Persistent memory for AI agents
Memory that remembers why it believes what it believes.
Text goes in; anchored, recallable claims come out — each one traceable to the exact evidence it came from. Search across the entire donto substrate.
try:
How it works
Three calls, one durable memory
/memorize
Extract
POST /memorize text → the OpenCode/GLM agent emits free, multi-directional claims, anchored to the source span, and ingests them as a durable job.
/search
Search
POST /search ranks across all 39M+ statements with a bounded full-text query — sub-second even on hot stopwords.
/recall
Recall
POST /recall composes a holder-scoped bundle for an agent — the durable working memory behind a conversation.
Try it
Memorize something
Add a fact to the substrate. Extraction runs asynchronously; once it lands, search for it above.
Extraction runs as a durable job (OpenCode · GLM) — facts land in the substrate asynchronously, then become searchable above.