logo
  • Environments
  • Enterprise
  • Pricing
BusinessJul 2, 2026

Automated VAT Return from Receipts and Invoices

EigentEigent
Automated VAT Return from Receipts and Invoices with Nebius Token Factory + Eigent
Automate Everything with
AI Workforce on Desktop
Download Eigent

From a Folder of Receipts to a Finished VAT Return

VAT recovery is the kind of finance work that eats an afternoon: open every receipt, read the numbers off crumpled photos and scanned PDFs, decide what's recoverable, and total it up without slipping a decimal. In this use case we handed the whole job to Eigent — running GLM-5.2 on Nebius Token Factory — and asked for two deliverables the finance team can actually use: a structured XLSX and a shareable HTML report.

The rule we cared about most: do not guess any uncertain information. Anything the model couldn't read with confidence had to be flagged for a human, not invented.

1Point Eigent at the Model — Nebius Token Factory

The demo starts in the model settings:

  1. Go to Home → Agents → Models → Nebius Token Factory.
  2. Enter your API key, then refresh the model list.
  3. Pick the model to run the task on — for this demo we selected GLM-5.2.

Nebius Token Factory serves the model behind Eigent's harness, so the exact same workflow runs on whichever model you select.

The Prompt

Back on the home page, we entered the task and attached the receipt and invoice files:

Please process all receipts and invoices in the "VAT" folder, including photos, scanned PDFs, and digital invoices.

The final output should include only two files:

vat_return.xlsx — one row per receipt or invoice; list all extracted fields; show whether each item is eligible for VAT recovery; show the recoverable VAT amount for each eligible item; include the exclusion reason for non-recoverable items; clearly flag items that require manual review; and include a summary sheet showing the total recoverable VAT amount.

vat_return.html — a self-contained HTML file that can be opened directly and shared with the accounting team, showing all VAT recovery items, the recoverable VAT amount for each, excluded items and the reasons for exclusion, items requiring manual review, and the total recoverable VAT amount.

Do not guess any uncertain information.

2Read Every File

Eigent starts by reading all of the uploaded files — mixing photos, scanned PDFs, and digital invoices in the same pass. No manual re-typing, no per-format handling from the user.

3Extract the Key Data

From each document it pulls out the fields that matter for a VAT return: supplier, date, net amount, VAT amount, VAT rate, and totals. Every row is one source document, so the output stays auditable back to the original receipt.

4Decide VAT Eligibility

This is the judgment step. For every receipt and invoice, Eigent determines whether the item is eligible for VAT recovery, records the recoverable amount where it is, and writes an exclusion reason where it isn't. Anything it can't read with confidence is flagged for manual review rather than guessed — exactly as instructed.

5Two Finished Deliverables

Finally, Eigent produces the two finance-operation documents:

  • vat_return.xlsx — a structured workbook with one row per document, all extracted fields, an eligibility flag, recoverable VAT per item, exclusion reasons, manual-review flags, and a summary sheet carrying the total recoverable VAT.
  • vat_return.html — a self-contained, interactive report that opens in any browser: recoverable items and amounts, excluded items with reasons, items needing review, and the headline total recoverable VAT — ready to send straight to the accounting team.

The Result

The final output is clean, clearly structured, and ready for the finance team to use. The XLSX is the auditable working file; the HTML is the shareable summary. Because uncertain items are flagged instead of guessed, a reviewer knows exactly where to spend their attention — and can trust everything that wasn't flagged.

Run It Yourself

  1. Put all your receipts and invoices — photos, scans, and digital files — into one folder.
  2. In Eigent, go to Agents → Models → Nebius Token Factory, add your API key, refresh, and select a model (we used GLM-5.2).
  3. Paste the prompt above, attach the folder, and send.
  4. Open vat_return.xlsx for the working file and vat_return.html to share with accounting.

What to Try Next

Add a currency column and convert every recoverable amount into your reporting currency using the invoice-date exchange rate.

Split the summary sheet by VAT rate (standard / reduced / zero) so the totals map directly onto the boxes of your VAT return form.

Generate a second tab listing only the manual-review items, with the specific field that was unreadable, so a human can clear them in one pass.

