AlQari Document Intelligence API

AlQari exposes a REST API over HTTPS to upload documents, run OCR with structured text, coordinates and confidence, validate documents, ask questions, and automate workflows — for Arabic and English documents. Response formats vary by endpoint (JSON, plain text, or HTML). Requests authenticate with a JWT access token or a qari_ API key.

Authentication

Create an API key (starts with qari_) from your dashboard, then send it in the Authorization header on every request. The full key is shown only once at creation. Never expose your key in client-side code or public repositories.

curl https://api.alqari.sa/documents \
  -H "Authorization: Bearer qari_••••••••••"

Alternative: session token via sign-in

For end-user apps, sign in via POST /auth/login with email_or_phone and password to receive an access_token, used in the Authorization header the same way. Session tokens are short-lived — use a qari_ key for long-running server integrations.

curl -X POST https://api.alqari.sa/auth/login \
  -H "Content-Type: application/json" \
  -d '{ "email_or_phone": "you@company.sa", "password": "••••••••" }'

A successful response returns: access_token, refresh_token, token_type, user_name, email, phone_number.

Base URL

All endpoints are relative to the base URL below. Always use HTTPS. Paths are currently unversioned (there is no /v1).

API Keys

Create an API key from your dashboard. Creating, listing, and revoking API keys requires an organization Admin. Keys start with qari_ followed by 40 hex characters, are shown in full only once at creation, then stored hashed — keep it safe. Keys have no scopes and no create-time expiration.

The create request body contains only a name. Management endpoints: POST /api-keys to create, GET /api-keys to list (metadata only, never the secret), and DELETE /api-keys/:id to revoke.

curl -X POST https://api.alqari.sa/api-keys \
  -H "Authorization: Bearer <JWT>" \
  -H "Content-Type: application/json" \
  -d '{ "name": "Production server" }'

Errors

The API uses conventional HTTP status codes. Domain errors carry a human-readable detail plus a stable machine-readable code — branch on code, not on message text. Validation errors are returned as 422.

// DomainError (fields may be optional depending on the error)
{
  "detail": "Document not found",
  "code": "DOCUMENT_NOT_FOUND",
  "violations": null,
  "context": null
}

// 422 Validation error
{
  "detail": [
    { "loc": ["body", "file"], "msg": "field required", "type": "missing" }
  ]
}

// 500 Unexpected
{
  "detail": "An unexpected error occurred. Please try again."
}

See the Error Codes section for the list of stable machine-readable codes.

n8n / Odoo integrations

To build pipelines outside the Al-Qari canvas (n8n, Odoo, or any HTTP client), use /services/integration/* endpoints. Each call runs one workflow node (OCR, extract, validate, human review, …). Chain steps by passing document_id from the previous response.

Request and response details for each endpoint are in the sections below.

Upload & OCR (integration)

Entry point for n8n/Odoo pipelines. Send multipart/form-data with file. output_format: plain-text (default) | json-layout | markdown.

curl -X POST "https://api.alqari.sa/services/integration/upload-ocr?output_format=plain-text" \
  -H "Authorization: Bearer qari_••••••••••" \
  -F "file=@invoice.pdf" \
  -F "language=auto"

{
  "document_id": "uuid",
  "filename": "invoice.pdf",
  "output_format": "plain-text",
  "text": "...",
  "pages": 1,
  "confidence": 0.97
}

Extract fields (integration)

Extract fields from a document after OCR. Pass document_id from upload. Optional state carries prior step output.

curl -X POST https://api.alqari.sa/services/integration/extract-fields \
  -H "Authorization: Bearer qari_••••••••••" \
  -H "Content-Type: application/json" \
  -d '{
    "document_id": "uuid",
    "fields": ["invoice_number", "total", "date"]
  }'

{
  "document_id": "uuid",
  "output": {
    "fields": { "invoice_number": "INV-001", "total": 124500 }
  }
}

Human review (integration)

Create a review task in the in-app Review Queue. Poll status via GET /workflows/reviews/{review_task_id}.

curl -X POST https://api.alqari.sa/services/integration/human-review \
  -H "Authorization: Bearer qari_••••••••••" \
  -H "Content-Type: application/json" \
  -d '{
    "document_id": "uuid",
    "assignee_email": "reviewer@company.com",
    "instructions": "Verify vendor name",
    "send_email": true
  }'

