Call with Tejas Channappa — 2026-06-07

Transcribed 2026-06-07 · ~15 min substantive content · participants: Jacob Cole + Tejas Channappa

📄 Full transcript: tejas-2026-06-07-transcript

TL;DR: Converging on a shared knowledge tool: meetings/conversations as raw streams → agents extract entities-with-provenance → those flow into living "knowledge products." Tejas frames the meeting tool as "augmenting short-term memory" inside an all-in-one Slack-like hub; advocates TiddlyWiki's malleable-software philosophy (agents erase the learn-Tiddlers friction). Jacob counters with Murmur — a shipped, minimalist thought-stream with per-note "send to agent." Decision: build the tool for ourselves first to stop spinning wheels.

💡 Ideas

• Streams → entities → knowledge products (Tejas's core architecture): raw transcripts/messages are streams; agents extract typed entities (insight / question / decision / feedback) with provenance; these atoms feed live, dynamic "knowledge products" (e.g. a requirements doc) that anchor where agents file new info.
• "Augmenting short-term memory" — the framing for the meeting tool (flagged #highlight). Tracks the threads you're running and helps you move through them.
• All-in-one Slack-like hub — meetings + conversations that automatically maintain your knowledge base (the "Greg Eisenberg Slack tool" reference).
• HUD feature — surfaces relevant atoms from your knowledge base mid-conversation ("this might be relevant right now").
• "Lovable for wikis" — Claude's HTML strength makes generated knowledge presentation richer than BI dashboards; build the tool while talking.
• TiddlyWiki / malleable software (Tejas): no separation between writing notes and changing the tool — agents now write the Tiddler language for you, removing the learning friction.
• Murmur (Jacob, shipped): minimalist thought-stream, voice/text note at top, optional "send to agent" per note → starts an agent chat that can bridge to Mac mini / cloud box / Hermes.

✅ Follow-ups (action items)

• Tejas: fix & re-share the broken demo (one of 3–4 prototypes).
• Jacob: Murmur is on TestFlight — test after tomorrow (mid-push). Get Tejas access.
• Jacob: share his "whole different take" on notes-vs-tool (deferred in the call).
• Both: decide on the grounding move — build the tool for ourselves first and dogfood.

🧵 Open threads

• Jacob's counter-take on "taking notes = changing the tool" — promised, never delivered on the call.
• Mobile story for TiddlyWiki — raised, left unresolved.
• Murmur ↔ knowledge-product architecture — one system or two? Murmur is lightweight capture; Tejas's entity/provenance engine is the heavy backend. Unmerged.
• Which substrate — TiddlyWiki vs. plain HTML-by-agent vs. Murmur's stream — undecided.

❓ Questions we should be asking

On architecture

1. What's the entity schema? (insight/question/decision/feedback — full set? who defines new types?)
2. How does provenance actually attach — per-utterance, per-thread, per-speaker — and is it queryable?
3. Are "knowledge products" user-declared up front or inferred by agents? Both?
4. Is Murmur the front-end capture layer for Tejas's backend, or a competing whole?

On the build
5. What's the smallest dogfoodable slice that grounds you this week?
6. Substrate decision: commit to TiddlyWiki, or stay on agent-generated HTML?
7. Mobile: required for v1 or desktop-first?

On scope/positioning
8. All-in-one Slack hub, or a focused meeting-memory tool? (Scope creep risk.)
9. What does the HUD trigger on, and how do you avoid noise?
10. Where does this sit relative to Noos (Jacob's existing entity-with-provenance graph)? Reinventing it, or should this be a Noos client?