of funds, newest first
GET /v0/transactions?limit=50             the global settlement feed
GET /v0/agents?q=<search>                 indexed ERC-8004 agents
GET /v0/agents/<wallet>                   one agent: identity + settlement stats
GET /v0/merchants/<wallet>                one merchant: inbound volume, customers
GET /v0/entities/<wallet>/settlements     raw settlements for one wallet
GET /v0/endpoints?q=<search>              the priced service directory
GET /v0/leaderboard?limit=25              agents ranked by trust
GET /v0/facilitators                      who relays payments, and how much
GET /v0/graph?days=7                      nodes and payer→payee edges

Two paging conventions, and they are not interchangeable:

  • Feeds use a cursor. next_before is an opaque token encoding the whole sort key including its tiebreaker. Pass it back as ?before=. Do not parse it, and do not construct one — a naive block_time cursor silently drops every row sharing the last row's timestamp, and on a 2-second chain most timestamps are shared.
  • The leaderboard uses an offset. A rank is a position, and score is nullable, so there is nothing stable for a cursor to compare against. ?offset= in, next_offset out.

An empty array is a real answer. A failed read is not: a rejected query returns 503 upstream_unavailable naming the resource, never 200 {"flows": []}. On a settlement index "no rows" reads as "nothing ever happened", which is the one lie the read path is built to prevent.

Read-only SQL#

For everything else. One statement, SELECT or WITH only, over the public data-layer tables.

bash
curl -sL -X POST "https://api.roundhouseai.io/v0/sql" \
  -H "authorization: Bearer $ROUNDHOUSE_KEY" \
  -H 'content-type: application/json' \
  -d '{
    "sql": "select date_trunc('day', block_time) as day, count(*) as settlements, sum(amount_usd) as usd from settlements where block_time > now() - interval '14 days' group by 1 order by 1 desc"
  }' | jq

The sandbox is enforced in the database, not in the request handler: a role that can read only the public tables, a read-only transaction, a statement timeout set by the tier you bought (15 seconds by default), and a hard 300-row cap. Write for those limits rather than discovering them — SQL over the index has the readable tables, the useful shapes, and the traps.

Use the rollups, not the raw table#

settlements has tens of millions of rows. An unbounded aggregate over it will hit the timeout. The pre-computed views exist so you do not have to:

ViewGrainGood for
mv_entity_rollupsone row per walletlifetime inbound/outbound volume and counts
mv_entity_dailywallet × dayan activity trend for one counterparty
mv_global_dailydaythe site-wide series behind /stats

They refresh on a schedule, so they trail the tip of the chain by up to an hour. Read surfaces fold in the newer settlements at read time, which is why a profile page and a raw mv_entity_rollups query can differ by a few rows — the page is more current, not wrong.

Rate limits and keys#

The API is keyed. Anonymous access covers exactly two surfaces — the Unified Services endpoints (/v0/unified*, no limit) and the service directory (/v0/endpoints, 30/min per IP) — so an agent can always discover and price services with nothing configured. Everything else answers 401 api_key_required without a key. Limits are per endpoint class: aggregate (graph, leaderboard, entity profiles, per-entity history, SQL) and feeds.

CallerAggregate / minuteFeeds / minuteSQL
Anonymous— (key required)— (key required)$0.005/query (x402)
Trial key from /v0/test/x4022 per key, for 30 days2 per key$0.005/query (x402)
Trial key, registered identity (ERC-8004 or KYA)10 per key, for 30 days10 per key$0.005/query (x402)
Organisation key, paid planplan rate per orgplan rate per org1 QU per query

SQL is not included in the free plan or in the trial identity key — for those callers every query is its own $0.005 x402 payment.

Send a key as authorization: Bearer rh_live_… or x-api-key. Organisation keys are created and managed in the dashboard; the trial key costs one cent, is bought by the paying wallet itself, and needs no account at all.

Next steps#