your
Excel
on one side.
your
documents
on the other.
AgentXL works the gap between them.
Point AgentXL at a folder of PDFs, agreements and statements. Open your workpaper. The agent sees both — cross-references rows against source documents, verifies the numbers, and writes the answer into your cell with a citation beside it pointing to the exact document, page and line. You never open a source file again.
A 47-second audit. Forty rows. Two source documents. Zero file switching.
A real lease-agreement audit captured end-to-end: AgentXL reads the SPV document pack, populates the extracted values, attaches a citation to the exact page for every row, and flags rows that don't reconcile. Pause anywhere — every number on screen is traceable.
Connected workbook, agent ready
Lease audit workpaper open. AgentXL taskpane docked, folder of source docs connected.
Reads, then plans
Agent describes its plan in plain English — populate the value and source columns, link every row to a citation.
Cells filled. Each one cited.
Extracted values land in column D. The Source column gets DOC1–6_ARX.pdf, p.2 precision next door.
Verdict, summarised
37 PASS · 1 FAIL · 2 FLAG. Each flagged row points the reviewer at the precise clause that didn't reconcile.
Every audit lives between two windows — and no tool has connected them.
On one side, a structured Excel workpaper. On the other, a folder of PDFs, agreements and statements where the truth lives. Between them sits a wall of manual labour — opening files, copying values, switching context hundreds of times per engagement.
The way it works today
- 01Open the workpaper. Open the document folder in a second window.
- 02Pick a row. Hunt for the matching PDF. Open it. Find the clause.
- 03Copy the value. Switch back to Excel. Paste. Note the source.
- 04Repeat 400 times. Reconcile. Re-open. Argue about what you saw.
- 05Cloud AI tools could help — but client data can't leave the machine.
One connection. Both sides visible.
- ✓Point AgentXL at the document folder once. The connection is persistent.
- ✓The agent sees the active sheet and the folder simultaneously.
- ✓Ask in plain English. Answers land in cells, with a citation pointing to the exact document, page and line right beside them.
- ✓Reconcile rows, populate columns, verify totals — across many docs at once.
- ✓Local-first. Only the relevant excerpt reaches the model. Nothing else leaves.
Four things no other tool does at once.
Chat tools make you bring the work to them. Excel plugins bolt onto formulas. Cloud copilots can't touch client files. AgentXL is the first to do all four — connected folder, both sides visible, fully local, and every answer cited not just to the document but to the exact page, section and line.
Connect once.
Compare everything.
Point AgentXL at a folder. Every document in it is now connected — persistently. The active Excel sheet is the anchor. The agent compares any row, cell, or section against any document in the folder. One connection replaces dozens of manual file opens.
Cross-sheet,
cross-document intelligence.
AgentXL works across multiple Excel sheets and many source documents in a single connected session. Reconcile a tickmark column against statements; populate a schedule from contracts; verify a total against its sources — all without touching a file.
Every answer cited.
Reviewers verify in one click.
Results never land alone. Every value written into a cell carries the source document and the exact location — file, section, page, even the line — in the column next door. A reviewer hovers, clicks once, and opens the precise paragraph. Audit trails are produced as a side-effect of doing the work.
Local-first.
Documents stay on your machine.
The folder lives on your machine. The Excel sheet lives on your machine. AgentXL runs locally. Only the small, relevant excerpt reaches the model — never your full document set. This clears the compliance bar that blocks cloud chat tools in regulated environments.
Self-correcting. Built for real document sets, not demos.
Folder index
Build a local, persistent index of every document. Match candidates by entity, date, section heading.
Targeted retrieval
Open the right pages, not the whole file. Pull clauses, tables and figures that bear on the question.
Cross-reference
Hold the active sheet and the retrieved excerpts side by side. Reconcile values across many documents.
Self-check
Re-derive the number from primary sources. Flag conflicts. Refuse to write what it can't cite.
Into the cell — with the receipt
Value goes in the target cell. The citation goes in the adjacent column — document, section, page, line. Reviewer clicks once and lands on the exact paragraph.
A chat agent makes you bring the work to it. AgentXL goes to where the work already lives.
General-purpose agents are remarkable. They are also browser-shaped. AgentXL is Excel-shaped — a taskpane, a connected folder, results in cells.
A finance or audit professional with an Excel sheet on screen and a folder of source docs on disk.
They spend a significant portion of their day manually bridging the two. They work in a regulated environment where cloud AI tools cannot touch client files.
The Analyst
2–7 years experience. Spends 40–70% of engagement time on mechanical document extraction. Discovers AgentXL through a demo video or a colleague who's already using it.
The Manager
Engagement speed bottlenecked by analyst throughput, not judgment. Needs results that are traceable and reviewable — not AI-generated guesses.
The Innovation Lead
Partners demanding AI adoption; cloud tools blocked by risk and compliance. AgentXL clears the bar: local-first, no telemetry, no account, open-source.
Three commands. Connect a folder. Open Excel.
No account. No telemetry. No infrastructure to set up. AgentXL runs entirely on your machine and ships with sensible defaults for audit, due diligence and fund accounting workpapers.