Pennyweight
Keep your human happy. They hate bookkeeping.
13 followers
Keep your human happy. They hate bookkeeping.
13 followers
Keep your human happy. They hate bookkeeping. Traditional accounting software was built for humans to operate – then AI was added as an afterthought. Pennyweight starts from the other end: a ledger whose primary user is your AI. Connect any assistant you already use. What's new: it remembers. Every decision is stored with its reasoning, your corrections are pinned so no future session undoes them, and every AI action is auditable. Other tools reset each chat. Ours picks up where it left off.
Free Options
Launch Team / Built With
Render: Get up to $100K in startup creditsThe fastest way to deploy and scale apps and agents
Promoted
Bookkeeping has been my most dreaded job in all the startups I've built. AI, please make me forget...
Pennyweight started as tooling I built for my own books, with one goal: think about bookkeeping as little as possible. It's focused on the piece every LLM will need and none of them have: persistent, well-structured financial data. It's built to outlast model generations — as AI evolves, Pennyweight stays the memory underneath and becomes better at it.
Pennyweight is different from traditional bookkeeping software: it's AI-centric, not human-centric. Connect it to the assistant you already use — Claude, ChatGPT, Gemini, Grok — link your bank, and ask.
Security is at the core: your data is encrypted with a key only your passphrase can derive — even a stolen copy of our database would be unreadable.
We want to make the tooling amazing, and that only comes from real usage and feedback. So the trial is six months, free, no card – your usage is on us; the feedback is what we're after.
I would love to hear your feedback, ideas, grievances...
Interesting product! How does financial transactions flow in the LLMs context/memory? Are using plaid? Does this work only for US bank accounts?
@manuel_angel great questions! I'll add these to our Q&A.
Context/memory flow: Short version: the model's context is a scratchpad, the ledger is the memory. Transactions live encrypted in Pennyweight; per question, MCP tools return just the needed slice — filtered rows for inspection, server-computed aggregates for reports, so even "summarize my year" doesn't push raw history through the context window. What persists between sessions lives in the ledger: decisions with their reasoning, your pinned corrections, the audit log. A fresh chat starts with zero context and full memory.
Plaid: yes, for live bank sync.
US-only?: live sync is US institutions via Plaid; outside the US you add accounts manually and import statements/CSV — multi-currency is supported natively. Import using LLMs is already pretty trivial, and we make it easier.