dsh-noteboard

Long conversations with a model tend to bury the truly valuable decisions, conclusions, and sparks of insight inside an ever-growing context window. dsh-noteboard turns a DeepSeek Harness session into an editable knowledge canvas: select any conversation text and save it as a raw Markdown note, or let AI distill it into a concise summary. Each note is a frontmatter-enriched .md file paired with a JSON Canvas layout file, so content and placement are fully decoupled and can be managed through plain-text editors and versioned in Git.

On top of the canvas sits a complete AI toolchain. Twelve registered model tools cover note creation, querying, modification, tag-based re-arrangement, and history restore. Five built-in workflows let the model execute multi-step tasks such as conditional extraction, topic-based tagging and colouring, side-by-side comparison of proposals, action-item decomposition, and deduplication with synthesis. Users can also send notes back into the input box as structured reference bubbles that travel with the message to the model for further reasoning, with an automatic guard at 32,000 cumulative characters to prevent context overflow.

It is designed for individuals and teams who work intensively with models inside Harness to brainstorm, review, and organise knowledge: pull key points into notes while writing, let AI compare trade-offs across notes, and auto-rearrange the board by tag. Everything is files and JSON with no separate database, so Git gives the team full traceability of every change.