Typed episodes
Store what happened, when it was observed and when it became available.
MADB gives agents durable episodic memory, current-truth recall and a causal record of why they acted. It runs locally, survives every session and keeps its uncertainty visible.
you Production uses api-v1.
agent Stored evt_41
you We migrated to api-v2.
agent evt_84 caused by evt_41
Episodes enter through the MADB encoding spine, form causal relationships, compete for bounded context and return with authority and uncertainty intact. Hover a node to inspect its functional 0.2.14 memory state. The visualization models structure—never private content.
Each path has a different job, success signal and commercial boundary.
Give Codex, Claude Code or another MCP client local memory it reaches for reflexively.
Add automatic recall and persistence to an Anthropic Python loop with one import change.
Embed the Rust engine inside an agent platform through a negotiated integration.
Brain-inspired engineering primitives without claiming biological equivalence.
Store what happened, when it was observed and when it became available.
Link decisions to the evidence, memory and outcomes that caused them.
Retrieve by subject, context, recency and causal proximity.
Replace stale authority without erasing the historical episode.
Release intentions only when time and context cues are satisfied.
Expose coverage, conflicts and abstention reasons before acting.
The 0.2.14 release matrix records hardware, corpus, thresholds and limitations. Results are measured-machine observations—not universal performance guarantees.
MADB records typed episodes, temporal availability and causal relationships, then assembles bounded context with explicit authority and uncertainty.