Skip to content

API keys and limits

Give each app the least it needs, and cap what it can spend.

A key is a password for your account's documents and credits. This page says what each kind of key can do, what it can cost, who can use it, and which guard to set. The links in these docs open the create form already filled in with safe limits. You review the form, change anything, and press the create button yourself.

On this page

Two kinds of key

A server key (dslp_priv_…) lives in backend secrets, CI or an agent's configuration. It can hold any permission you grant, including uploads and, only if you opt in, organization management.

A browser key (dslp_pub_…) lives in page HTML. It can only search and, if you add it, answer questions, and only inside the projects and websites you name. Anyone can copy it from your page source, so treat every browser key as public and keep its limits tight.

Relative risk, lowest to highest

Each permission adds what a key can do and what it can cost. Choose the lowest row that your app works with.

PermissionWhat it can doWhat it can costWho can trigger itGuard to set
Keyword searchMatch words in your documents and return cited passages.A very small charge per search, the lowest search price. No AI model is called. It is not free: the spend limit and the searches-per-day cap bound it.Anyone on your allowed websites for a browser key; anyone with a server key.Searches per day, spend limit, project restriction.
Hybrid searchKeyword search plus meaning-based matching. It embeds each query.A small charge per search, which includes embedding the query.As above.Searches per day, spend limit, project restriction.
Deep searchHybrid search plus AI reranking and query rewriting. On server keys it also allows paid web tools.The most per search: AI reranking, query rewriting and graph expansion. On a server key it also unlocks paid web search tools.Server key holders. Give it to a browser key only if you accept the cost of every visitor's search.Deep searches per day, spend limit, expiry, IP allowlist.
Answers (chat)Write cited answers with an AI model. It also lets an MCP client or the chat endpoints answer.Each answer is written by an AI model and billed by its length. The retrieval each answer runs is billed too.Anyone on your allowed websites for a browser key, as often as the limits allow; anyone holding a server key with answers. A browser key with answers is high risk even with a spend limit: the limit caps what visitors can spend, not who can.Spend limit, chats at once, answer length, project restriction.
UploadAdd documents to projects. It also lets the key manage the documents it can reach.Billed per page processed; project settings can add AI steps per page.Server key holders only.Pages per day, spend limit, expiry, IP allowlist, project restriction.
Billed extraction featuresRun optional pipeline steps that cost more per page, such as structured extraction.Uploads may use extraction features beyond the project settings, billed per page.Server key holders only.Spend limit, pages per day. Grant it in the form; links cannot pre-fill it.
On-demand indexingA deep search can ask for pages that were deferred to be indexed now.Searches may parse pages on demand and spend credits doing it.Holders of a key whose searches may request it: deep search, or a key whose on-demand indexing settings allow it (website keys included).Spend limit and the key's on-demand indexing settings in the form.
Organization and project managementRead or change organization details, projects, members and keys.No direct charge, but it can change who has access and what projects exist.Server key holders only. Any member can grant read access; changing the organization or its projects needs an owner or admin to grant it.Expiry, IP allowlist, a short life for one-off jobs. Revoke it when the job is done.

Guidance

  • Least privilege. Start from the permission rows above that your app needs, and nothing more. A search box needs search, not answers. A CI job needs upload, not search.
  • One key per app. You can then revoke one without breaking the rest, and the spend and last use of each stay separate.
  • Set a spend limit and an expiry. Every link here that can spend credits sets a spend limit (12 of 15). The other 3 (Algolia settings import, Read organization, Manage organization) hold no permission that spends credits. Links set an expiry on server keys only; add one to a browser key in the form (Advanced, Expiry) if you want it. A spend limit stops billed work from one key when it is reached; an expiry limits how long a forgotten key stays useful.
  • Restrict projects. Browser keys must name their projects. For server keys, name them too unless the key really needs every project.
  • Restrict where it works. Add allowed websites to a browser key, and an IP allowlist to a server key that runs from fixed addresses.
  • Rotate before you revoke. Rotation with an overlap of up to seven days gives you a new key with the same settings while the old one keeps working, so you can deploy the new secret first.
  • Revoke at once if a key leaks. Do not use overlapping rotation for a leaked key: it stays usable during the overlap. Revoke it, remove it from wherever it was exposed, and issue a replacement.
  • Never put a server key in a page. Minifying does not hide it. If a page needs search with a server key's reach, search through your own server instead.

Default limits in the links

Every link on this page and in the guides fills the form with these limits unless a recipe says otherwise. They are starting points: edit them in the form under permissions and limits before you create the key. A key a link creates carries only the limits you see in the form.

