Start here
Deciding whether to adopt it
The decision-maker brief →
The problem, what it measurably does, what it does not, and what saying yes commits you to. Fifteen minutes, no specialist vocabulary.
Understanding the mechanism
How it works, walked through →
One memory followed from write to recall to correction, each piece introduced by what it does before it is named. Thirty minutes; every number sourced.
Visual explainers — for engineers & researchers
For builders & researchers
Agent Playbook & Findings →
Why AWM exists (the context-window wall), the PRIME→ACT→VERIFY→LEARN harness, the full agent feature surface, how multi-hop is solved, and the honest benchmark results.
How it works inside
Pipeline Walkthrough →
A visual tour of the recall pipeline — candidate generation, composite scoring, the wide rerank pool, the cross-encoder, ACT-R decay, and the abstention gate.
Reference documentation
- Quickstart Install, setup, first session
- Architecture Pipelines, schema, storage backends, roadmap
- API & config reference HTTP routes, env vars, recall tuning, schema
- AWM-Native Harness pattern Build an agent on the substrate
- Using AWM at scale Large corpora, multi-hop, write-for-recall
- SQLite ↔ PGlite parity Backend capabilities & selection
- Changelog Release history & design notes
- README Overview, benchmarks, why it's different