Questions, with straight answers.

You send it files and it reads them: every page becomes Markdown, a page image and searchable chunks, and, if you give it a JSON schema, structured fields.

Then you search those pages with citations, or ask questions through a chat endpoint that answers from your documents and cites the page each claim came from.

Sign up and drop a file into the app. New accounts get 250 starter credits and no card is needed. 1 credit = 1¢, so a typical document costs a few cents.

From a terminal, install the CLI with npm install --global @docslurp/cli, then run docslurp upload ./file --watch and docslurp search.

No. Every organization has a default workspace, and any request that leaves out workspaceId uses it. Anyone in the organization can read and upload to it.

Create more workspaces when you want to keep collections apart, such as contracts and product manuals. A new workspace names itself from the first files you upload into it.

The REST API at https://docslurp.io/v1, the Node SDK (@docslurp/sdk), the docslurp CLI, and an MCP server for AI agents. The SDK also ships embeddable search and chat components for React and Web Components.

There is no Python SDK. From Python or any other language, call the HTTP API directly. The SDK and CLI need Node.js 20 or newer.

PDF; Word, Excel and PowerPoint (both current and legacy formats); EPUB, HTML, Markdown, plain text, RTF, LaTeX, CSV, TSV, JSON and JSONL; PNG, JPEG, TIFF and WebP images; MP3, M4A, WAV and WebA audio; and MP4 and WebM video.

Images and scanned pages are read with OCR. Audio and video are transcribed.

Up to 500 MB per file. Simple multipart uploads are capped at 25 MB; larger files go through direct upload sessions, which the SDK and CLI use for you.

Markdown for every page, an image of every page, chunks with embeddings for search, and, unless you turn it off, a summary of the document. With a workspace JSON schema you also get structured fields, each checked against the page it came from.

File links stream through an authenticated proxy and only work with your API key. They expire after about 75 minutes.

Uploading the same filename to the same folder in a workspace creates a new version. Search and chat see only the newest one; older versions stay readable until they age out of retention.

Text-layer documents work in any language. OCR covers Latin scripts and scripts such as Chinese, Japanese and Korean. Each upload reports the languages it detected and can suggest a multilingual embedding profile when your documents need one.

Hybrid search: keyword and vector results are combined and reranked. Each result has the passage text, a score and a citation with the filename and page number.

Keyword search covers a document as soon as its text is chunked. Vector search joins when indexing finishes, and every response says which parts are ready.

Yes. Point an OpenAI client at https://docslurp.io/v1 and call chat completions with the model workspace:default, or workspace:<slug> for a specific workspace. The model picks a workspace, not an inference model.

Streaming works. Answers end with a list of sources, and a docslurp extension in the response carries structured citations. Tool calls, image inputs and the Responses API are not supported.

@docslurp/contrib-search-compat provides drop-in search clients for all four, so existing search UIs keep working against your DocSlurp index. There is also a native Astro component and a WordPress plugin.

Any client that speaks MCP over Streamable HTTP, at https://docslurp.io/v1/mcp. There are setup guides for Claude Code, Codex, Cursor, VS Code, OpenCode and Claude, plus a stdio bridge for other clients.

Agents can list workspaces, upload files, check upload status, search and chat. Sign-in uses OAuth, or a server API key for unattended agents.

Yes, in the app. You speak and the answer comes back as text. Recordings stop after 2 minutes and the audio is not saved.

Optional, per workspace. When it is on, DocSlurp builds a graph of entities and relationships from your documents, and search can follow it to find related passages. It is off by default.

Yes. Free gives you 250 starter credits once, 50 concurrent workers and 1 GB of storage. No card needed. Paid plans add monthly credits, more workers, more storage and auto-recharge.

No. 1 credit = 1¢ on every plan, and a page costs the same number of credits whichever plan you are on. A plan's monthly fee comes back as included credits, with a bonus on the larger plans; what you buy is speed, in workers that read pages at the same time.

A worker slot processes one page task at a time: parsing, layout, OCR, chunking or embedding. Your plan sets how many slots your organization can run at once, across all workspaces and runs.

Workers run on a shared fleet. The number is a ceiling on concurrency, not machines set aside for you. When the fleet is busy you may run below your ceiling; reserved capacity is an Enterprise agreement.

