Retab – AI-powered document automation for developers

Retab split beats Google - BenchmarkEnd-to-end automation for your hardest

document workflows.

Extract, validate, and route documents at scale — with configurable business rules, cross-document checks, and human review for exceptions.

Try now

Get a demo

Read our docs

We process millions of pages for teams from AI startups to Fortune 500 companies

+ many more

Automate end-to-end document-driven processes in minutes.

Platform

MCP

Integrates with your coding agent

Retab exposes workflow runs, source documents, validation results, and inspection tools through MCP, so agents can inspect, debug, and improve document automations without leaving the loop.

MCP

codex mcp add retab --url https://mcp.retab.com/mcp

CLI

curl -fsSL https://retab.com/install.sh | sh

Agent Skills

npx skills add https://github.com/retab-dev/retab --skill retab

Terminal

Claude Codev2.1.3

Opus 4.7

> Create a workflow to extract data from bank statements

●Let's inspect this bank statement and turn the repeatable parts into a workflow.

↳files_inspect{ "file_id": "stmt_042", "args": { "kind": "pdf_image", "pages": [1] } }

⎿3-page statement · checking account · May 2026 · two-column transaction table

●I need the stable anchors first: account holder, statement period, balances, and transaction rows.

↳files_grep{ "file_id": "stmt_042", "pattern": "Opening balance|Closing balance|Deposits|Withdrawals" }

⎿✓anchors found · opening balance · deposits · withdrawals · closing balance

●The layout is consistent. I'll create a workflow, then add a parse step that preserves line order before extraction.

↳workflows_create{ "name": "bank_statement_reconciliation" }

⎿✓workflow_id: wf_bank_stmt · draft created

↳workflows_nodes_add{ "workflow_id": "wf_bank_stmt", "kind": "parse", "name": "parse_statement_pages" }

⎿✓parse_statement_pages · page text and table lines preserved

●Next I need structured fields and transaction rows with dates, descriptions, debit, credit, and running balance.

↳workflows_nodes_add{ "kind": "extract", "schema": "bank_statement_v1" }

⎿fields: account_number · period_start_at · period_end_at · transactions[]

↳workflows_nodes_add{ "kind": "validate", "rule": "opening + credits - debits = closing" }

⎿✓balance_check · duplicate transaction guard · required date guard

●Let's run the sample through the draft workflow and see if the validation catches anything.

↳workflows_runs_create{ "workflow_id": "wf_bank_stmt", "file_id": "stmt_042" }

⎿run_91bd · parsing pages · extracting transactions

↳workflows_runs_get{ "run_id": "run_91bd" }

⎿✓status: completed · 143 transactions · balance_check passed

●Good. I'll add a review gate only for balance mismatches or rows with missing dates.

↳workflows_nodes_add{ "kind": "human_review", "when": "balance_check.failed || transactions[].date.missing" }

⎿✓review_bank_statement · triggers only on reconciliation failures

●Workflow drafted: bank statement reconciliation.

[PDF bank statement]
        |
        v
[parse pages] -> [extract fields + transactions]
        |                  |
        v                  v
[balance check] ----fail----> [human review]
        | pass
        v
[clean transaction JSON]

▎

Get started for free. No credit card needed.

Start free trial Contact Sales

Retab split beats Google - BenchmarkEnd-to-end automation for your hardest

document workflows.

Extract, validate, and route documents at scale — with configurable business rules, cross-document checks, and human review for exceptions.

Try now

Get a demo

Read our docs

We process millions of pages for teams from AI startups to Fortune 500 companies

+ many more

Automate end-to-end document-driven processes in minutes. Extract data, and validate it with your business rules.

Maximize straight-through processing by detecting and routing edge cases for human review.

Retab contains everything you need to go from prototype to production without stitching brittle tools together.

Automate end-to-end document-driven processes in minutes.

Platform

Document workflows

End-to-end orchestration for complex pipelines. Build multi-step workflows that parse, split, extract, validate, and route with versioning and durability out of the box.

Agent builder

Describe your document pipeline in natural language. Our agent scaffolds the entire workflow — from ingestion through validation to output — in seconds.

Set up a workflow that reads a source PDF, uses borrower JSON and a short instruction, routes exceptions for review, and sends approved output to a webhook.

Great, I will draft the steps, connect the handoffs, and fill in the key settings for you.

Action: Add step

Action: Connect steps

Done. Settings are in place. Run a test batch, review outputs, and publish when the team is comfortable.

Human-in-the-loop

Flag uncertain extractions for human review. Set confidence thresholds, route edge cases to reviewers, and approve or correct results before they hit your systems.

Evals & monitoring

Benchmark extraction accuracy across document types, track drift over time, and ship changes with confidence using built-in evaluation suites.

Confidence scoring

Quantify extraction certainty with our novel k-LLM consensus approach — run multiple vision language models on the same document and score agreement field-by-field before it reaches your pipeline.

Smart routing

Automatically match each document to the right model tier based on complexity. Optimize cost and accuracy without manual configuration.

Source grounding

Trace every extracted field back to the exact region in the original document. Visual proof that builds trust and simplifies audits.