I Built an AI That Remembers the Promises We Forget

Description and use

PromiseKeeper uses a local Qwen model to extract and manage promises from notes and chats, enabling researchers to quickly locate and confirm commitments.

Problem

Researchers often forget informal agreements made in meetings or chats, making it hard to track commitments and deadlines.

Outcome

PromiseKeeper successfully extracts three actionable promises from five synthetic note snippets, surfaces them with source quotes, and allows a reviewer to approve one overdue report and draft follow‑ups.

Workflow

• Import notes or text
• Local Ollama/Qwen extracts candidate promises
• Model proposes fields (action, owner, recipient, deadline, condition, certainty, evidence quote, next step)
• Pydantic validates schema
• API verifies evidence quote is a literal substring of source
• Human reviewer approves, edits, or dismisses each promise
• Approved promise becomes a draft follow‑up
• User reviews and sends draft (no sending integration)

Evidence

Import meeting notes, a chat transcript, or a text file. A local language model proposes commitments.

Source

The reviewer can approve it, correct it, dismiss it, or mark it complete.

Source

The app is built around the difference between a possible promise and a confirmed one.

Source

The recording follows a researcher importing five snippets of synthetic notes. Actual local Qwen inference finds three candidates: a revised field report, cleaned survey figures, and a conditional appendix.

Source