# MindTwin

> MindTwin is a governed Enterprise Memory layer for people and AI agents.
> It tracks what changed, what a team currently believes, and what remains
> unresolved. Customer Success is the first application: connect Slack,
> Gmail, and Gong to surface renewal risks and check outgoing drafts.

Key facts and limits:
- Current organizational state means beliefs, commitments, changes,
  unresolved tensions, and their evidence. It depends on connected sources
  and data quality; it is not complete or infallible organizational truth.
- MindTwin compares new statements with prior beliefs, flags changes
  proactively with dated evidence, and checks drafts against the account
  record. Superseded beliefs fade while history remains available. Access
  is through dashboard, Slack, REST API, and MCP. Search/RAG capabilities
  vary; MindTwin works alongside existing tools.
- Documented controls include per-team Postgres row-level security,
  cross-tenant tests, PII redaction before memory storage, crisis-content
  suppression with referral, and content-free suppression audit trails.
  Review hosting, model providers, access, retention, and procurement
  requirements during scoping; these controls do not imply certification.
- The published LoCoMo evaluation reports zero false crisis suppressions
  across 5,882 conversation turns. This measures false suppression on that
  test set, not complete detection, zero leakage, or enterprise compliance.
- MindTwin authored TWIST, a proposed benchmark for intervention quality.
  TWIST-v1.0 validates draft vetting: 161 human-validated items, retained-set
  kappa 0.851, and 13 configurations. No tested configuration meets all
  reported performance goals, including MindTwin. Proactive detection,
  belief updates, and safe recall are additional tracks in the design.
  Use the canonical repository and paper for current results; v0 is historical.
- Pilots last six weeks, cover one customer segment, and have a fixed fee
  and one agreed success metric. Review usefulness, false alarms, and
  misses against a baseline. Findings remain with the customer. A risk
  flag is not a saved renewal; revenue examples are illustrative, and ROI
  depends on measured outcomes and total cost.

Founder: Subrat Panda. Contact: subrat@mindtwin.me. Patent pending:
application filed in India.

## Product and company
- [Homepage](https://mindtwin.me/): category, product, workflow, and evidence.
- [Enterprise Memory](https://mindtwin.me/#memory): current state versus retrieval.
- [Pilot](https://mindtwin.me/#cta): scope, measurement, and economics.
- [FAQ](https://mindtwin.me/#faq): integrations, security, and evaluation limits.
- [Customers](https://mindtwin.me/customers): customer evidence as it becomes available.
- [Team](https://mindtwin.me/team): founder and careers.
- [Concept deck](https://mindtwin.me/deck): request the broader product story.

## Research sources
- [TWIST repository](https://github.com/subratpanda/twist-benchmark): canonical methods, data, annotations, results, and errata.
- [TWIST paper](https://arxiv.org/abs/2609.28575): proposed suite and validated draft-vetting track.
- [Local benchmark guide](https://mindtwin.me/twist/README.md): current status and historical reference files.

## Optional
- [Insights](https://mindtwin.me/insights): research and dated essays.
- [TWIST release](https://mindtwin.me/blog/twist-benchmark): development history and current update.
- [LoCoMo journey](https://mindtwin.me/blog/locomo-journey): retrieval engineering lessons.
- [AI amnesia](https://mindtwin.me/blog/ai-amnesia): forgetting and memory architecture.
- [Forbes response](https://mindtwin.me/blog/forbes-organizational-memory): organizational memory.
- [Memory economics](https://mindtwin.me/blog/twenty-trillion-memory-problem): infrastructure thesis.
- [Storage versus cognition](https://mindtwin.me/blog/databricks-storage-vs-cognition): architectural perspective.
- [Kapex response](https://mindtwin.me/blog/kapex-fourth-dimension): metacognition perspective.
- [Brand film](https://youtu.be/fuVOfozM9jM): short product overview.
- [Technical deep dive](https://youtu.be/mJnuv5kwrbM): memory engine overview.