Productisation Track (Epic #1029)
Hybrid Memory already has strong memory depth. This track exists to make that depth easy to see, trust, and demo.
Goal
Make hybrid-memory feel like a first-class product: immediately understandable, inspectable, and demoable without weakening the underlying trust model.
Important: Epic #1029 is a coordinating issue, not one giant implementation branch. Product work lands through focused child issues.
Current phase view
| Phase | Goal | Status | Child issues |
|---|---|---|---|
| Phase 1 — Foundation | Viewer, top-of-repo legibility, session visibility | Partially shipped | #1023 ✅, #1024 ✅, #1025 open |
| Phase 2 — Polish | Messaging, demos, and simple public surface | Partially shipped | #1027 ✅, #1028 open |
| Phase 3 — Maturation | Explicit retrieval strategy and layered terminology | Planned | #1026 open |
What is already shipped
Product entry points
- Memory Viewer / Mission Control via the local dashboard and viewer routes documented in
extensions/memory-hybrid/README.md(Issue #1023, closed) - README / onboarding refresh with the capture → store → recall → inspect → control mental model, trust/privacy links, and persona-based start paths (Issue #1024, closed)
- Simple public API / export surface documented at
PUBLIC-API-SURFACE.md(/health,/search,/timeline,/stats,/export) (Issue #1027, closed)
Trust and operator surface already available
- Local-first storage and inspection paths
- Verification, provenance, and deletion controls
- Quick start, operations, backup/restore, and trust/privacy documentation
- Session distillation and narrative docs that explain how memory is captured and reused over time
What remains open
#1025 — Session timeline / observability
Still needed for the product story:
- one coherent session timeline
- capture vs injection visibility
- skipped/suppressed write explanations
- a human-readable “why this was recalled” surface
Status update (this implementation pass):
- unified observability service + tool surface is in place
- CLI now exposes
openclaw hybrid-mem audit sessionwith summary/timeline/json formats - remaining polish is primarily UI visualization depth
#1028 — Messaging, visuals, and demo story
Still needed for the presentation layer:
- tagline and elevator pitch
- hero screenshots / proof points
- 60-second and 5-minute demo scripts
- terminology cleanup for first-time readers
#1026 — Filter → rank → hydrate retrieval
Still needed for the explicit retrieval model:
- named constrained-search mode in code and docs
- structured filters before semantic ranking
- clearer explanation of why results matched and how they were ranked
Status update (this implementation pass):
- constrained-recall mode is documented and exposed as user-facing retrieval mode
- filter → rank → hydrate explanation is returned in tool output
Recommended execution order
- Finish #1025 so users can inspect capture, recall, injection, and skips without log-diving.
- Finish #1028 so the now-visible product has a sharper story, screenshots, and demo flows.
- Finish #1026 so retrieval becomes a named, teachable product capability instead of an internal implementation detail.
Guardrails for every phase
- Local-first remains the hero path — no hosted-memory dependency required for the baseline experience.
- Keep the rich tool API — the product surface should complement it, not replace it.
- Do not weaken verification, provenance, or decay just to make the UI or docs simpler.
- Prefer layered explanations — simple first, deep internals one click away.