Letta (formerly MemGPT) pioneered stateful, self-editing memory for an individual agent. Caura targets the layer above the single agent: governed memory shared across a fleet — multiple agents, teams, and vendors on one auditable plane, with an auto-extracted knowledge graph and contradiction handling built in.
| Capability | Caura | Letta |
|---|---|---|
| Multi-fleet support | ||
| Agent trust tiers + keystone policies | ||
| Cross-vendor memory sharing | ||
| Contradiction detection + supersession | ||
| Per-agent retrieval tuning | ||
| PII detection & quarantine | ||
| Audit trail / provenance | ||
| Knowledge graph (auto-extracted) | ||
| MCP-native | Partial | |
| OSS license | Apache 2.0 | Apache 2.0 |
Competitive comparisons reflect our reading of each product's publicly available documentation, to the best of our knowledge as of June 2026. Capabilities change — verify current details with each vendor.
Letta pioneered stateful, self-editing memory for an individual agent. Caura operates one layer up: governed memory shared across a fleet of agents, teams, and vendors, with an auto-extracted knowledge graph and contradiction handling built in.
Letta's memory lives inside a single agent's state. Caura is built for fleet-scoped shared memory, governed across many agents and tenants on one auditable plane.
Yes. Caura extracts entities and relations on every write and expands recall through graph hops. It also adds trust tiers, keystone policies, contradiction detection with supersession, and audit trails.
Yes. Caura and Letta are both Apache 2.0 licensed. Caura is self-hostable via Docker Compose or available as a managed service at caura.ai.
Comparison reflects our reading of each project’s public documentation as of June 2026. Mem0, Zep, and Letta are solid projects for single-agent memory. Spot something out of date? Open an issue or PR and we’ll correct it.