CombinatorPublic verdict
MemoryOS
Promising memory AI, but the founders must narrow the wedge and prove trust before this feels venture-backable.
PASS
64
/ 100
The idea is real, but right now the founders sound like they want the upside of a trust product without yet proving the wedge, the metrics, or the safety model.
Green flags
- Clear pain point: people forget decisions, commitments, and context across many tools
- Strong AI-native wedge if retrieval and citations are actually trustworthy
- Potentially sticky product if it becomes the default place to recover lost context
- Natural expansion path from personal memory into team knowledge workflows
- Source-backed recall is especially compelling for trust-sensitive or regulated use cases
Red flags
- Founders did not answer the wedge question with an actual wedge; 'yes that can be done' is not a plan
- No benchmark, error budget, or evaluation method was provided for retrieval quality
- Permissions and isolation were answered with 'everything remail in local,' which is vague and not credible
- Business model answer 'we will use users storage' does not address compute, transcript costs, or pricing power
- Scope is extremely broad across email, Slack, WhatsApp, meetings, browser history, screenshots, docs, and GitHub
- Privacy, security, and trust failures would be existential for this product