Why are balance and monthly cap headroom missing?

This connection reads rated costs only: balanceAmount=null, officialLimit=unavailable/not_queried. That is not zero balance or unlimited funds. Fireworks has a public quota API, but this version does not read it. Internal budget differences are not provider allowance.

Why is today's cost low or suddenly higher?

Daily aggregation can lag a day or more and cache for minutes; a jump may be delayed accounting rather than new traffic. queryEnd is tomorrow's exclusive UTC boundary, not proof that data is reported through then. Do not fill gaps with token-price estimates.

Follow step 4's private GET /v1/accounts?readMask=name,displayName and copy the selected name after accounts/. The public model owner fireworks, a deployment/key ID or display name is not your paying account ID. Never share the key or account-list response in chat.

Does 403 require Admin?

No automatic elevation. Official billing docs say any key on the target account works, and Inference User can create self-owned keys; actual minimum-role behavior is still unverified. Check account ownership, validity and existing access. Stop if unresolved; do not create an administrator key or change roles.

How do units, nanos and usageBuckets work?

Top-level lineItems[].totalCost is authoritative USD Money: one unit is one dollar and a nano is one billionth. Zero units plus 125000000 nanos is $0.125; signed adjustments are preserved. usageBuckets repeats a breakdown and must not be added again or reconstructed from quantity/price.

What about 401, 404, 429 or a changed response?

Check expiry/revocation and copying for 401, account ID for 404, and allow backoff for 429. Missing totals, mixed currencies or errors fail the read and retain the old successful time instead of treating unknown as zero. Use manual billing if the ordinary key's risk is unacceptable, not payment or expanded access.

Fireworks AI costs & secure API key setup guide

Costs only · Not a balance

Guide updated 2026-10-16

Official billing reader implemented. Real key creation, minimum-role behavior, signed-in navigation and account-cost reconciliation await user testing. No supplier login, paid model test, top-up or privilege increase.

Required credentials: Dedicated ordinary API key on the target account
Monitoring currency: USD · rated costs, not balance
What is monitored: Account costs reported this UTC month; Fixed GET https://api.fireworks.ai/v1/accounts/{account_id}/billing/summary with Bearer authentication. startTime is this UTC month's first day; endTime is tomorrow's UTC midnight (exclusive). Date aggregation ignores time-of-day. Sum only top-level lineItems[].totalCost: USD units + nanos / 1000000000. units=0, nanos=125000000 is $0.125, not cents divided by 100, quantity × price or usageBuckets added again. balanceAmount=null. fireworksCost.queryEnd is a requested boundary, NOT a reported-through timestamp.

Check officially without sharing a key with Quota Monitor

Open the official console, select the actual billing account and inspect Total Spend in its existing usage/billing view. Do not pay or change auto-recharge to test.
Choose the same current UTC-month dates and account-wide costs, recording the amount/time. A different console zone or range is not an identical API window; costs can lag a day or more and today is incomplete.
Inspect balance/Credits, quotas, notifications and invoices separately. If ordinary-key access is unacceptable, keep checking manually without a higher role or plan purchase.

