2Ori Mnemos – an open-source persistent memory for AI agentsMost agents start every session from zero. Ori is an open-source, markdown-native memory layer that fixes this. local-first, no cloud, no vendor lock-in.- Three-signal retrieval: semantic embeddings + BM25 + PageRank fused via RRF- Graph-aware forgetting: ACT-R vitality model + spreading activation +Tarjan's articulation point protection (notes that matter stay alive, their neighbors stay warm, the rest fades) - Agent identity: name, goals, methodology persist across sessions and clients - ~850 tokens per query regardless of vault size - 14 MCP tools, Apache-2.0, 396 tests npm install -g ori-memory https://github.com/aayoawoyemi/Ori-Mnemos
Most agents start every session from zero. Ori is an open-source, markdown-native memory layer that fixes this. local-first, no cloud, no vendor lock-in.- Three-signal retrieval: semantic embeddings + BM25 + PageRank fused via RRF- Graph-aware forgetting: ACT-R vitality model + spreading activation +Tarjan's articulation point protection (notes that matter stay alive, their neighbors stay warm, the rest fades) - Agent identity: name, goals, methodology persist across sessions and clients - ~850 tokens per query regardless of vault size - 14 MCP tools, Apache-2.0, 396 tests npm install -g ori-memory https://github.com/aayoawoyemi/Ori-Mnemos
Most agents start every session from zero. Ori is an open-source, markdown-native memory layer that fixes this. local-first, no cloud, no vendor lock-in.
- Three-signal retrieval: semantic embeddings + BM25 + PageRank fused via RRF
- Graph-aware forgetting: ACT-R vitality model + spreading activation +
Tarjan's articulation point protection (notes that matter stay alive, their neighbors stay warm, the rest fades)
- 14 MCP tools, Apache-2.0, 396 tests