Memorry roadmap

Roadmap

Dates are targets, not commitments.

Month 4 after kickoff target

MVP

Prove that a memory created in one agent is usefully recalled in another within a day of setup.

  • remember, recall, forget, and context_pack MCP tools, remote MCP and REST API.
  • Consented browser extension for chatgpt.com, claude.ai, and grok.com with context-pack injection.
  • Typed extraction, deduplication, hybrid semantic and keyword recall, and personal, team, and workspace scopes.
  • Console timeline, search, edit, pin, and forget, plus connected-agent visibility.
  • Owner, admin, and member roles; encryption in transit and at rest; workspace isolation; immutable read/write audit; deletion on request; JSON/Markdown export.
  • Free and Pro self-service billing and paid Team plans from month 6.

Month 8 target

v1

  • Slack, Gmail, Notion, Google Drive, and meeting-note import.
  • Nightly consolidation, versioned conflict resolution, decay, and confidence scores.
  • Python and TypeScript SDKs and OpenAI Agents SDK, Claude Agent SDK, and LangGraph adapters.
  • ChatGPT App listing after developer-mode submission and planned Claude Connectors Directory submission.
  • Team spaces, shared-memory approval queue, sharing rules, and Team seat management.
  • SAML SSO, SCIM, retention policies, PII redaction, and per-tenant encryption keys.

Month 12 target

v2

  • CRM and ticketing connectors, webhooks, and bulk context-document import.
  • Memory-quality analytics, per-agent recall tuning, and weekly change digest.
  • Agent-to-agent handoff packets, offline export bundles, and Gemini/Copilot paths.
  • Admin analytics, policy templates, and per-agent usage reports.
  • EU/US data residency, VPC, DLP integrations, and a planned SOC 2 Type II report.
  • Enterprise contracts and usage-based API billing.

Memory lifecycle

How each memory changes over time

MVP target

Capture with consent

Capture submits content only with user consent.

MVP target

Extract typed candidates

Extraction turns submitted text into typed candidates and keeps the original episode as a pointer.

MVP target

Check and link

Checking detects duplicates, updates, and contradictions, and links people, companies, or projects. PII rules are in verification; detection and redaction are v1 targets.

MVP target

Store with provenance

Storage preserves source, author, time, scope, and history, and audits every write.

v1 target

Consolidate nightly

Nightly consolidation merges duplicates, lets newer dated evidence supersede older versions, and sends unresolved conflicts to review.

v1 target

Decay by workspace policy

Workspace-policy decay archives unused low-confidence memories after the configured period; pinned memories do not decay.

Planned; target not set

Learn from use

People and agents can rate recalls to adjust confidence.

MVP target

Recall within scope

Recall ranks relevance, recency, confidence, and scope within the caller’s token budget.

Control status

Data use, access, and assurance

  • Model trainingCurrent policy: Customer memories are not used to train models.
  • Memory typeProduct rule: Memories are reference data rather than instructions.
  • Unconfirmed writesMVP target: Unconfirmed webpage, email, and file writes stay quarantined until confirmed.
  • Browser extensionMVP target: It reads only sites you enable and your own conversations, with a visible indicator.
  • Encryption and workspace isolationMVP target: Encryption in transit and at rest, with workspace isolation.
  • Audit and deletionMVP target: Immutable read/write audit and deletion on request.
  • PII redactionv1 target: PII redaction is a v1 target.
  • DPAEarly access: A DPA is available at early access.
  • Data residency and VPCv2 target: EU and US data residency options and VPC are v2 targets.
  • SOC 2Planned for 2027: The audit is planned for 2027; a SOC 2 Type II report is a v2 target.