Open official Fireworks AI · account costs page

Connect Fireworks AI · account costs to Quota Monitor

Confirm the existing account and accounting scope
Open the Fireworks console and select the existing account that pays for your API activity. An account is the top-level billing resource shared by its users, models, deployments and training—not a Quota Monitor project. The accounts/fireworks prefix on public models is not your paying account.
This guide reads rated account costs, not wallet funds or Fire Pass remaining allowance. Without existing access, or if ordinary-key exposure is unacceptable, use the manual option below. Do not purchase access, accept new invitations or upgrade a plan.
Open official Fireworks AI · account costs page(opens in a new tab)
Use existing privileges without granting more access
Official Usage & Cost Breakdown says any API key on the target account works. The roles table permits Inference User, Contributor, User and Admin to create self-owned keys. Even the least-privileged listed Inference User can run paid inference; it is not a billing-only reader.
Use your existing authorized identity and a dedicated ordinary key. No documented billing-only scope or permission JSON is available. Do not become Admin, create a service account or change roles. For 403, check account/key matching first; actual minimum-role behavior remains pending user verification.
Open official Fireworks AI · account costs page(opens in a new tab)
Create and privately save a dedicated ordinary key
Open the API Keys page below, verify the same account, and select Create API key. Use a recognizable name such as QuotaMonitorCosts. Complete the current form and privately save the full value immediately: it is shown only once. The API documents expireTime, but do not assume a matching UI control exists.
A name or expiry does not make the key read-only. Do not regenerate, delete or overwrite another application's key, or run the Quickstart's paid model examples. Copy only the ordinary account key—not an Authorization line, Cookie, Fire Pass-specific credential or session password.
Open official Fireworks AI · account costs page(opens in a new tab)
Privately obtain account_id instead of guessing a model owner
In your own non-shared local HTTP client, disable cloud sync, history and request logging. Create GET https://api.fireworks.ai/v1/accounts?readMask=name,displayName. Put the key only in Authorization: Bearer, without a body or redirects. Match displayName to the intended account and copy only the account_id after accounts/ in name.
If nextPageToken is present, use pageToken on that same official URL with the same readMask; never follow a token as a URL. IDs have 1–63 lowercase letters/digits/hyphens, start with a letter and cannot end with a hyphen. Already-configured firectl users can privately verify with firectl account get. Quota Monitor calls neither identity nor account-list endpoints. Close the local client afterward.
Open official Fireworks AI · account costs page(opens in a new tab)
Securely add the key and account binding
In your Quota Monitor project, add a connection and choose Fireworks AI · account costs (USD). Enter a recognizable name, paste the full key in the masked input and put step 4's slug in the project binding's required account ID field. Do not include accounts/, a display name, model/deployment ID, key ID or URL.
Currency is USD, with no project/key filters or daily-burn estimate. For MCP, save the key through the secure link first, then specify the account ID on the project binding; never give an assistant the secret. Credentials are encrypted. The help drawer shares this guide and opening it does not submit a connection or read the provider.
Wait for the first success and reconcile identical dates
After saving, wait for the first sync or use the existing refresh action. Check that the successful observation time advances. The request runs from this UTC month's first day to tomorrow's midnight, exclusive: on October 16, endTime is October 17 to include October 16's reported portion. queryEnd does not prove source coverage through that time.
Compare rated costs for the same account/date range. lineItems[].totalCost with units=0 and nanos=125000000 is $0.125. Preserve nine decimal places and signed adjustments; never add usageBuckets or reconstruct quantity × unitAmount. REST/MCP share the snapshot. Failures, an empty object, missing money or non-USD data preserve the old value/time; only an explicit empty lineItems array means zero reported costs.
Open official Fireworks AI · account costs page(opens in a new tab)
Check Credits, lag and provider limits separately
Use the account's existing official usage/billing view to compare Total Spend. Match the account and dates; costs can lag a day or more, responses are cached for minutes, and today is partial. A different console time zone or range is not directly equivalent. For exact reconciliation, use step 4's private client with the documented summary GET and identical startTime/endTime, retaining only amounts and times.
Check wallet funds, Credit grants/purchases, monthly limits and invoices separately. not_queried means this connection did not read monthly-spend-usd—not an absent cap. Enterprise spend notifications differ from enforced caps on other accounts. An optional internal budget gives only a reference difference; it changes no funds, provider caps or reset times.
Open official Fireworks AI · account costs page(opens in a new tab)
Rotate and troubleshoot without escalation or paid calls
When rotating, create only a replacement monitoring key on the same account. Update it through the secure form, confirm a successful sync, then revoke only the old monitoring key. Leave other application keys alone. For 401 check expiry/revocation and the complete value; for 403 check key/account ownership; for 404 check account ID and the official contract; allow 429 backoff.
Share only error codes, times and redacted symptoms—not keys, raw responses or screenshots. A failed sync retains the previous successful snapshot/time; missing data is not estimated zero. Do not troubleshoot with administrator access, purchases, console Cookies, guessed endpoints or a billable model test. Real-account reconciliation remains pending.
Open official Fireworks AI · account costs page(opens in a new tab)

What Quota Monitor reads from Fireworks AI · account costs

Account-wide rated usage across the report's serverless, dedicated and training categories, not per-model/key costs, wallet funds, granted/purchased Credits, Fire Pass allowance or a final invoice. Projects can share one account: do not sum duplicate account reports or estimate missing costs from token prices.
The current UTC month includes today's partial reported usage. Official documentation says rated costs and Total Spend can lag a day or more, with responses cached for minutes; no maximum freshness is promised. queryStart/queryEnd are requested dates, not a source watermark or provider reset schedule.
This connection does not read the documented monthly-spend-usd quota. officialLimit=unavailable/not_queried is not not_configured and does not mean Fireworks lacks a quota API. Enterprise spend alerts differ from enforced non-Enterprise caps. No cap headroom is inferred across unmatched periods/time zones. Internal budget differences are reference-only, never funds, changed caps, guaranteed access or runway.
The dashboard shows the latest successful cost snapshot. Refresh cannot remove upstream reporting delays. An old cost and timestamp after a failed query are not a new report.

Fireworks AI · account costs FAQ and troubleshooting

Why are balance and monthly cap headroom missing?: This connection reads rated costs only: balanceAmount=null, officialLimit=unavailable/not_queried. That is not zero balance or unlimited funds. Fireworks has a public quota API, but this version does not read it. Internal budget differences are not provider allowance.
Why is today's cost low or suddenly higher?: Daily aggregation can lag a day or more and cache for minutes; a jump may be delayed accounting rather than new traffic. queryEnd is tomorrow's exclusive UTC boundary, not proof that data is reported through then. Do not fill gaps with token-price estimates.
Where is account_id?: Follow step 4's private GET /v1/accounts?readMask=name,displayName and copy the selected name after accounts/. The public model owner fireworks, a deployment/key ID or display name is not your paying account ID. Never share the key or account-list response in chat.
Does 403 require Admin?: No automatic elevation. Official billing docs say any key on the target account works, and Inference User can create self-owned keys; actual minimum-role behavior is still unverified. Check account ownership, validity and existing access. Stop if unresolved; do not create an administrator key or change roles.
How do units, nanos and usageBuckets work?: Top-level lineItems[].totalCost is authoritative USD Money: one unit is one dollar and a nano is one billionth. Zero units plus 125000000 nanos is $0.125; signed adjustments are preserved. usageBuckets repeats a breakdown and must not be added again or reconstructed from quantity/price.
What about 401, 404, 429 or a changed response?: Check expiry/revocation and copying for 401, account ID for 404, and allow backoff for 429. Missing totals, mixed currencies or errors fail the read and retain the old successful time instead of treating unknown as zero. Use manual billing if the ordinary key's risk is unacceptable, not payment or expanded access.

Official Fireworks AI · account costs references

Prepared by Quota Monitor, not an official provider tutorial. Interfaces and permissions may change; check official pages. Verification is limited to the scope stated here.