{
  "document_id": "uuid",
  "output": {
    "status": "awaiting_review",
    "review_task_id": "uuid",
    "assignee_email": "reviewer@company.com",
    "review_reasons": ["..."],
    "workflow_run_id": "uuid"
  }
}

MCP server for AI assistants

The official Model Context Protocol (MCP) server connects AlQari to Claude, ChatGPT, Codex, Cursor and other MCP-compatible assistants. Ask in plain language to OCR a file, search your documents, summarize a document, extract its fields or tables, classify or validate it, or run one of your workflows. The server runs on your computer and calls the AlQari API with your company's API key.

1. Create an API key

Sign in with an Admin account and open Settings → API Keys. Create a key, copy it and keep it somewhere safe: it starts with qari_ and is shown only once.

2. Install

Requires Python 3.10 or newer. This installs the alqari-mcp command; find its full path with where alqari-mcp on Windows or which alqari-mcp on macOS and Linux.

pip install alqari-mcp

3. Connect your assistant

In Claude Desktop, open Settings → Developer → Edit Config and add the server to claude_desktop_config.json. Set command to the full path of alqari-mcp, then restart the app. Cursor uses the same block in ~/.cursor/mcp.json.

{
  "mcpServers": {
    "alqari": {
      "command": "C:\\path\\to\\alqari-mcp.exe",
      "env": {
        "ALQARI_API_KEY": "qari_your_api_key"
      }
    }
  }
}

In Claude Code, and in the ChatGPT desktop app and Codex, add it with one command:

# Claude Code
claude mcp add --transport stdio alqari --env ALQARI_API_KEY=qari_your_api_key -- alqari-mcp

# ChatGPT desktop app and Codex
codex mcp add alqari --env ALQARI_API_KEY=qari_your_api_key -- alqari-mcp

Tools

Need a workflow to run from your assistant? Build one with AI, then run it with run_workflow. Package page on PyPI

Upload & run OCR

Upload a file and run OCR in one step. Send as multipart/form-data with a file field and do not set the Content-Type header manually. Usually answers HTTP 200 with the result. If OCR is not finished when the request returns (e.g. very large documents), it answers HTTP 202 with status processing — poll included_outputs.text_url until it returns 200 (404 means not ready yet).

curl -X POST "https://api.alqari.sa/services/upload-ocr?mode=premium&language=ar&process_for_chat=true" \
  -H "Authorization: Bearer qari_••••••••••" \
  -F "file=@invoice_2024.pdf"

{
  "document_id": "doc_9xKpL3mN",
  "file_name": "invoice_2024.pdf",
  "pages": 3,
  "processing_time": 4.21,
  "total_words": 512,
  "document_language": "ar",
  "text": "فاتورة ضريبية ...",
  "credits_consumed": 3,
  "remaining_credits": 4977,
  "included_outputs": {
    "text_url": "https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/text",
    "markdown_url": "https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/markdown",
    "html_url": "https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/html",
    "blocks_url": "https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/blocks"
  },
  "premium_outputs": {
    "layout_url": "https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/layout",
    "extraction_url": "https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/ocr"
  }
}

OCR text output

Retrieve the full extracted text as plain text. This is the canonical plain-text endpoint and returns text/plain.

curl https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/text \
  -H "Authorization: Bearer qari_••••••••••"

Content-Type: text/plain

فاتورة ضريبية
Invoice No: INV-0001
...

OCR structured output

Retrieve the structured OCR regions as JSON. Each element carries the text, its confidence, a 4-point bounding box, and the page number.

curl https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/ocr \
  -H "Authorization: Bearer qari_••••••••••"

{
  "results": [
    {
      "text": "Invoice No: INV-0001",
      "confidence": 0.984,
      "bbox": [
        [72.0, 118.0],
        [268.0, 118.0],
        [268.0, 138.0],
        [72.0, 138.0]
      ],
      "page": 1
    }
  ],
  "credits_consumed": 0,
  "remaining_credits": 4980
}

Bounding Boxes

OCR regions return results[].bbox as a 4-corner polygon in order: top-left, top-right, bottom-right, bottom-left. The origin is the top-left of the page.