One run fans out to at most 128 page tasks. Gold already allows 250, so a single 1,000-page book takes about 30 s on Gold and 30 s on Platinum.

Platinum's 1,000 workers pay off when you upload many files at once: a back catalog, an archive, a nightly sync.

A run that has started finishes, even if that takes your balance below zero. New runs wait until you add credits. A negative balance is paid down first by your next monthly grant or top-up.

On a paid plan, turn on auto-recharge to buy credits automatically when your balance falls below a threshold you choose, up to a monthly limit you set.

Included credits are spent first and expire at the end of each billing period. Prepaid credits (top-ups, auto-recharge and the starter grant) don't expire while your account is active. They lapse after 12 consecutive months without billable activity, where the law allows.

Upgrades take effect immediately. You pay the prorated difference for the rest of the period and receive credits in proportion to what you paid, bonus included.

Downgrades and cancellations take effect at the end of the current period, and you keep your plan until then. Change or cancel at any time from the billing portal in the app.

They stay. Storage keeps accruing daily at the rate card against your prepaid credits until you delete the workspaces you no longer need.

If your balance stays negative for 30 days with no payment method on file, we email you, and we may delete stored content 14 days after that notice.

No. Prices are in US dollars and exclude tax. Checkout calculates sales tax, VAT or GST from your billing address, and you can add a business tax ID.

Plan fees and credits aren't refundable, except where the law requires it. If a charge looks wrong or a run failed on our side, tell us and we'll review it. A refunded payment removes the credits it granted.

Yes, on Enterprise: reserved capacity, fleet isolation, an SLA, a DPA, security review and invoiced billing. Email [email protected].

From a planning model with a per-page cost for each pipeline and page type (prose, tables and figures, scans), anchored to internal measurements of text-layer parsing, CPU layout analysis, GPU OCR and vision models. It assumes 85% parallel efficiency and 15 seconds of fixed overhead per batch.

Real documents vary. Every run in the app reports its measured pages per minute, pages per $1 and credits, and those figures are what you are billed on.

No. We do not sell your content and we do not use it to train general-purpose AI models.

Some features send parts of a document to an AI model provider: reading complex pages with a vision model, writing summaries, answering chat questions and transcribing voice. Which ones run depends on your workspace settings.

Workspaces can run in a local-only or strict-privacy mode that limits what leaves our infrastructure. There is no zero-data-retention provider tier today.

We are based in the United States. We and our subprocessors (hosting and storage, AI model providers, payments, email, error monitoring and product analytics) may process data in the United States and other countries.

Each organization gets its own search index table.

Delete documents and workspaces from the app, or through the API. Records are deleted from our database immediately, and a background sweep removes indexes, page images and original files shortly after. Backups age out within 35 days.

Deleting an organization or account removes everything in it. For privacy requests, email [email protected].

Encryption in transit, hashed passwords, scoped API keys and access controls. You can sign in with GitHub, Google or company SSO. API keys are scoped, for example to upload or to search, and browser widget keys are separate from server keys.

Need a DPA, a security review or an SLA? Those come with Enterprise. Email [email protected].

Yes. The self-host stack runs on one machine with Docker Compose: Postgres, Redis, S3-compatible storage, the API and workers. Local embeddings work without any provider key; features that call language models need one.

Self-hosted DocSlurp has no web dashboard. You use it through the API, CLI, SDK and MCP. It is a single-host setup, not a high-availability cluster.

The server, worker and CLI are source-available under SSPL-1.0, and each version is also licensed under Apache-2.0 three years after its release. The SDKs and client integrations are MIT.

You can self-host for your own use. Offering DocSlurp to others as a hosted service triggers section 13 of the SSPL.

Create keys in the app. Each key is scoped (for example ingest for uploads, search for search and chat) and uses its organization's default workspace unless a request names another.

Send the key as a Bearer token. Browser widget keys are separate keys for embedded search, and they do not grant MCP access.

Yes, per minute and set by your plan, with separate limits for uploads, search, chat, reads and writes. Paid plans get higher limits than Free.

You can also put your own caps on a key, by request count or by spend, over windows of up to 31 days.

Yes. Register a webhook and DocSlurp sends run.started, run.progress, run.completed and run.failed events. You can test a webhook, rotate its secret, and see and replay past deliveries. The CLI's --watch flag follows a run from the terminal instead.