Skip to content

API

Keys, limits and errors

Everything the remake API expects from a call, and everything it can answer. For a first walk-through, start with the API quickstart.

Base address

https://app.tryremake.com/v1

Every call goes over HTTPS and sends and receives JSON. The API is for servers and scripts: it doesn't accept calls from a web page on another site.

GET /v1/openapi.json returns the full schema, with no key. The API reference renders it.

Keys

Item Value
Where Your AI › API keys › Create a key
Format hf_ followed by 32 characters
Shown Once, when you create it. Later the list shows only its first 8 characters.
Header Authorization: Bearer hf_…, with a capital B
Live keys per account 5
Revoking Revoke key in the key's ⋯ menu stops it at once
Who The account owner. Viewer seats can't create or revoke keys.

A key has every permission: read, change and spend. Keep it on a server, never in a web page.

AI assistants connected by sign-in hold their own tokens, also starting hf_. Those don't count toward the 5 keys; an account can connect 10.

Each one works only within the permissions you granted: read, or write, which includes spending. See Connect your AI assistant.

Rate limit

Calls are free. One limit applies, and it is the same for every account.

Limit Value
Calls per account 120 in any 60 seconds
Counted across Every API key and connected app on the account, API and MCP together
Over the limit 429 rate_limited: "Too many requests. Try again in N seconds."

Every answer to a call with a working key carries these headers:

Header What it says
X-RateLimit-Limit Calls your account may make in any 60 seconds
X-RateLimit-Remaining Calls still open right now
X-RateLimit-Reset The time the next call frees up, in Unix seconds
Retry-After On a 429 only: seconds to wait

Nothing is charged for a refused call and nothing is lost. The limit clears by itself as your earlier calls pass the 60-second mark.

GET /v1/me returns the limit as rate_limit. The remake app in your browser is never limited.

Runs at once

An account may have 3 runs queued or running at a time, across every brand. A brand may have 1 run going at a time. Past either limit, the confirm answers 409, nothing runs, and nothing is charged.

A scheduled run that comes due at the account limit waits and starts in turn as other runs finish.

Conventions

Routes such as /runs and fields such as run_id hold your runs.

Thing Format
Brand id The brand's website domain, such as samsung.com. www.samsung.com also works. AI assistants may pass a full address.
New brand's website When you add or import a brand: a public address such as acme.com. A port number, a login or a private or local network address answers 400 and says which.
Workspace id A number, from GET /v1/workspaces
Run id The run's token: 16 characters, which can include - and _. Prompt-writing jobs use ids starting gen_.
Run status queued, running, done or failed. done and failed are final.
Times ISO 8601 in UTC, such as 2026-10-07T07:08:51+00:00
Money Mills: thousandths of a dollar. 558 is $0.558, shown in the app as $0.56.
Request body A JSON object. Routes that need no input accept an empty body.

A brand that isn't on your account answers 404, the same as one that doesn't exist.

Pagination

These lists take limit and offset:

  • GET /v1/workspaces
  • GET /v1/brands/{brand}/runs
  • GET /v1/brands/{brand}/runs/{token}/answers
  • GET /v1/looker/data

The default page size is 50 rows, and the most is 200. Offsets start at 0.

The answer wraps the rows as {total, limit, offset, items}. The answers route names its rows answers instead of items.

GET /v1/brands/{brand}/prompts returns every prompt at once. GET /v1/brands/{brand}/visibility returns the latest 24 runs. GET /v1/wallet returns the latest 20 transactions.

Errors

Every error has the same shape:

