Roadmap / Self-growing memory

Self-Growing Memory Roadmap

A product blueprint for memory that can cite sources, reconcile drift, discover relationships, and grow from trusted connectors.

Memorose is moving from durable recall toward a governed memory runtime: repeated input is skipped, source context is preserved, conflicts become visible, and connector workflows can feed reviewed long-term knowledge.

Read the docs View GitHub
Capture streams connectors Reason hash guard conflicts graph jobs Review provenance approval Promote memory insight tasks opt-in active behavior + audit trail + budgets
Product path

From passive recall to governed memory growth

The roadmap keeps the current runtime shape intact while adding trust primitives first, then active insight workflows, and finally connector-driven research loops.

01

Hash Guard

Skip duplicate payloads and detect drift before embedding work.

02

Provenance

Attach source, connector, timestamp, and confidence to memory units.

03

Conflict Timeline

Show how preferences, claims, and task assumptions changed.

04

Graph Insights

Find cross-stream links that vector recall would miss.

05

Knowledge Gaps

Queue missing context for review or research workflows.

06

Connector Framework

Bring GitHub, docs, local files, chat, CRM, and web context in safely.

Milestones

Phased work, visible product outcomes

Each phase produces a usable surface instead of a hidden backend rewrite: better data quality, better review tools, and clearer integration points.

v0.2 Foundation

Self-Correcting Memory

  • Content hash dedupe and drift detection.
  • Provenance fields on memory units and graph edges.
  • Conflict detection with dashboard timeline views.
  • Review queue for semantic updates and derived memories.
v0.3 Insight layer

Self-Growing Memory

  • Proactive graph insights across streams and domains.
  • Knowledge gap candidates with confidence and source needs.
  • Connector framework boundaries, budgets, and plugin contracts.
  • Runtime policies for cost, rate limits, and active behavior.
v0.4 Expansion

Research-Aware Memory

  • Repository, documentation, local file, and browser connectors.
  • Optional deep research plugin that produces source-backed notes.
  • Promotion workflow from research jobs into durable memory.
  • Local privacy and encryption guidance for sensitive deployments.
Connectors

Integrations become first-class memory inputs

Connector work is part of the roadmap, but it should land behind explicit scopes, provenance, rate limits, and review rules so external context does not pollute long-term memory.

GitHub

Issues, pull requests, commits, code review decisions.

Docs

Product docs, API references, changelogs, release notes.

Local files

Private notes, markdown folders, project artifacts.

Chat

Slack, Discord, Feishu, and team decision streams.

CRM and support

Tickets, accounts, objections, escalations, outcomes.

Browser and web

Clips, citations, research trails, source snapshots.

Delivery surface

Where the work lands

The roadmap should be visible in the runtime, dashboard, APIs, SDKs, and optional plugins rather than living as a disconnected research track.

Implementation map

Core runtime Hash guard, provenance, conflict model, graph insight jobs.
Dashboard Timeline, review inbox, connector health, budget visibility.
API Connector ingest contracts, review actions, insight promotion endpoints.
SDK Typed connector clients and local development helpers.
Plugins Arbitrator, research, encryption, and domain-specific enrichers.