// results[].bbox — 4-corner polygon
"bbox": [
  [x1, y1],
  [x2, y2],
  [x3, y3],
  [x4, y4]
]

Confidence Scores

results[].confidence is OCR recognition confidence for the region, in the range 0.0–1.0. This is not a business-field confidence score. Layout block confidence is returned under a different key: layout.blocks[].conf.

Layout output

Retrieve the document layout analysis as JSON: blocks, tables, figures, and page sizes. The layout block box is a rectangle [left, top, right, bottom] — different from the OCR 4-point polygon.

curl https://api.alqari.sa/services/ocr-output/doc_9xKpL3mN/layout \
  -H "Authorization: Bearer qari_••••••••••"

{
  "source": "invoice_2024.pdf",
  "pages": 1,
  "page_sizes": { "1": [826, 1280] },
  "blocks": [
    {
      "id": 0,
      "text": "Invoice No: INV-0001",
      "conf": 0.984,
      "bbox": [72, 118, 268, 138],
      "block_type": "paragraph",
      "page": 1,
      "raw_label": "paragraph"
    }
  ],
  "tables": [],
  "figures": [],
  "credits_consumed": 0,
  "remaining_credits": 4980
}

Tables & Figures

The layout response may contain detected table structures and figures, as part of document layout analysis. This is not a generic or custom-schema table extraction API.

// layout.tables[] — table structures detected in layout output
{
  "id": "table_1",
  "page": 1,
  "bbox": [72, 300, 760, 640],
  "markdown": "| Item | Qty | Total |\n| --- | --- | --- |\n| ... | ... | ... |",
  "rows": [ ["Item", "Qty", "Total"] ]
}

// layout.figures[]
{
  "id": "figure_1",
  "page": 1,
  "bbox": [499, 30, 772, 278],
  "caption": ""
}

HTML / Markdown / Blocks Outputs

Beyond text and structured JSON, additional rendered formats derived from the same OCR job are available.

AI validation

Run validation rules against an already-processed document. Pass document_id and rules_text as query params. Results are not always perfect — have a human review critical outcomes.

curl -X POST "https://api.alqari.sa/services/ai-validate?document_id=doc_9xKpL3mN&rules_text=1.%20Check%20required%20fields" \
  -H "Authorization: Bearer qari_••••••••••"

{
  "document_id": "doc_9xKpL3mN",
  "overall_verdict": "FAIL",
  "summary": "1 of 2 rules failed.",
  "results": [
    {
      "rule_number": 1,
      "rule_text": "Check required fields",
      "verdict": "PASS",
      "calculation": null,
      "detail": "All required fields present.",
      "evidence": "Invoice No: INV-0001",
      "bounding_boxes": [
        { "page": 1, "bbox": [[72,118],[268,118],[268,138],[72,138]] }
      ]
    },
    {
      "rule_number": 2,
      "rule_text": "Validate dates",
      "verdict": "FAIL",
      "calculation": null,
      "detail": "Issue date is after due date.",
      "evidence": null,
      "bounding_boxes": null
    }
  ],
  "credits_consumed": 1,
  "remaining_credits": 4976
}

Document Q&A

Ask a natural-language question about a document that has been processed for chat. Upload the document with process_for_chat=true to enable chat on it.

curl -X POST https://api.alqari.sa/services/chat/doc_9xKpL3mN \
  -H "Authorization: Bearer qari_••••••••••" \
  -H "Content-Type: application/json" \
  -d '{ "message": "ما إجمالي الفاتورة؟" }'

{
  "document_id": "doc_9xKpL3mN",
  "question": "ما إجمالي الفاتورة؟",
  "answer": "إجمالي الفاتورة هو SAR 124,500.00",
  "tokens_used": { "input": 812, "output": 24 },
  "credits_consumed": 1,
  "remaining_credits": 4975
}

Build workflows with AI

Describe what a workflow should do, and optionally attach a use-case document, sample documents or both, and the AI assistant builds it for you as a Draft. Keep chatting to change it until it's right. Start a conversation with this endpoint (pass workflow_id to work on an existing workflow), then send your messages to it.

curl -X POST https://api.alqari.sa/workflows/ai-builder/sessions \
  -H "Authorization: Bearer qari_••••••••••" \
  -H "Content-Type: application/json" \
  -d '{}'

