Inspiration

A promise often arrives with a condition attached. Someone agrees to send a draft after another person shares the figures, then a later conversation moves the date. When those exchanges live in separate transcripts, a reminder can lose the reason the plan changed. THREADLINE grew from the wish to keep the words, the condition, and the later update together so a person can return to a commitment with its history intact.

What it does

THREADLINE reviews commitments beside their exact source messages, including who is responsible, what must happen first, and how a later conversation changed the plan. Its local Simulated Alexa+ assistant can answer a supported question using the current commitments, source messages, and dependency tools. In the synthetic mural example, it explains the source-stated Friday-to-Monday change and cites the messages that support it. A person can open the evidence, acknowledge the proposed review, and approve it into the same persistent commitment history.

How we built it

Node.js serves a dependency-free browser workbench with source-linked persistence and a read-only Bee adapter. The new bounded English interpreter plans a small sequence of local engine queries and composes the actual returned evidence. It preserves the current commitment and source identities through explicit human review, then recomputes its answer from the saved state. Source refresh, locking, and unsupported requests retire the assistant context instead of carrying an old answer forward. The demonstrated conversations are clearly labeled synthetic material.

Challenges

A source can change while someone is considering a review. The assistant therefore needs the exact current message and commitment identities, and a delayed answer must not authorize a stale action. The original promise and its prerequisite also need to remain visible when later evidence questions the plan. Real Bee data and live Alexa service use remain unavailable in this demonstration.

Accomplishments

The actual browser course answered what changed about the mural plan with exact m01, m02, and m04 citations, two reviewed nodes, and the stored prerequisite. Explicit acknowledgment and approval appended the existing contested evidence and updated the same map and history. Reload preserved the review and dependency. Unsupported requests offered no tool action, and source refresh and lock retired the old assistant words. The suite passed 319 native tests, and the launched smoke check passed.

What we learned

The strongest assistant answer is one a person can trace back to the conversation and revise through the ordinary review flow. Sharing the workbench’s state and keeping review separate from interpretation makes a later change easier to understand.

What's next

We want to extend the supported questions, receive authorized real conversation material, and test an actual assistant client. The current experience is a deterministic local text simulation. It does not establish live Alexa service use, eligible Bee receiving, external calendar synchronization, or independent delivery of reminders.

Built With

Share this project:

Updates

Submission history