Skip to content

Instantly share code, notes, and snippets.

@arthyn

arthyn/MEMORY_COMPANION_ROADMAP.md

Created February 22, 2026 05:31
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save arthyn/dcf8369ee485b90ae82323830ae8909f to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save arthyn/dcf8369ee485b90ae82323830ae8909f to your computer and use it in GitHub Desktop.
Download ZIP
Memory Companion execution roadmap
Raw
MEMORY_COMPANION_ROADMAP.md

Memory Companion Roadmap

Last updated: 2026-02-22 Project: Memory Companion (Willow) Status: Draft → Execution

How to use this doc

  • Keep tasks small and testable.
  • Update Status and Notes as work progresses.
  • Run regression checks before merging behavior changes.

Status key:

  • TODO
  • IN_PROGRESS
  • BLOCKED
  • DONE

Phase 0 — Alignment & Scope (Target: 1 day)

ID Task Owner Est Status Notes
P0-1 Define pilot patient profile and caregiver profile Hunter 1h TODO Stage, routine complexity, distress triggers
P0-2 Lock MVP features for first pilot Hunter + Nimbus 1h TODO Wake word, Q&A, pending-memory queue, dashboard
P0-3 Define hard safety boundaries Hunter + Nimbus 1h TODO Never invent personal facts; deceased handling rules
P0-4 Write 2-week pilot success criteria Hunter 30m TODO e.g. <2 critical errors/week

Exit criteria:

  • MVP scope signed off
  • Safety boundaries documented

Phase 1 — Architecture Freeze (Target: 1–2 days)

ID Task Owner Est Status Notes
P1-1 Create single source-of-truth architecture doc Nimbus 2h TODO Components, ports, startup order
P1-2 Resolve naming drift (Willow vs placeholder wake words) Nimbus 30m TODO No mixed naming in docs
P1-3 Resolve path drift (~/clawd vs actual paths) Nimbus 1h TODO Make all docs runnable as-is
P1-4 Add one-command bringup script (make up or script) Nimbus 2h TODO Covers required services

Exit criteria:

  • New contributor can boot stack from docs in one pass

Phase 2 — Memory Graph Hardening (Target: 2–4 days)

ID Task Owner Est Status Notes
P2-1 Finalize schema for nodes/edges Nimbus 2h TODO people/places/events/routines/comforts
P2-2 Add provenance fields (source, confidence, approved_by) Nimbus 2h TODO Required for trust
P2-3 Enforce pending → approved write path Nimbus 3h TODO No direct unreviewed truth writes
P2-4 Build backup script + retention policy Nimbus 2h TODO Daily snapshots
P2-5 Build restore test script Nimbus 1h TODO Must prove recovery

Exit criteria:

  • Backup/restore passes
  • Unapproved memories cannot enter live retrieval

Phase 3 — Agent Safety & Prompt Contract (Target: 2–3 days)

ID Task Owner Est Status Notes
P3-1 Convert prompt rules into explicit decision policy Nimbus 2h TODO Retrieval-first logic
P3-2 Create response templates for uncertainty/deceased/distress Nimbus 2h TODO Keep short/calm
P3-3 Build regression test set (10–20 scenarios) Nimbus 3h TODO Include known edge cases
P3-4 Add pass/fail harness for prompt changes Nimbus 3h TODO Gate merges on safety checks

Exit criteria:

  • Safety regression suite passes before deploy

Phase 4 — Voice Loop Reliability (Target: 2–4 days)

ID Task Owner Est Status Notes
P4-1 Tune wake word sensitivity + false positives Nimbus 2h TODO Real home noise test
P4-2 Tune barge-in/interrupt behavior Nimbus 2h TODO Reduce overlap confusion
P4-3 Tune TTS voice for clarity/calm Hunter + Nimbus 1h TODO Stability/speed settings
P4-4 Add fallback behavior when voice path fails Nimbus 2h TODO Text fallback/caregiver alert

Exit criteria:

  • Stable voice round-trip in real environment

Phase 5 — Caregiver Dashboard MVP (Target: 3–5 days)

ID Task Owner Est Status Notes
P5-1 Build pending queue UI (approve/reject/edit) Nimbus 4h TODO Fast triage flow
P5-2 Build approved memory browser Nimbus 3h TODO Filter by type/source
P5-3 Add quick-edit comforts/routines panel Nimbus 3h TODO Distress-critical
P5-4 Add activity/audit log view Nimbus 2h TODO Who changed what

Exit criteria:

  • Caregiver can manage memory lifecycle without CLI

Phase 6 — Ops, Security, and Deployment (Target: 2–3 days)

ID Task Owner Est Status Notes
P6-1 Add health checks for all services Nimbus 2h TODO Agent, graph, dashboard
P6-2 Add startup preflight (.env validation) Nimbus 1h TODO Fail fast on config issues
P6-3 Add secrets handling guidelines Hunter + Nimbus 1h TODO No plaintext in repo
P6-4 Add monitoring basics (error logs + key metrics) Nimbus 2h TODO hit rate, fallback rate

Exit criteria:

  • System can be operated predictably and audited

Phase 7 — Pilot Runbook (Target: 1–2 weeks)

ID Task Owner Est Status Notes
P7-1 Create daily pilot checklist Hunter 1h TODO safety/accuracy UX checks
P7-2 Define incident severity + response playbook Hunter + Nimbus 2h TODO including pause/disable flow
P7-3 Run pilot and log issues Hunter Ongoing TODO structured daily notes
P7-4 Weekly retro and patch plan Hunter + Nimbus 1h/wk TODO prioritize safety first

Exit criteria:

  • Pilot outcomes documented
  • Go/no-go decision for wider use

Immediate next actions (this week)

  • Complete Phase 0 (scope + safety boundary signoff)
  • Complete P1-2 and P1-3 (naming/path cleanup)
  • Land P2-3 (pending→approved enforcement)
  • Draft first 10 regression scenarios (P3-3)

Risk register (living)

Risk Impact Likelihood Mitigation Status
Hallucinated personal fact High Medium Retrieval-first policy + tests OPEN
Caregiver overload in review queue Medium Medium Better triage UX + confidence ranking OPEN
Voice false positives Medium High Wake threshold tuning + environment test OPEN
Data loss/corruption High Low-Med Verified backups + restore drills OPEN

Definition of Done (project-level)

  • Safety suite consistently green
  • Pending-memory approval enforced technically
  • Backup/restore tested and documented
  • Caregiver dashboard usable for daily workflow
  • Pilot meets agreed success metrics
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment