Background interview mode
The interview agent can now listen during live work — a pricing review, an incident, a customer call — and stitch what it learns back into the memory graph.
- Background mode for the interview agent, with opt-in consent and pause controls.
- Speaker diarization for multi-party sessions on the call recordings adapter.
- Topic coverage planner is 38% faster on long sessions.
- Resolved a race condition where reflections could fire on incomplete turns.
Customer-managed keys (KMS)
Enterprise tenants can now encrypt the memory graph at rest with keys they control in AWS KMS, GCP KMS, or Azure Key Vault.
- Customer-managed keys with one-click revocation and a guided onboarding.
- BYO LLM routing for Anthropic, OpenAI, Bedrock, Azure OpenAI, and Vertex.
- Audit log retention extended up to seven years on Enterprise.
Drift detection
When a new transcript contradicts an existing SOP, Mnemos flags the drift and routes it to the canonical owner with the conflicting span highlighted.
- Drift detection for SOPs with span-level diffing.
- Freshness score uses observed activity from connected source systems, not just doc age.
- Improved entity merge precision when names collide across departments.
Time-travel in the graph explorer
Rewind the memory graph to a date. See what the company looked like before a reorg, an acquisition, or an incident.
- Time-travel for the graph explorer, scoped to the asker's permissions at the historical moment.
- Snapshot export for any subgraph to CSV, JSON, or a static report.
SCIM provisioning and custom roles
Enterprise IAM features now generally available. Custom roles let you express the exact least-privilege boundaries your security team needs.
- SCIM 2.0 provisioning for Okta, Azure AD, Google Workspace, and OneLogin.
- Custom roles, with a permissions matrix builder in admin.
- SSO setup wizard reduces enterprise onboarding by 45 minutes on average.
Audit lineage on every AI answer
Every AI answer now renders with a clickable audit ID linking the question, the retrieved set, the filter decisions, and the model used.
- Audit ledger UI for compliance teams.
- Audit ID linking and shareable, redacted audit views for external review.
- Permission filter latency reduced 22% on large tenants.
Onboarding Copilot
A personalized copilot built from the memory graph for every new hire. Day-one productivity with a 90-day journey baked in.
- Onboarding Copilot with role-aware reading lists and shadow-pair scheduling.
- Manager dashboards for ramp visibility and question intelligence.
Knowledge graph explorer
An interactive UI for navigating people, workflows, systems, and dependencies — with risk overlays and one-click pivot.
- Knowledge graph explorer with permission-aware traversal.
- Risk overlays — bus-factor, freshness, ownership gaps, access concentration.
- Graph rendering is now WebGL-accelerated for tenants over 50k entities.
Microsoft Teams app
First-class Mnemos experience in Microsoft Teams — search, chat, and capture without leaving the app.
- Microsoft Teams native app, including search command and capture bot.
- Slack message extraction now preserves thread context for better retrieval.
ACL inheritance from source
Mnemos now respects ACLs from Google Drive, Notion, SharePoint, and Confluence at retrieval time. Permission-aware AI, by design.
- ACL inheritance for the four major doc systems, refreshed on webhook.
- Per-question filter ledger surfaced in the admin audit view.
- Resolved a Confluence space-permission edge case with restricted parent pages.
EU data residency
Enterprise tenants can now pin to EU-West. All data — graph, transcripts, embeddings — remains in region.
- EU-West region for Enterprise tenants.
- Cross-region replication block for compliance-strict deployments.
- Lower latency for European customers on Slack and Notion sync.
AI Search GA
Hybrid retrieval, permission-aware filtering, and cited answers — now generally available on every plan.
- AI Search GA with citation enforcement on every claim.
- Slack and Teams native search commands.
- Re-ranker model upgraded — average answer relevance up 14% on the internal eval set.