Tips for Better Results

  • Keep it to two outputs. Asking for exactly vat_return.xlsx and vat_return.html keeps the deliverables clean and predictable instead of a scatter of intermediate files.
  • Forbid guessing explicitly. "Do not guess any uncertain information" turns ambiguity into a manual-review flag — which is what makes the result trustworthy for finance.
  • Demand an exclusion reason per item. Requiring a reason for every non-recoverable line makes the report auditable, not just a number.
  • Separate the working file from the shareable one. The XLSX stays editable and auditable; the self-contained HTML is what you actually hand to the accounting team.

Other use cases

Long-Horizon Task: GLM-5.1 vs GLM-5.2 on Eigent

Long-Horizon Task: GLM-5.1 vs GLM-5.2 on Eigent

Do a deep-dive research on 26 companies in the AI infrastructure ecosystem — the most certain main thread of the entire AI value chain. Cover these 6 sub-sectors (pick representative companies in each, from large-cap leaders down to smaller players): AI Data Center (compute infrastructure / build-out); GPU / AI Chips (training & inference silicon, ASICs, IP); Servers, Networking & Optical Modules (switches, NICs, optical interconnect); Power, Liquid Cooling & Energy Storage (power supply, thermal, energy management); AI Cloud / Compute Platform (hyperscalers, GPU clouds, compute-rental platforms); Supporting Ecosystem (HBM / advanced packaging, foundry, connectors & other critical components). For each company, research: company name, sub-sector, HQ / country; core products and its specific role in the AI chain; public or private (ticker + exchange if listed; if private, note latest valuation / funding round); market cap or valuation size (used for ranking); positioning and moat in the ecosystem (1–2 sentences); key customers / competitors. Ordering: within each sub-sector, rank from largest to smallest (by market cap / valuation). Structure the whole thing top-down: from the full hardware-ecosystem landscape → down to each individual company. Output requirements: First, generate a structured data file ai_infra_data.json — containing all 26 companies with the fields above, the 6 sub-sector classifications, a public/private flag, and a cross-company comparison matrix (sub-sector × key dimensions). Then generate a polished HTML report from that JSON: include an ecosystem landscape / layered diagram, sector sections, company cards, a clear visual indicator for public vs. private (tags or color coding), a market-cap ranking chart, and a sortable/filterable comparison table. Make the design professional, information-dense, and interactive. Verify the research data for accuracy first (listing status, tickers, valuations — use the latest figures and cite sources), then generate the report. Send the task in single-agent mode.

Build 10 Chinese New Year HTML5 Games with Eigent

Build 10 Chinese New Year HTML5 Games with Eigent

Build 10 separate and COMPLETE games with topics related to 2026 Chinese New Year (Horse) in HTML, CSS and JS (no libraries). Games must be fun, original, polished, mobile-friendly. Include scoring, scaling difficulty, restart buttons, and smooth visuals. Cover: arcade, puzzle, endless runner, reaction, strategy, memory, 2-player local, idle, retro pixel, and 1 experimental game.

Build a 3D Snow Bros Platformer with Gemini 3.1 Pro

Build a 3D Snow Bros Platformer with Gemini 3.1 Pro

Create a modern 3D side-scrolling platformer inspired by Mario, combined with Snow Bros mechanics. The player can shoot snow projectiles to freeze monsters into snowballs, then kick them to chain into other enemies. Include a scoring system, lives display, scaling difficulty, and a restart function with rich 3D layered environments.

Automate everything with AI workforce on desktop
Download Eigent

Try Eigent today

Download the open-source desktop app. Your AI workforce, running on your machine.

Download Eigent
Eigent

Get the latest updates, tutorials, and releases on AI workforce automation.

ProductEigentEnvironmentsPricingEnterprise
ExploreSolutionsUse CasesSkillsPluginsBlogs
DevelopersDocsGitHubCAMEL-AIOpen Source FundPartner
DownloadFor open source
CompanyAbout UsBrandCareersTerms of UsePrivacy PolicySecurity & TrustCookie PolicyRefund & Trial Policy

All rights reserved © 2026 EIGENT UK LTD

Eigent 1.0 New Version Released !download