Amulet

Amulet Labs describes conflict-safe publishing as a compare-and-swap operation on a workspace head. Each agent publishes only if its starting revision remains current; a stale writer receives an explicit conflict and retains its session. The approach includes per-commit provenance and isolated session forks for parallel agent fleets.

Jul 2026 · Amulet LabsNews article

Amulet Labs explains how copy-on-write branching creates writable views of existing data without copying the underlying bytes. The model uses shared immutable blocks and isolated branches, allocating storage only when a branch changes data. The article connects this approach to AI training and evaluation workflows, including dataset snapshots, isolated experiments, and rollback to prior revisions.

Jul 2026 · Amulet LabsNews article

Amulet Labs compares ways to preserve AI-agent state across disposable compute, including paused sandboxes, platform volumes, and mounted durable workspaces. It examines portability, concurrent workers, lifecycle coupling, and versioned state, and presents Amulet’s workspace as a durable filesystem mounted into isolated sandbox sessions, with published results represented as attributable commits.

Jul 2026 · Amulet LabsNews article

Amulet Labs compares s3fs, goofys, Mountpoint for S3, JuiceFS, EFS, Archil, and Amulet for workloads that need filesystem behavior over durable data. The discussion focuses on object-store round trips, rename behavior, concurrent writes, and operational tradeoffs; Amulet describes its positioning for parallel AI workloads as including isolated forks, versioned history, conflict-safe publishing, and usage-based pricing.

Jul 2026 · Amulet LabsNews article