LimitDefault in links
Daily spend cap$5 a day, on every key that can use billed features
Monthly spend cap$50 over a rolling 30 days, on the same keys
Expiry90 days on server keys. Links never set one on browser keys; you can add one in the form (Advanced, Expiry) when you create or edit the key
Chats at once4 on browser keys and 16 on server keys with answers. Each server process keeps its own count, so a key can run more at once when several servers share the traffic
Searches50,000 a day, in the key's mode, on browser keys
Requests per minuteNot set, so your plan limits apply
Pages per day2,000 on keys that can upload
Answer length1,000 tokens at most per answer, on keys with answers

A link can only lower the pages-per-day cap from the form's own 10,000. Whoever runs this site sets these defaults with environment variables whose names start with PUBLIC_DOCSLURP_KEY_LINK_; a self-hosted deployment lists them in .env.example. A value the form cannot accept is ignored.

When a limit is reached

A key that reaches a limit gets an error until the window passes. It does not stop other keys.

  • A spend limit returns 429 with spend_limit_exceeded and a Retry-After header. Billed work (uploads, search, answers) stops; reads such as run progress and document lists keep working.
  • A request limit (a requests rule or requests per minute) returns 429 with rate_limited and stops every call of that kind, reads included, until the window passes.
  • A chats-at-once limit refuses a new answer while that many are already running.
  • A searches-per-day limit returns 429 with search_limit_exceeded. A cheaper search mode keeps working under its own limit.
  • The pages-per-day cap stops uploads with daily_page_limit.
  • An organization that is out of credits gets 402 on billed work, whatever the key's own limits.
  • Spend is recorded when work finishes, so a request already running can carry a key slightly past its spend limit.

Open Connections to see each key's permissions, last use and address, and to edit or revoke it. Billing shows the credits the work used.

Create a key

Each link opens the create form filled in for one guide. Nothing is created until you press the create button, and you can change every field first. Risk follows the same rule as the create form: the highest row from the table above that the recipe touches, and high for any browser key that can run answers or deep search.

Browser keys

For a page on your own website. Anyone who views the page can copy the key.

  • Website keyword search. A search box on your site that matches words and never calls a model. Risk: low. Pre-fills keyword search only, 50,000 keyword searches a day, $5 a day, $50 every 30 days.
  • Website hybrid search. A search box on your site that also understands meaning, which costs a little per search. Risk: medium. Pre-fills hybrid search only, 50,000 hybrid searches a day, $5 a day, $50 every 30 days.
  • Website answer widget. Search with cited AI answers on your site. Every answer costs credits. Risk: high. Pre-fills hybrid search, answers, 50,000 hybrid searches a day, $5 a day, $50 every 30 days, 4 chats at once, answers up to 1,000 tokens.

Server keys

For your backend, CI, a command line or an AI client. They never go in a page.

  • Get started. Your first key: upload documents and search them from your own code. Risk: high. Pre-fills uploads, hybrid search, $5 a day, $50 every 30 days, 2,000 pages a day, expires in 90 days.
  • Command line. Upload files and search from a terminal with the docslurp CLI. Risk: high. Pre-fills uploads, hybrid search, $5 a day, $50 every 30 days, 2,000 pages a day, expires in 90 days.
  • SDK quickstart. Upload, search and ask questions from Node.js with the SDK. Risk: high. Pre-fills uploads, hybrid search, answers, $5 a day, $50 every 30 days, 16 chats at once, 2,000 pages a day, answers up to 1,000 tokens, expires in 90 days.
  • Chat backend. Answer questions from your own server, or through an OpenAI-compatible client. Risk: medium. Pre-fills hybrid search, answers, $5 a day, $50 every 30 days, 16 chats at once, answers up to 1,000 tokens, expires in 90 days.
  • AI client (MCP). Let an agent search your projects and ask for cited answers. It cannot upload. Risk: medium. Pre-fills hybrid search, answers, $5 a day, $50 every 30 days, 16 chats at once, answers up to 1,000 tokens, expires in 90 days.
  • AI client that can upload. The same as the AI client key, and the agent may also upload files. Risk: high. Pre-fills uploads, hybrid search, answers, $5 a day, $50 every 30 days, 16 chats at once, 2,000 pages a day, answers up to 1,000 tokens, expires in 90 days.
  • Search through your server. Your backend searches for the page, so the key never reaches a browser. Risk: medium. Pre-fills hybrid search only, $5 a day, $50 every 30 days, expires in 90 days.
  • Upload from CI. Upload documents from a pipeline. It cannot search or ask. Risk: high. Pre-fills uploads, $5 a day, $50 every 30 days, 2,000 pages a day, expires in 90 days.
  • Algolia settings import. Copy index settings, synonyms and rules into a project, then discard the key. Risk: high. Pre-fills managing projects, expires in 30 days.
  • Read organization. Read organization and project details for reporting. It changes nothing. Risk: low. Pre-fills reading organization details, expires in 90 days.
  • Manage organization. Create and change projects and organization settings from automation. Risk: high. Pre-fills managing the organization, managing projects, expires in 90 days.

Chat keys use hybrid search, because answers retrieve in their own mode. Add deep search only if your requests ask for deep retrieval and you accept its cost.

Jump to