A reasoning agent that remembers why, not just what.
Most memory systems for AI agents store facts as isolated snippets. Tragent stores facts as a graph, so when it recalls something, it also recalls how that thing connects to everything else it has learned, and can reason across those connections instead of just retrieving the nearest match.
Two memory systems, one decision layer.
A vector store is good at finding things that sound similar. A knowledge graph is good at explaining how things relate. Tragent runs both side by side: Qdrant handles semantic recall, Memgraph holds the structured relationships between entities, and TimescaleDB keeps a time-ordered record of what the agent has observed, so "what changed since last week" is a real query, not a guess.
The result is an agent that can answer not just "have I seen this before," but "what else does this connect to, and has that changed."
From a raw question to a grounded answer.
1. Recall
The query is embedded and matched against Qdrant for anything semantically close to what's being asked.
2. Traverse
Every match is looked up in Memgraph to pull in the entities and relationships actually connected to it.
3. Reconcile
TimescaleDB is checked for anything that's changed since the match was last confirmed. Stale facts get flagged, not trusted.
4. Respond
Only after recall, traversal, and reconciliation agree does the reasoning core commit to an answer.