HTTP/1.1 201 Created

{
  "id": "aib_3Fk9",
  "workflow_id": null,
  "workflow": null,
  "messages": [],
  "sources": [],
  "created_at": "2026-10-17T09:00:00Z",
  "updated_at": "2026-10-17T09:00:00Z"
}

Send a message to the assistant

Send multipart/form-data with message, files, or both. The assistant answers, asks a question back, or builds the workflow. When it builds, the workflow is saved as a Draft and workflow_changed is true. A file that can't be read is listed in unreadable instead of failing the message, and isn't charged. Each message is billed: billing breaks down the charge (OCR of the files and AI usage) and reply.credits is its total. Set your HTTP timeout to a few minutes.

curl -X POST https://api.alqari.sa/workflows/ai-builder/sessions/aib_3Fk9/messages \
  -H "Authorization: Bearer qari_••••••••••" \
  -F "message=Extract the invoice number, date, VAT number and total from each invoice and export them to Excel" \
  -F "files=@invoice_sample.pdf" \
  --max-time 300

{
  "reply": {
    "role": "assistant",
    "content": "Your workflow is ready.",
    "action": "build",
    "workflow_version": 1,
    "notes": [],
    "attachments": [],
    "unreadable": [],
    "credits": 77
  },
  "workflow_changed": true,
  "billing": {
    "ocr_pages": 3, "ocr_credits": 60,
    "llm_input_tokens": 20000, "llm_output_tokens": 3000, "ai_credits": 17,
    "credits_consumed": 77, "remaining_credits": 4923
  },
  "session": {
    "id": "aib_3Fk9",
    "workflow_id": "wf_7dR2",
    "workflow": {
      "id": "wf_7dR2",
      "name": "Invoice extraction",
      "status": "draft",
      "graph": { "nodes": [ … ], "edges": [ … ] }
    },
    "messages": [ … ],
    "sources": [
      { "filename": "invoice_sample.pdf", "chars": 1840, "truncated": false }
    ]
  }
}

Get a conversation

The messages so far, the names of the attached files, and the workflow the assistant is building, with its graph.

curl https://api.alqari.sa/workflows/ai-builder/sessions/aib_3Fk9 \
  -H "Authorization: Bearer qari_••••••••••"

Run a workflow and wait for the result

Send a document (multipart) and get the result in the same response, without polling. The request stays open while the workflow runs (OCR, extraction, AI steps), then answers 200 with the whole run: the same object GET /workflows/runs/:runId returns, including output, billing and node_runs. The run id is id here, not run_id. Only Active workflows run, and there is no Idempotency-Key. Set your HTTP timeout to at least 5 minutes. For several files, repeat file or send several files fields; the answer is { related_files, run_count, runs: [...] } with a full run per file.

curl -X POST https://api.alqari.sa/workflows/wf_7dR2/run-upload \
  -H "Authorization: Bearer qari_••••••••••" \
  -F "file=@invoice_2024.pdf" \
  --max-time 300

HTTP/1.1 200 OK

{
  "id": "run_5Qm1",
  "workflow_id": "wf_7dR2",
  "status": "completed",
  "execution_mode": "api",
  "trigger_type": "upload_document",
  "output": {
    "filename": "invoice.pdf",
    "fields": { "invoice_number": "INV-2041", "invoice_date": "2026-09-28", "total_amount": "1150.00" },
    "ocr_text": "…",
    "_workflow": {
      "run_id": "run_5Qm1",
      "workflow_id": "wf_7dR2",
      "workflow_name": "Invoice extraction"
    }
  },
  "error": null,
  "error_code": null,
  "created_at": "2026-10-04T08:48:09Z",
  "started_at": "2026-10-04T08:48:10Z",
  "finished_at": "2026-10-04T08:48:14Z",
  "billing": {
    "ocr_pages": 1, "ocr_credits": 1,
    "llm_input_tokens": 2140, "llm_output_tokens": 180, "llm_credits": 1,
    "credits_consumed": 2, "remaining_credits": 4998
  },
  "node_runs": [
    { "node_id": "upload-1", "node_type": "upload_document", "label": "Upload Document", "status": "completed", "output": { } },
    { "node_id": "ocr-1", "node_type": "ocr", "label": "OCR", "status": "completed", "output": { } },
    { "node_id": "extract-1", "node_type": "extract_fields", "label": "Extract Fields", "status": "completed", "output": { } },
    { "node_id": "output-1", "node_type": "output", "label": "Output", "status": "completed", "output": { } }
  ]
}

Run status in the answer

A 200 means the request worked, not that the run succeeded. Always check status.

