Adaptive Recall

Adaptive Recall is a hosted memory service for AI applications that goes well beyond the standard store-embeddings-and-search-by-cosine-similarity pattern. Built by AI Apps API and marketed as patent pending, it runs four retrieval strategies in parallel — vector similarity, temporal recency, full-text keyword, and knowledge graph traversal — and learns over time which strategies to prioritize for each type of query. Results are ranked with ACT-R activation modeling from cognitive science, so recency, access frequency, entity connections, and validated confidence all influence which memories surface first. Entities and relationships are extracted automatically into a knowledge graph, and memories have a real lifecycle: they progress through stages, gain or lose confidence as corroborating evidence arrives, and fade naturally when no longer accessed. An ML pipeline trains on your usage patterns, validates parameter changes against real query history, and monitors its own retrieval quality. For builders, the surface is deliberately simple: eight tools (store, recall, update, forget, graph, status, snapshot, feedback) exposed both as an MCP server for Claude Code and other MCP-capable clients and as a plain HTTP REST API with bearer-token authentication, so assistants and agents can gain persistent, self-improving memory with minimal integration work. Pricing starts with a free tier of 500 memories and no credit card, with paid plans from $19.99 per month scaling storage, request rates, consolidation frequency, and dashboard tools like the Memory Explorer, Graph Explorer, and Data Browser.