feeds

Goldsky Feeds

Feeds answers one question about one wallet over HTTP: what it holds, or what it sent and received, across the supported chains in a single request. Goldsky indexes, normalizes, and prices the rows. The caller does not run a pipeline.

The same host and the same key also serve Polymarket and block-header feeds. Those are listed under "Other feeds on this key", and they do not follow the conventions below.

The installed CLI has no feeds command. Do not look for goldsky feeds, and do not mint the key with goldsky edge create. The endpoint name feeds is reserved on the Edge create path, so that command is rejected. The CLI login token is a different credential and does not authenticate these requests.

When this is the wrong tool

• The rows need to land in the user's own database or webhook. That is a Turbo pipeline (/turbo-builder, /datasets). Feeds does not stream into a sink; streaming is not offered yet (https://docs.goldsky.com/feeds/streaming).
• The question is a Turbo dataset name or chain prefix. That is /datasets. Those prefixes differ: Feeds says polygon and arbitrum_one, Turbo says matic and arbitrum.
• The question is JSON-RPC. That is /edge, or /boost when they want to keep their current provider.

Key

A project has one Feeds key, and every feed answers it. Creating that key is the setup. Do it with the API. Do not send the user to the dashboard, and do not run goldsky edge create.

The call authenticates with the project token from goldsky login, which is stored at ~/.goldsky/auth_token. Do not read that file into the chat, do not pass --token, and do not ask the user to paste a token. If goldsky project list says they are not logged in, use /auth-setup first. The caller has to be an Editor on the project.

create_raw=$(curl -sS -X POST \
  -H "Authorization: Bearer $(cat "$HOME/.goldsky/auth_token")" \
  -w '\n%{http_code}' \
  https://api.goldsky.com/api/v1/feeds/api-key)
printf '%s\n' "${create_raw##*$'\n'}"

Do not print create_raw. The line above is the HTTP status. The rest of the variable is { "data": { "name": "feeds", "api_key": "<plaintext or null>" } }.

• When api_key is a string, that call created the key. Export it as GOLDSKY_FEEDS_API_KEY. Do not print it, and do not write it into a file.
• When api_key is null, the project already has the key and this endpoint will not return it again. Reveal it, export .data.api_key, and do not print that response either. The reveal body is { "data": { "api_key": "<plaintext>" } } and has no name. Run this only in that case:

reveal_raw=$(curl -sS \
  -H "Authorization: Bearer $(cat "$HOME/.goldsky/auth_token")" \
  -w '\n%{http_code}' \
  https://api.goldsky.com/api/v1/edge/feeds/api-key)
printf '%s\n' "${reveal_raw##*$'\n'}"

The two paths differ on purpose and are not interchangeable: create is POST /api/v1/feeds/api-key with no edge/, reveal is GET /api/v1/edge/feeds/api-key with it. The other two combinations are 404. Do not "correct" either path.

Do not call POST /api/v1/feeds/key/rotate unless the user asks: rotate revokes the current key immediately.

• 401 means they are not logged in. Use /auth-setup.
• 403 means this user is not an Editor on the project.
• 409 means an Edge endpoint named feeds exists and is not the Feeds key. That endpoint has to be deleted before this call can create one.

A 401, 403, or 409 body is an error and has no key, so it can be shown. A 200 body has the key, so leave it in the variable.

Send GOLDSKY_FEEDS_API_KEY in the x-api-key header. ?key= and Authorization: Bearer are also accepted on the feed itself. Prefer the header so the key does not land in a URL log.

The codes above are the mint and reveal calls. The feed itself answers a missing key with 402 and an x402 payment-required body, not 401, because that path is also offered pay-per-request. A key that is present but wrong is 401. So on a 402, the header did not arrive; re-read it before assuming the key is bad.

Calls

Read https://edge.goldsky.com/data/openapi.json before using a parameter or response field that the quickstart does not show. That document is the contract. The guide is https://docs.goldsky.com/feeds/quickstart and the overview is https://docs.goldsky.com/feeds.

Balances returns the latest balance of each token the wallet holds, native assets included, plus total_value_usd:

curl -sS -H "x-api-key: $GOLDSKY_FEEDS_API_KEY" \
  "https://edge.goldsky.com/data/feeds/wallets/balances?address=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045"

Transfers returns that wallet's movements, newest first. direction is in or out relative to address:

curl -sS -H "x-api-key: $GOLDSKY_FEEDS_API_KEY" \
  "https://edge.goldsky.com/data/feeds/wallets/transfers?address=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045"

Both feeds can also answer 502 when ClickHouse upstream fails, and balances can answer 503 when native balances are unavailable on some chains. The 503 names them in error.chains and sets Retry-After; it is a partial outage, so retry rather than reporting the wallet as empty.

address is required and is exactly one wallet. An EVM address is 0x plus 40 hex digits. A base58 Solana address passes validation but has no chain behind it yet, so it answers 200 with an empty data. A non-200 body is an error, not an empty wallet.

