Our products

Make agents do less, remember better, and prove what changed.

A durable coordination layer for AI work. ldgr keeps an agent's tasks, runs, evidence, and decisions in a single SQLite ledger that survives restarts and context resets. A guardrail against drift, false completion, and context collapse.

Built in Rust as a single binary, ldgr wraps an agent's work in a structured loop (work, run, observe, decide) backed by a local SQLite ledger. It adds dependency-aware work queues, prompt provenance, a local web cockpit for human oversight, and a signed adapter ecosystem for extending into code, research, and security workflows.

Event-first memory for AI systems.

An experimental memory layer that stores streams of events as durable temporal evidence and retrieves them by structural cue, instead of collapsing them into static embeddings. Early research shows it preserves reusable temporal patterns more robustly under noise.

Ecphory explores an "event-first" approach to machine memory. Rather than flattening observations into static graphs or vectors, it preserves the raw stream of events as durable evidence and retrieves it by structural signature, echoing the psychological term for cue-triggered recall. It's active research, with early results on when this preserves reusable temporal structure better than exact recovery.