Connect your books.Ask your AI.
Sync QuickBooks into a graph Claude, ChatGPT, or any MCP client can reason over. Statements that tie, a plan that rolls off your actuals, and your numbers beside public companies. Nothing writes back until you post an entry. When you trust it, it drafts the close.
Nothing reaches QuickBooks until you post an entry. Read, analyze and plan first.
Add the SEC graph next to your books and ask how your margins compare to public companies in your niche.
Statements export as validated reports, straight from your ledger.
Your board pack is out of date the day you send it
An export keeps the numbers and loses everything behind them. Keep the books on a ledger your AI can query, and the answer is current and you can ask why.
Start with the answers. The close comes last.
Reading your books writes nothing back, so you can put RoboLedger to work on questions today and hand it the close once you trust it.
Connect your books
Link QuickBooks with one-click OAuth, then add one MCP address to Claude, ChatGPT, or any MCP client and sign in. Add the SEC filings connector beside it to compare against public companies.
Ask, share, plan
Ask why a number moved and open it down to the facts behind it. Send the board a statement that ties, roll a plan forward from your actuals, and see where you stand against peers.
Approve the close
When you trust it, the entries arrive drafted and the rule engine checks the period. You review and approve; nothing writes back to QuickBooks until you post an entry.
From the first question to the close
Six views of the same live books. Read the statements, open any number, plan forward, and ask in plain English. When you trust it, the close and the inbox are there too.
Live statements & validated reports
See the Balance Sheet and Income Statement straight from the ledger on any day of the month, no close required, with Cash Flow and Equity as your transactions are classified. When it is time to send, generate a validated report package.
- Multi-period and comparative reports with a Draft → Under Review → Filed lifecycle
- Export Tavi and XBRL 2.1 reports; verification view checks US-GAAP concept relations
- Share snapshots with your board, lenders, and investors through publish lists
Open up any number and see how it was built
When someone asks where a number came from, open it. Every statement, schedule, metric, and disclosure is an Information Block, and the Explorer renders any of them through the same view projections, so the rendered figure and the facts behind it are one toggle apart.
- Six views per block: rendered, chart, facts, elements, validation, and business rules
- Compute a metric for a new period on the spot and watch its time series extend
- Deep-linkable state (?block= / ?view= / ?scenario=) and CSV export on every view
Plan forward from the books you just closed
Your closed months and your forecast in one monthly grid. Switch to a scenario and see what it does to cash, with its levers sitting right beneath the statements they drive.
- Income Statement, Balance Sheet, and Cash Flow month by month — actuals and forward columns in the same row
- Switch scenarios in place; the assumption levers driving the forecast render in the same grid
- Trailing-window control, CSV export, and shareable scenario links
Ask about your books in plain English
Get the answer yourself instead of waiting on a report. Query your ledger in the AI Console, or connect Claude, ChatGPT, or any MCP client and ask from the chat you already use.
- Natural language → Cypher, with the generated query shown and results one copy away
- Grounded in your actual ledger — transactions, trial balance, statements
- Add one MCP address to your AI client and sign in to run RoboLedger tools — no install
Close the period with guard rails
The entries get drafted; you review and approve. A guided, sequential close backed by a fiscal calendar and a rule engine, so no one locks an unbalanced or out-of-order period.
- Fiscal-calendar bootstrap with human-readable blockers (period incomplete, sync stale, out of sequence)
- Schedules post depreciation & prepaid entries to draft; balanced-draft gating before you lock
- Optional write-back to QuickBooks on close, and reopen-with-audit-reason
An inbox for your books
Every transaction lands as a typed business event — captured, classified, then committed. AI proposes the entry; you set the rules and nothing posts without a decision.
- AI pre-classifies invoices, bills, payments, and receipts
- Preview exactly what would post — matched handler, planned journal entries, validation errors — before it hits the GL
- Approve or reject in one click — or enable autopilot for trusted sources
Your agent has the wider surface
Every write operation on the platform becomes an MCP tool the day it ships. Screens follow where they earn it — so there are things your AI assistant can do today that don't have a button yet.
One connector per company, side by side in the same chat. Your AI reads each set of books and answers across all of them.
Author a forecast scenario
Name the scenario, set the horizon and base period, assert the driver levers, then derive the forward slice. The Plan grid picks it up on the next load.
Define a new metric
Declare a metric block and its operands once; the Block Explorer computes and charts its standing time series from then on.
Write an event-handler rule
Teach the ledger how a new kind of business event should post — the same DSL rows the Inbox reads when it pre-classifies a transaction.
Push an approved entry back out
Publish a committed event to the source-of-truth system, and manage the obligations and schedules that feed the close.
Compare across companies
Build a fact grid over canonical concepts to line your numbers up against SEC filers on the same taxonomy.
Run the graph itself
Rebuild the analytical graph after a batch of writes, take a backup, or stand up an isolated workspace to try something out.
