Trove
Clean-room institutional knowledge for AI agents.
Trove turns sensitive team communication into governed, sanitized, agent-readable knowledge without exposing the private emails, calls, CRM notes, or chat threads that produced it.
Context broker
Policy checked answer assembly
Example request
How should I respond to this rate-limit concern?
Private context
Team pattern
Canonical docs
AI can read the docs. It still misses the field truth.
The most useful operating knowledge is scattered through private conversations: buyer objections, implementation surprises, customer-specific workarounds, and product nuance that never makes it into official documentation.
Not enterprise search
Search respects existing permissions, but it does not create a reusable shared knowledge asset.
Not basic redaction
DLP can detect sensitive data. Trove turns safe patterns into structured, reviewed knowledge.
Built for the people closest to the work.
Trove starts with Sales Engineering because those teams sit where product truth, buyer friction, competitive nuance, and integration reality all collide.
Rep question
How should I respond to Acme's rate-limit concern?
The rep's own account history, open opportunity notes, and recent email thread stay permission-bound to that rep.
Similar enterprise objections are generalized into safe guidance with source diversity and reviewer status attached.
Tools such as search_team_knowledge, find_similar_team_patterns, and trace_sanitized_knowledge expose only policy-approved results.
How the clean room works.
Trove separates raw source material from reusable knowledge at rest, during processing, and at retrieval time. Only approved, anonymized artifacts cross into the team corpus.
Ingest private sources
Email, CRM, chat, docs, tickets, transcripts, and approved product sources land in quarantine.
Detect sensitive data
Customer names, personal data, secrets, deal terms, and re-identification clues are found before publication.
Extract durable knowledge
The system keeps useful claims, patterns, implementation caveats, objections, and best practices.
Gate by policy and review
Policy version, source diversity, k-anonymity, and reviewer state decide what is eligible to share.
Publish approved corpus
Agents receive structured team knowledge cards, not the private source material underneath them.
Evidence attached to every shared item
- Sanitization result and sensitivity level
- Source count and source diversity
- Policy version and reviewer status
- Freshness, confidence, and approved citations
- Tenant-scoped lineage fingerprints for dedupe
What never crosses the boundary
Customer names, deal terms, personal data, secrets, raw transcripts, private messages, and account-specific identifiers stay out of the shared team layer.
Designed for security teams before launch.
Trove is built as split-context, policy-enforced infrastructure so CISOs and platform teams can understand exactly where sensitive data lives, who can access it, and what agents are allowed to retrieve.
Split control plane and data plane
Trove SaaS can manage tenant configuration, policy, billing, connectors, and upgrades while sensitive processing runs in a dedicated data plane.
BYO Azure NetApp Files option
Regulated customers can keep quarantine, raw private artifacts, sanitized artifacts, and audit evidence on customer-owned ANF volumes.
Token-scoped MCP access
Every tool call validates identity, audience, scopes, requested corpus, sensitivity level, and policy before returning context.
Audit-ready operations
SSO/SCIM, retention controls, audit exports, review queues, lineage, and deletion workflows are designed into the launch path.
Pre-launch roadmap.
- Foundation Tenant model, policy engine, audit events, connector framework, and local proof pipeline.
- Sales Engineering MVP Email ingest, private vault, clean-room extraction, review queue, team corpus, and MCP server.
- Enterprise hardening SSO/SCIM, retention, BYO data plane, private endpoints, Key Vault, and audit exports.
- Ecosystem expansion Slack, Teams, CRM, transcript, docs, SDK, API, and broader agent platform integrations.