Parameters

Re-read the OpenAPI spec when a request is rejected. The chain set lives on the chains parameter there.

• Omit chains to query every supported chain. Pass canonical slugs, comma-separated. matic, arbitrum, and mainnet are aliases for polygon, arbitrum_one, and ethereum. Matching is case-insensitive, and - is read as _. Any other spelling returns 400.
• token_symbol matches a symbol, so two contracts can both match — usdc returns both the native and the bridged contract on several chains at once. Use token_address when the token must be a specific contract; it takes up to 100, comma-separated. On balances, the zero address selects the native asset of every chain in scope.
• include_unknown_price defaults to false, which leaves out tokens Goldsky has no price source for. It is not a promise that every row is priced: at the default, rows whose token is priceable still come back with price_usd and value_usd null. Handle a null price on every row.
• include_historical defaults to false. Set it to true on balances to also get zero rows: tokens fully sold, and natives the wallet holds none of. On one wallet here that took 373 rows to 517. "Has this wallet ever held X" needs it; "what is it worth now" does not.
• min_value_usd drops rows below that USD value and also drops unpriced rows. It applies to total_value_usd too.
• On transfers, from and to are inclusive RFC 3339 bounds on block_timestamp. from_block and to_block require exactly one chain, because block numbers are not comparable across chains; without one they are 422 CONFLICTING_FILTERS, not 400. transfer_type is native, erc20, or both, comma-separated. The NFT types (erc721, erc1155) return 400. spl is accepted and returns 200 with an empty data, because no Solana chain is served yet: that empty page means the filter matched nothing on the EVM chains, not that the wallet is empty. direction is in or out.
• page_size defaults to 100 and the maximum is 1000. A larger value is clamped to 1000 rather than rejected, so read pagination.page_size back instead of assuming the request size was honored. The next page is page_token set to the previous response's pagination.next_page_token. A token is only valid for the filters that produced it. Each page is a separate billed request (https://docs.goldsky.com/pricing/summary#feeds). Do not walk the whole history unless the user asked for that range.

Other feeds on this key

/polymarket/activity, /polymarket/positions, /polymarket/balances, /blocks/{chain} and /blocks/{chain}/head are served by the same host and the same key. A Polymarket question does not have to become a pipeline: reach for /turbo-builder only when the rows have to land in the user's own database. Read the OpenAPI spec before calling one, because all three of the conventions above change:

• Pagination is limit and cursor, not page_size and page_token, and the next cursor is next_cursor at the top level, not pagination.next_page_token. page_size is accepted and silently ignored, so a request meaning to ask for 2 rows returns the default 100 — and every one of those is billed. Check what came back.
• The chain vocabulary is not the wallet feeds' vocabulary. /blocks serves 21 chains and spells Polygon matic; polygon, mainnet and every arbitrum spelling are 400 there. The wallet feeds serve 7 and spell it polygon, with matic only an alias. A slug that works on one is not known to work on the other.
• Errors are not always the {"error":{"code","message"}} envelope. A bad query string on these paths comes back as a bare string (Failed to deserialize query string: ...), so reading .error.message gets null. Branch on the status code, not on the body shape.
• Numbers are scaled differently per path. /polymarket/activity serves JSON numbers already in decimals (price: 0.65, amount_shares: 84.615385, amount_usdc: 55.0, and amount_usdc / amount_shares == price). /polymarket/positions and /polymarket/balances serve strings in raw 1e6 integer units, so avg_price: "100000" is $0.10 and amount: "12120000000" is 12120 shares. Divide by 1e6 before showing either to a user. positions.amount and balances.balance share that scale but are not the same quantity and disagreed on half the rows sampled here: amount is the trade-derived net position, balance is the conditional-token balance actually held, and splits, merges, redeems and plain transfers move tokens without a fill. Pick the one that answers the question and do not substitute one for the other.
• /polymarket/activity refuses an unselective scan. It needs one of address, token_id, a closed block window of 300000 blocks or fewer, a closed time window of 604800 seconds or fewer, or limit of 100 or fewer with no filter. Anything else is 400 MISSING_SELECTIVE_FILTER.
• Three values carry warnings in the spec. activity.fee's basis is inconsistent across some maker/taker pairs in the source data. positions.last_updated_at is pipeline upsert time, not block time, so rows with different block_number can share it; use block_number to order. balances.as_of_block is the latest balance at or before that block by insert order, not a reconstruction of that block.

Writing code against a response

• Balances are the latest amount of each token, not a history. Several transfers in one block become one balance row. Reconstruct movement from the transfers feed.
• A transfer object has no id. Do not build a primary key that includes block_number or block_timestamp: a transaction that survives a reorg comes back with different block fields. Do not assume one transaction emits a transfer only once.
• price_usd and amount_usd are computed when the request is served. A later read of the same transfer can show a different USD value.