Field study 04 · Organizational memory
Let the workspace organize itself as the conversation grows
Topic-aware threads connect ongoing conversations to source documents and collaborators, giving people a navigable map of the work without asking them to maintain another taxonomy.
The situation
Shared memory is useful only if the next person can find the right part. A flat conversation list works for the last few days, then becomes a feed of document names, greetings, and unrelated questions. Manual folders help with artifacts but rarely capture the conversation that explains why an artifact changed.
Dana4 treats a topic as a lightweight graph node. Meaningful public threads are connected to one primary topic. The thread already points to its source document; document and authorship edges then connect the topic to artifacts and collaborators.
Topic → conversation → document → people
The product shows the useful result—a grouped thread list and document navigator—without requiring the user to operate a graph visualization.
How assignment works
- The classifier reads the latest meaningful task episode and source-document title.
- Greetings, mentions, URLs, UI words, and action scaffolding are removed.
- Existing workspace topics act as priors; a topic is reused only with enough specific overlap.
- The topic-to-channel edge is stored and exposed through the workspace graph API.
The algorithm is incremental classification, not batch clustering. Every new meaningful conversation is compared with topics the workspace already knows. That makes topic names more stable and reduces near-duplicates.
Threads such as “Hi,” “thanks,” or a time question remain under Other conversations. A specific document title can strengthen a short real request such as “summarize this,” but a generic document called “Workspace” cannot promote a trivial exchange into a topic.
The onboarding moment
Imagine a researcher joining a workspace six weeks into a project. Instead of reading every conversation chronologically, they open Threads → Grouped, choose a subject, and see:
- the conversations assigned to that subject;
- the documents those conversations belong to;
- how recently each document was active;
- the people associated with the connected work;
- a direct path into the source document and its discussion.
This is navigation, not an automatically generated answer. The person can inspect the evidence and continue from the artifact itself.
Good fits
| Workspace | Topics that emerge |
|---|---|
| Product discovery | Customer problems, competitive alternatives, experiments |
| Research program | Methods, evidence clusters, unresolved questions |
| Creative account | Positioning, visual direction, channel production |
| Launch room | Scope, technical readiness, narrative, customer readiness |
| Client delivery | Requirements, workstreams, approvals, handoff |
Privacy boundary
Only public workspace threads eligible for the shared thread list enter topic assignment. Private conversation titles, messages, summaries, counts, and assignments are not included. Topic and graph endpoints require workspace membership.
What to look for
- Related conversations reuse a stable topic rather than creating variants.
- Low-information chat stays out of the durable map.
- Every shown document has an explainable path through at least one assigned conversation.
- A teammate can move from topic to source document in one interaction.
- Private threads never influence shared topic metadata.
Try it
- Start substantive conversations from two related documents.
- Return to Threads after the replies complete.
- Open Grouped and choose the resulting topic card.
- Inspect the connected documents and collaborators.
- Use Recent when chronology matters more than subject.