Today the server waits until the run ends or pauses, so running doesn't come back yet. The wait may later be limited to 5 minutes; a run still going then comes back as running. Handle that case now.

Which one should I use?

• Get a run id (POST …/runs) for live systems. It answers at once, can be retried safely with Idempotency-Key, runs Draft workflows, and handles long documents and many files without holding a connection open.
• Wait for the result (POST …/run-upload) when you want the result in the same response: scripts, one-off jobs, short documents. It needs an API key and an Active workflow.
• Either way, a run that needs Human Review isn't finished until someone decides. Its final result is read later.
• /run-upload returns the run id as id; /runs returns it as run_id.

Timeouts and retries

• Set your HTTP client's timeout to at least 5 minutes. A short document takes seconds; a long one can take minutes.
• There's no Idempotency-Key on this endpoint. If the request times out, the connection drops, or you get a 5xx answer, the run may still finish and be billed. Resending runs the document again and bills it again.
• Before resending, look for the run in GET /workflows/{workflow_id}/runs (newest first, same API key). If you need safe retries, use Get a run id instead.

Several documents and Batch

• Related files off (default): each file gets its own run, but they run one after another while you wait. The answer lists them in runs[]; check each run's status.
• If one of several files is refused (for example 402 OUT_OF_CREDITS), the whole request answers with that error, even though earlier files already ran and were billed. Send one document per request, or use Get a run id for several.
• Related files on (needed for Batch Document Processing): all files go into one run. Each file's result is in output.documents[]; the package summary is in output.package. The run succeeds or fails as a whole.

Errors: wait for the result

Errors come back as JSON with a detail field: either a message string, or an object with message plus code or error_code. Some errors (402) put code next to detail. Read whichever is present.

Async runs

Send now, collect later — recommended for integrations and high volume. Same files, rules and billing as run-upload, but it answers immediately with HTTP 202 and a run_id in status pending. Then poll GET /workflows/runs/:runId every ~30 s until status is completed or failed.

curl -X POST https://api.alqari.sa/workflows/wf_7dR2/runs \
  -H "Authorization: Bearer qari_••••••••••" \
  -F "file=@invoice_2024.pdf"

HTTP/1.1 202 Accepted

{
  "run_id": "run_5Qm1",
  "workflow_id": "wf_7dR2",
  "status": "pending",
  "execution_mode": "api"
}

Several documents and Batch

• Related files off (default): each file gets its own run. The answer lists them in runs[]; poll each run_id.
• Related files on (needed for Batch Document Processing): all files go into one run with one run_id. Each file's result is in output.documents[] (file name, document type, confidence, fields). The package summary is in output.package (types found, required types missing, unclear files).
• A Batch run succeeds or fails as a whole: if one file can't be read, the run is failed.
• With many files, set the Output node's payload template to return only what you need. Otherwise the response includes every file's OCR data.

202 Accepted
{
  "related_files": false,
  "run_count": 2,
  "runs": [
    { "run_id": "run_5Qm1", "status": "pending", … },
    { "run_id": "run_8Lp4", "status": "pending", … }
  ]
}

By default each file gets its own run. If the workflow's Upload node has Related files on, all files go into one run and you get the single-run answer.

Run status

Retrieve the status and result of a workflow run — this is the endpoint to poll. Statuses: pending (queued, not started yet), running (being processed), completed (result in output), failed (see error and error_code), paused (waiting for a Human Review decision in AlQari — keep checking every few minutes; the run continues after the review). Stop polling on completed or failed. A run can stay pending or running for a while.

curl https://api.alqari.sa/workflows/runs/run_5Qm1 \
  -H "Authorization: Bearer qari_••••••••••"

{
  "id": "run_5Qm1",
  "workflow_id": "wf_7dR2",
  "status": "completed",
  "execution_mode": "api",
  "output": { "...": "the workflow result" },
  "error": null,
  "error_code": null,
  "started_at": "2026-10-04T09:12:03Z",
  "finished_at": "2026-10-04T09:12:41Z",
  "created_at": "2026-10-04T09:12:02Z",
  "node_runs": [
    { "node_id": "ocr_1", "status": "completed", "output": {} }
  ]
}

Run statuses

A run can stay pending for a while.

