
OfinTech Desk
Finance AI where every figure traces to its source
12 followers
Finance AI where every figure traces to its source
12 followers
Open-source AI workspace for finance teams where every figure traces to its source. Connect QuickBooks, Xero, Zoho Books or Odoo read-only and get statements, month-end close, KPIs and an assistant in Arabic or English. The AI explains; it never invents a number. Pro adds payment settlements, learning from your team's decisions and the Ofin model: no GPU, 0.03 ms per bank line on a 2-core CPU. Runs on your own servers.
This is the 2nd launch from OfinTech Desk. View more

OfinTech Desk
Launching today
Open-source AI workspace for finance teams where every figure traces to its source. Connect QuickBooks, Xero, Zoho Books or Odoo read-only and get statements, month-end close, KPIs and an assistant in Arabic or English. The AI explains; it never invents a number. Pro adds payment settlements, learning from your team's decisions and the Ofin model: no GPU, 0.03 ms per bank line on a 2-core CPU. Runs on your own servers.







Free
Launch Team

Hi Product Hunt š I'm Mohammed Ahmed, founder of OfinTech.
Every finance team I've worked with asks the same question about every report: "where did this number come from?" Most AI tools answer with confidence, not evidence. So I built OfinTech Desk around one rule: every figure has a source. Click any number and open the chain behind it ā the computation, the ledger rows that fed it, the document, and the exact page.
How it works
⢠Deterministic code does the math ā same input, same answer.
⢠A classifier makes closed decisions, with a measured probability.
⢠The AI only explains. It never invents a number. Use any LLM provider, a local model, or none.
Free & open source (Apache-2.0)
⢠Desktop app + self-hosted Team Server
⢠Read-only connectors: QuickBooks, Xero, Zoho Books, Odoo + an open API
⢠Statements, month-end close, 35 KPIs, cost centres, VAT return figures
⢠Document search with OCR, audit log, encrypted storage
⢠Arabic, English, Spanish and French
Pro & Enterprise
⢠Payment settlements: payouts matched to the bank, with every difference labelled
⢠Learns from your team's own accept/reject decisions ā month two is more accurate than month one
⢠Accept matches without a person, above a threshold you set, fully audited
⢠The trained Ofin model: no GPU needed. On a 2-core CPU it classifies a bank line in 0.03 ms ā a 500-page statement in under a second ā and learns from your data in 0.3 s.
⢠Embedded read-only finance agents for your own product
Everything runs on your own servers ā your data never leaves your network. Licences at ofintech.ai.
I'd love your feedback, especially from accountants and finance leads: what's the one report you never fully trust? š
click any number and see the ledger rows behind it, love that Mohamed. congrats!