Add this address as a connector in Claude, ChatGPT, or any MCP client, sign in, and choose the company. Add it again for each company you look after. It can read your books and drive these tools directly — with you approving each write. No install, no key to paste. The same tools back the in-app AI Console, so nothing is locked behind a desktop app.
https://api.robosystems.ai/v1/mcpPowered by RoboSystems
Your numbers live in a financial knowledge graph — a unified three-block model of taxonomy, events, and information — so every figure is traceable to its source, queryable in plain English, and validated before it's filed
Data Sources
- • Chart of accounts
- • Journal entries
- • Transaction data
- • 10-K/10-Q filings
- • XBRL standards
- • Taxonomy data
- • Direct bank & card transactions
- • Account balances
Three-Block Model
Applications
Accounting and financial reporting
Ask, plan, and close from Claude, ChatGPT, or any MCP client
Python, TypeScript, and REST clients
Automated Sync
Keep connected sources in step with your ledger — sync on demand or on a schedule
Unified Data Model
Single source of truth connecting accounting, reporting, and regulatory data
Graph Database Performance
Fast, interactive queries on complex accounting relationships powered by knowledge graph technology
Open by default
RoboLedger is open source and built so you're never trapped.
Open source
Apache-2.0. Inspect every line — see exactly how each number is produced.
You own your data
Your books, your graph. Nothing locked in a vendor black box.
No lock-in
Fork it and take it in-house whenever you want.
Your people run it. Harbinger trains them.
Harbinger FinLab implements RoboLedger and trains the people who’ll work in it every day: your own accountant, or the fractional CFO or bookkeeper you work with, and their staff. They work from the AI chat they already use. AI drafts, they approve, and they do every step. It starts with a live demo, at no cost.
Meet Harbinger FinLabRun RoboLedger across your whole client book. We walk you through the first client on a screenshare and train your staff; your clients stay yours.
Your people set up RoboLedger with us on a screenshare, then we train them to run it. No controller yet? We’ll point you to a practice that runs RoboLedger.
Each company keeps its own books on RoboLedger, run by its own accountant, and its filed reports land in your graph as data, not PDFs. The investor side is early, and we say so.
Harbinger runs its own month-end close on RoboLedger, and so does RoboSystems. We train the workflow because we work in it every month.
Nobody from Harbinger posts, approves or closes a period on your books, or signs in to them. Sign-off is yours, by name.
Questions, answered
The things finance teams ask before they connect their books.
Do I have to leave QuickBooks?
No. RoboLedger syncs your QuickBooks chart of accounts and transactions. Reading, analysis, and planning write nothing back; approved entries go back to QuickBooks only when you post them. It's an AI layer on top of your books, not a replacement for them.
Where does my financial data live?
In your own knowledge graph on the RoboSystems platform. RoboLedger is open source, so you can inspect exactly how every number is produced — and you can fork it and take it in-house whenever you want.
Do I need to know XBRL or US-GAAP?
No. AI Auto-Map proposes the US-GAAP mapping for your chart of accounts with confidence scores, and the rule engine validates every entry before you lock a period. You review and adjust — the taxonomy work happens for you.
What does the AI actually do?
It answers questions about your books in plain English, opens any number down to the facts behind it, rolls plans forward, and compares you to public companies. When you're ready, it also triages each transaction in the inbox and drafts the close. You stay in control at every step — one-click approve/reject, dry-run previews of what would post, and balanced-entry checks before anything commits.
Can I forecast, or is this only historical accounting?
You can plan forward. Once you've closed a month, the Plan surface puts your monthly statements and a scenario's assumptions in one grid spanning the actuals/forecast seam — so the forecast is driven by the same ledger the close produced, not a spreadsheet copied out of it. Scenarios are authored from your AI assistant over MCP today; the in-app editor is on the way.
Are there things my AI assistant can do that the app cannot?
Yes — deliberately. Every write operation on the platform becomes an MCP tool the day it ships, while screens follow where they earn their place. Authoring a forecast scenario, defining a new metric, and writing an event-handler rule all run from Claude, ChatGPT, or any MCP client today without a dedicated form in the app. Anything you drive that way lands in the same ledger and shows up on the same screens.
I look after several companies. Can I see them all here?
Yes. Each company's books live in their own isolated graph — connect QuickBooks per company and switch between them from one account. In your AI chat, add the RoboLedger connector once per company and ask across all of them. It's built for fractional CFOs, controllers, and firms who look after more than one company. Harbinger FinLab walks you through your first client on a screenshare and trains your staff, so the second client is a playbook, not a project.
How do I get started, and what does it cost?
Create an account, then provision a graph on the RoboSystems platform — enabling RoboLedger gives you access to this app. Plans and provisioning live on the RoboSystems side; once you've enabled RoboLedger, connect QuickBooks and ask your AI the first question. Nothing writes back until you post an entry.
What if we don't have anyone to run it?
RoboLedger is run by your own accountant, bookkeeper or fractional CFO, working from the AI chat they already use. If you don't have one yet, Harbinger FinLab will point you to a practice that runs RoboLedger. Harbinger implements and trains; it never keeps anyone's books.
See it on your own books
Connect QuickBooks and ask your AI the first question. Nothing writes to QuickBooks until you post an entry.
Already have an account? LoginView on GitHub