Designing systems that remember, recover & govern.
Great systems are not defined by model intelligence alone. They are defined by the architecture that surrounds it: the memory that persists, the pathways that recover, the orchestration that stays deterministic under load.
Intelligence is abundant and improving on a schedule no engineer controls. What remains scarce is structure: the discipline to make a capable model behave predictably, remember what it learned, and act within governed bounds.
Memory is architecture.
Interlude — on persistenceHow should a system remember across sessions without accumulating noise, drift, or contradiction?
PersistenceWhen execution is interrupted, how does a system recover its exact position and continue as if uninterrupted?
RecoveryHow do many agents, tools, and policies coordinate so the outcome stays deterministic and governable?
OrchestrationMemex: a memory that keeps its shape.
A temporal codebase graph that carries engineering context across coding-agent sessions.
Tree-sitter, Graphiti, and Neo4j connect code relationships and decisions. A read-only integration adds bounded context with timeout and fail-open handling; v0.9 was evaluated on 8 paired engineering tasks.
Cairn: pathways that survive interruption.
A v1.0.0 Python library for agent recovery after crashes and context loss.
Durable checkpoints, workspace restoration, and external-effect reconciliation support recovery. Evaluated across Claude and Codex, with zero duplicate effects observed in the recovery runs.
Nexus: determinism under coordination.
The earlier orchestration system, preserved in the engine room of Pinocchio.
Nexus explored policy-gated execution, approval workflows, and audit trails. Its repository is now Pinocchio: a Claude Code plugin under construction for checking agent completion claims against independent evidence.
Systems remember so people don't have to.
Interlude — on trustOne substrate,
three concerns.
Read left to righttop to bottom: a request enters, the orchestrator plans against persistent memory, executes through recoverable steps, passes every action through policy, and returns a result that can be replayed exactly.
The model will keep getting better on its own. The parts that decide whether you can trust it (memory, recovery, governance) are the parts we still build by hand.
From the editor's note, Vol. 01