AI agents for finance teams that show their work.
We build agents and automation for research, documents and reporting. Every figure comes with its source, and a person approves what matters.
Working headline. Final tagline pending Alex's decision.
Practical agents for the work finance teams repeat every week.
Research with citations
Reads filings, reports and notes, and answers with the source next to each claim.
[ services → ]Document processing
Turns PDFs and scans into spreadsheets or a database, with every value traced to its page.
[ services → ]Routine multi-step workflows
Chains the steps people do by hand, with approval points where a decision belongs to a person.
[ services → ]Reporting and reconciliations
Prepares reports and matches records, and lists every difference it could not explain.
[ services → ]Internal assistants
Answers questions over your own documents, and says when the documents do not contain the answer.
[ services → ]Private setups
Runs on open-source models inside your environment when data cannot leave it.
[ services → ]Four rules every agent we build follows.
Every figure has a source
Page, line and quote are stored with the value.
Checks run in code
Quotes and totals are verified by plain code, not by the model.
A person decides
Anything that matters waits in a review queue.
Limits are stated
Missing data is reported as missing, not guessed.
A wrong figure costs more than a slow one.
Finance teams work with documents that must be right: filings, statements, data rooms, board packs. We design for that. The agent shows where each number came from, code checks what can be checked, and a person approves the result.
The studio is led by someone with nine years in finance, so the workflows start from how analysts actually work. More on finance →
Two public samples you can open and inspect.
Both run on synthetic or public data. Model replies in the offline demos are scripted.
PE deal screening agent
Reads a teaser and returns a deal card where each figure carries a page and a checked quote.
[ see sample → ][ SAMPLE · SYNTHETIC DATA ]Financial statement spreading agent
Moves statement lines from a PDF into a workbook and keeps a trace sheet back to the source page.
[ see sample → ]From a written brief to a monitored deployment.
You describe the process and share examples. A written brief is enough to start.
A first version on your sample documents, so you see real output early.
We test against cases you provide and report what passes and what does not.
The agent goes live with logging, and we watch how it behaves.