{"error": {"code": "not_found", "message": "No such brand on this account."}}
Status code When
400 invalid_request A parameter or body is missing or malformed
400 budget_floor A monthly cap below the price of one run
400 unreachable The website couldn't be read
400 unknown_market A prompt's country is not one of the brand's countries
400 name_not_real, names_in_questions, not_lookalike A pitch name or pitch prompt can't be used. The message says why.
401 unauthorized No key sent
401 invalid_key The key is wrong or revoked
401 key_expired An assistant's sign-in token has expired
402 balance_empty Your wallet can't cover the spend
402 spend_limit_reached Your monthly spending limit on Your AI is reached. Nothing was charged; the message says where to raise it.
403 insufficient_scope An assistant's token lacks the permission
403 audience_mismatch A route for agency accounts, called by another account
403 viewer_readonly A Viewer seat tried to spend or change something
404 not_found Not on your account, doesn't exist, or the brand has no schedule
405 method_not_allowed The wrong HTTP method
409 quote_changed The quote is missing, expired, used or out of date. A fresh quote comes with it.
409 too_many_pending 3 runs are already queued or running
409 already_running This brand's run, or a re-check of its site's fixes, is already going
409 budget_reached The brand's monthly cap is reached
409 paused The schedule is paused
409 pick_country The brand has no country yet. Give it one first.
409 prospect_readonly The site is a prospect. Change it in its pitch, or convert it first.
409 request_id_reused The request_id was used for another action
409 request_in_progress The first call with this request_id hasn't finished
409 other A conflict named in the message, such as no_new_prompts, name_taken, already_launched or not_paused
410 expired A brand's 7-day restore window has passed, or a pitch report has already expired
429 rate_limited Your account made more than 120 calls in 60 seconds
429 cooldown A fix re-check on the same site within a minute
500 draft_failed Saving written prompts failed. The charge is refunded.
502 vendor_failed, check_failed The work failed. A failure on our side is refunded. A fix re-check of a site that was down, slow, broken or too large to read is charged, and the message says so.
503 not_ready MCP only: the MCP address is restarting. Try again in a moment.

A 429 and a cooldown both send Retry-After.

Spending

Calls to the API are free, reads and changes alike. Only the work a call starts costs money, at the prices on the rate card.

Work your wallet can't cover answers 402 balance_empty and nothing runs. Work past your monthly spending limit answers 402 spend_limit_reached, and nothing is charged. Reads keep working either way.

These 10 calls spend credit, each quoted first:

Call What it buys
POST …/prompt-spend/run-now One run of the brand's schedule, now
POST …/prompt-spend/resume Resumes the schedule, with a run within about a minute
POST …/prompt-spend/launch Starts a schedule. Nothing now; each scheduled run is charged.
POST …/prompt-spend/edit Changes a schedule. Nothing now; it changes later runs.
POST …/verify-fix A fix re-check on the live site
POST …/prompts/generate Prompts written with AI
POST …/competitors/suggest Up to 8 real rival businesses with their websites
POST /v1/pitches/{id}/site-read A pitch filled in from the prospect's website, on agency accounts
POST /v1/pitches/{id}/questions/generate A pitch's 25 prompts written with AI
POST /v1/pitches/{id}/run One pitch, asked of ChatGPT, Perplexity and Google AI Overviews

POST …/prompt-spend/delete adds no price of its own but follows the same 2 steps, because it can't be undone.

Quote, then confirm

  1. Call the route without confirm. Nothing happens. The answer is a quote: price_mills, balance_after_mills, would_refuse, the card rows and a quote_token.
  2. Call it again with "confirm": true and that quote_token.

A quote lasts 30 minutes

A quote confirms once. A quote for one brand, route or key doesn't work for another. A confirm with a missing, expired or used quote answers 409 quote_changed, with a fresh quote, and charges nothing.

Retries never redo the work

Add a request_id to a confirm. It takes up to 64 letters, digits, ., _, : or -. remake remembers it for 24 hours, per key and brand.

  • Sending the same call again returns the first answer, with "replayed": true. The work isn't charged again.
  • The same id on a different action answers 409 request_id_reused.
  • An id whose first call failed or was refused is free to use again.

How charges land

A run is charged one answer at a time as it runs. Answers that fail on our side are refunded. The run object carries no cost; GET /v1/wallet shows your balance, your latest transactions and this month's spend by brand.

What is charged, and what isn't

A call costs nothing, whether it reads or changes something. That covers bringing data in, the quote step of every spending call, and a call refused with a 429.

The 10 spending calls above charge their quoted price when you confirm. Nothing else is charged.

FAQ

Can I call the API from a web page?

No. It doesn't answer calls from a browser on another site. Call it from your server, and keep the key there.

Does a quote reserve the price?

For 30 minutes. If the price moves before you confirm, the confirm returns the new quote. Nothing is charged.

Is there a sandbox?

No. Reads and quotes are free, so you can test them on any account. The spending calls above charge their own price, and each quotes first.