Getting started
What does TaxClaw actually do?
TaxClaw reads common tax PDFs/images (W-2s, 1099s, K-1s) and extracts the key boxes into structured data you can review and export. It's built to save you from manual re-typing.
How do I install it?
Install TaxClaw as an OpenClaw skill, then open the local app at http://localhost:8421. You'll also need Ollama running for Local mode.
What tax forms does it support?
W-2, 1099-NEC, 1099-INT, 1099-DIV, 1099-DA, 1099-B, 1099-MISC, 1099-OID, K-1, and consolidated-1099.
Does it work on Windows / Linux / Mac?
Yes-any OS that can run OpenClaw + Ollama should work. Local performance depends on your CPU/GPU and the OCR model you choose.
Privacy & data
Does my tax data leave my computer?
In Local mode, no-everything stays on your machine. In Cloud mode, the document is sent to Claude for extraction.
What's the difference between Local and Cloud mode?
Local mode uses Ollama (recommended: glm-ocr) and keeps data fully private, but accuracy can vary by scan quality. Cloud mode is typically more accurate, but requires an API key and a privacy acknowledgment.
Where is my data stored?
Locally, in the TaxClaw/OpenClaw workspace on your computer (plus whatever exports you download). Cloud mode still saves your results locally, but the uploaded document is processed by the cloud provider.
Extraction quality
Why are all my fields empty after uploading?
Most often the OCR model isn't running/selected, or the document is a low-quality scan (blurry, skewed, tiny text). Switch to glm-ocr, re-run extraction, and try a cleaner PDF/image if you can.
What's the best local model to use?
Use glm-ocr in Ollama-it's the best default balance of OCR + form understanding for TaxClaw. If you're not getting text, confirm Ollama is running and the model is actually installed.
What do the confidence scores mean? (🟢🟡🔴)
They're a quick "how sure is the model" signal for each field. 🟢 = likely correct, 🟡 = plausible but check it, 🔴 = treat as a guess.
What does "Needs review" mean?
TaxClaw flags a doc as "Needs review" when one or more important fields are missing or low-confidence. It's not an error-just a heads-up to eyeball the highlighted fields.
My doc was classified wrong - how do I fix it?
Change the document type (e.g., 1099-INT vs 1099-DIV) and re-run extraction. Classification matters because each form has different boxes and field rules.
Can I edit extracted fields if they're wrong?
Yes-edit the fields in the review UI, then export again. Your edits are what get exported.
1099-DA & crypto
I uploaded a 1099-DA. What do I do next?
Use TaxClaw to extract the proceeds and transaction details, then reconcile cost basis in a crypto tax tool (we recommend Koinly). Export your data and use it as your starting point.
What is cost basis and why do I need it?
Cost basis is what you originally paid for the asset (plus certain fees/adjustments). You need it to calculate gains/losses-proceeds alone only tells you the sale side.
TaxClaw extracted my proceeds but not cost basis - is that normal?
Yes-many 1099-DA forms don't include complete cost basis, or the broker reports it separately/partially. That's why cost basis reconciliation (e.g., in Koinly) is usually still required.
Exporting & filing
How do I export my data?
Go to Exports and choose CSV (wide or long), JSON, or download the original ZIP. Export after you've reviewed and corrected any flagged fields.
What's the difference between the export formats?
CSV (wide) = one row per document with lots of columns; CSV (long) = one row per field/value (better for analytics and pipelines). JSON preserves structure and is easiest for developers; ZIP is the original files.
Can I share exports with my tax preparer?
Yes-CSV is usually the easiest for a preparer to scan, and the ZIP is helpful if they want the originals. Always spot-check totals and key IDs (SSN/EIN, payer name, amounts) before you send.