YantrikDB

YantrikDB is a cognitive memory database for AI agents. It is described as a way to model how memory works, rather than treating stored items as equal records for simple embedding and top-k retrieval. The page links it to temporal decay, semantic consolidation, contradiction detection, a knowledge graph, and proactive triggers.

Several capabilities are named explicitly. Relevance-conditioned scoring lets relevance gate other signals. A cognitive state graph uses typed nodes for beliefs, goals, intents, and preferences, with typed edges such as supports, contradicts, causes, and predicts. The system also includes autonomous cognition functions such as consolidation of related memories, conflict detection, and pattern mining, exposed through db.think(). Proactive triggers are described as acting on decaying memories, unresolved conflicts, and emerging patterns. Five unified indexes are listed: vector (HNSW), graph, temporal, decay heap, and key-value, all in one embedded SQLite database.

YantrikDB can be embedded in Rust or Python, run as an MCP server, or deployed as a Raft-replicated cluster. The page also says it can be used through a browser-based WebAssembly build, and the live demo shows record(), recall(), relate(), and think() in the browser. For MCP use, it is presented as persistent memory for Claude Code, Cursor, Windsurf, and other MCP-compatible AI agents. The server variant is described as providing an HTTP API, multi-tenant database operation, Raft replication, automatic failover, and encryption at rest. It also mentions a FastAPI dashboard for browsing, configuring, and safely maintaining Hermes agent memory, with read-only browsing by default, an opt-in admin mode, and an optional dashboard password.

Installation commands are shown for pip, cargo, Homebrew, and Docker. The bundled embedder is said to work out of the box, with a default potion-base-2M static embedder and optional potion-base-8M and potion-base-32M downloads. The page names a GitHub project and shows version references such as v0.8.17, but it does not state a pricing model or license.