Response examples

200 OK
{
  "id": "run_5Qm1",
  "workflow_id": "wf_7dR2",
  "status": "completed",
  "execution_mode": "api",
  "trigger_type": "upload_document",
  "output": {
    "filename": "invoice.pdf",
    "fields": {
      "invoice_number": "INV-2041",
      "invoice_date": "2026-09-28",
      "total_amount": "1150.00"
    },
    "ocr_text": "…",
    "_workflow": {
      "run_id": "run_5Qm1",
      "workflow_id": "wf_7dR2",
      "workflow_name": "Invoice extraction"
    }
  },
  "error": null,
  "error_code": null,
  "created_at": "2026-10-04T08:48:09Z",
  "started_at": "2026-10-04T08:48:10Z",
  "finished_at": "2026-10-04T08:48:14Z",
  "billing": {
    "ocr_pages": 1, "ocr_credits": 1,
    "llm_input_tokens": 2140, "llm_output_tokens": 180, "llm_credits": 1,
    "credits_consumed": 2, "remaining_credits": 4998
  },
  "node_runs": [
    { "node_id": "upload-1", "node_type": "upload_document", "status": "completed", "output": { … } },
    { "node_id": "output-1", "node_type": "output", "status": "completed", "output": { … } }
  ]
}

The result is in output. Its keys depend on the workflow's nodes and the Output node's payload template. _workflow is included unless Include metadata is off on the Output node.

Still going: the same shape with "status": "pending" or "running" and "output": null.

Failed: "status": "failed", with a readable message in error and a machine code in error_code. A failed step uses no credits.

Exported PDF/DOCX/Excel: if the Output node exports files, output.exported_files[] lists them. Download each with GET {BASE}{download_path} and the same API key.

Webhooks

Add a Webhook Output node inside a workflow to deliver run results to your own HTTPS endpoint via an HTTP POST with a JSON payload.

POST https://your-app.com/webhooks/alqari
Content-Type: application/json

{
  "workflow_id": "wf_7dR2",
  "run_id": "run_5Qm1",
  "status": "completed"
}

Get the result by webhook instead

Instead of polling, set the workflow's Output node to Output type: webhook and enter your URL. When a run reaches the Output node, AlQari POSTs the output JSON to that URL (30-second timeout, redirects not followed).

• Treat the webhook as a notification: read _workflow.run_id from the body and confirm the result with GET /workflows/runs/{run_id}.
• It's sent only when the run reaches an Output node set to webhook. A failed run sends nothing. After a Human Review decision, it's sent only if the path taken after the decision leads to that Output node. Keep a slow poll as a fallback.

File Support

Accepted upload file types are governed by an allowlist: .pdf .png .jpg .jpeg .tif .tiff .bmp .webp .docx .doc .txt .md .csv .json .xml .html. Not all types behave identically for OCR.

Workflow run limits

• File types: PDF, PNG, JPG/JPEG, TIFF, WEBP, DOCX, XLSX
• Up to 20 MB per file
• Up to 100 files per request (the Upload node may set a lower limit)

Status Enums

Document processing status is persisted as one of the values below. Keep the enum families (document, run, execution mode, workflow) separate.

Idempotency

An Idempotency-Key can be supplied for upload requests: POST /services/upload-ocr and POST /services/ai-validate. Its scope is billing de-duplication only — a technical retry with the same key is billed once, while a new request with a different or absent key is billed normally. It is not an OCR/validation response re-send. Use a new key for a new upload.

curl -X POST https://api.alqari.sa/services/upload-ocr \
  -H "Authorization: Bearer qari_••••••••••" \
  -H "Idempotency-Key: 9f2b7c4a-••••" \
  -F "file=@invoice_2024.pdf"

Billing Fields

Processing responses return credits_consumed and remaining_credits to reflect credit usage. For current pricing, see the pricing page.

Error Codes

Branch your logic on the code field, not the message text. The stable codes are below:

Workflow run errors

Errors come back as JSON with a detail field: either a message string, or an object with message plus code or error_code. Some errors (402) put code next to detail. Read whichever is present.

Versioning

Paths are currently unversioned: there is no /v1/ or /v2/ prefix, no version header, and no versioned host. This may change in the future.