PocketLedger
A local AI bookkeeping tool for freelancers. It uses a local vision-language model to extract structured expense information from receipt images or text, with data stored in a local SQLite database and no cloud account required.
Timeline (last 90 days)
PocketLedger is a project submitted by developer SaikatDash in the second week of the Hacktoberfest open-source AI challenge.
PocketLedger targets solo operators such as independent designers, developers, photographers, and consultants, positioned as an alternative between heavy accounting software and cloud receipt tools.
PocketLedger's core flow is: the user uploads a receipt image or pastes text, local AI extracts the merchant, date, amount, currency, category, and description to generate an editable draft, and the user reviews and confirms it before writing to the ledger.
PocketLedger stores data in a local SQLite database and requires no cloud account registration.
PocketLedger's backend uses the Python standard library and its frontend uses HTML, CSS, and vanilla JavaScript, introducing no complex frameworks.
PocketLedger calls the Qwen2.5-VL 3B model via Ollama for receipt understanding and constrains the output fields with a structured JSON schema.
PocketLedger's product design is that AI only generates drafts, and every expense requires human confirmation before being written to the ledger, making no irreversible bookkeeping decisions.
The business model given by the PocketLedger author is a one-time-paid desktop deployment, supplemented by optional onboarding guidance and support services.
The author candidly admits that PocketLedger has not yet completed a formal first-user trial.
PocketLedger is open-sourced on GitHub, and the code repository is SaikatDash/sideHustlers.
1 more timeline events
All Facts (11)
Using local inference means users need sufficient local storage and hardware to run the models, and AI results still require item-by-item manual review.
50%UnverifiedPocketLedger targets solo operators such as independent designers, developers, photographers, and consultants, positioned as an alternative between heavy accounting software and cloud receipt tools.
50%UnverifiedPocketLedger's core flow is: the user uploads a receipt image or pastes text, local AI extracts the merchant, date, amount, currency, category, and description to generate an editable draft, and the user reviews and confirms it before writing to the ledger.
50%UnverifiedPocketLedger stores data in a local SQLite database and requires no cloud account registration.
50%UnverifiedPocketLedger's backend uses the Python standard library and its frontend uses HTML, CSS, and vanilla JavaScript, introducing no complex frameworks.
50%UnverifiedPocketLedger calls the Qwen2.5-VL 3B model via Ollama for receipt understanding and constrains the output fields with a structured JSON schema.
50%UnverifiedPocketLedger's product design is that AI only generates drafts, and every expense requires human confirmation before being written to the ledger, making no irreversible bookkeeping decisions.
50%UnverifiedThe business model given by the PocketLedger author is a one-time-paid desktop deployment, supplemented by optional onboarding guidance and support services.
50%UnverifiedThe author candidly admits that PocketLedger has not yet completed a formal first-user trial.
50%UnverifiedPocketLedger is open-sourced on GitHub, and the code repository is SaikatDash/sideHustlers.
50%UnverifiedPocketLedger is a project submitted by developer SaikatDash in the second week of the Hacktoberfest open-source AI challenge.
50%