Independent deployments will determine whether Dropstone SDK provides reliable persistence, recovery, and continuity for long-running agents beyond disposable session-based runtimes.
state: expiredheat: lowuncertainty: highknownscott: lowagent-harnesses long-running-agents agent-memoryDropstone
What is this?
Dropstone announced SDK 1.0 as a persistence layer for agents, centered on “one persistent memory per account” shared across CLI, chat, and VS Code. The supplied material frames this as infrastructure for agents that preserve state, recover from failures, and continue work across sessions—capabilities the broader snippets identify as central to long-running-agent engineering. However, none of the search results independently examines Dropstone or reports a real deployment, so its reliability, recovery behavior, and continuity claims remain unverified here.
Why it matters to Scott
Known via ip:framework.long-running-agents and dev:concept.resumable-agent-job-control-plane: Scott already treats durable external state, restartability, and recovery evidence as the basis of multi-session agent continuity, while the radar tracks near-identical validation questions for AgentHarness and Hermes Missions. Dropstone is a directly relevant implementation candidate, but without an independent deployment or benchmark it adds only another unverified product claim and would not yet change what Scott builds or argues.
ip:framework.long-running-agentsip:concept.checkpoint-disciplinedev:concept.resumable-agent-job-control-planeip:concept.model-plus-harness-benchmark-unitradar:concept.long-running-agentsradar:concept.agent-memoryradar:pi-agentharness-durable-runtimeradar:hermes-missions-durable-agent-execution
queries asked of Scott's wikis
- durable state and checkpointing for coding agents
- agent memory versus workflow state
- cross-session continuity in agent harnesses
- recovery and rehydration after sandbox failure
- shared memory across agent interfaces
- benchmarks for long-running agent reliability
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (3) — ⭐ canonical anchor
Interpretation history
2026-08-21T04:32:35Z
No independent deployment, recovery test, or architectural evidence appeared within the observation window; Relay remains only adjacent evidence. The episode has faded and should be reopened only if real-world Dropstone validation emerges.
2026-08-19T04:29:28Z
Relay adds adjacent evidence that developers want cross-device agent continuity, but its live-session remote-control model neither validates Dropstone nor demonstrates checkpointing, restart recovery, or durable rehydration. The case remains an unverified vendor implementation awaiting an independent deployment or failure-recovery test.
2026-08-19T04:22:39Z
evidence attached: hn.story.49356310 — Relay is a concrete open-source implementation of persistent, remotely controlled coding-agent sessions across devices, materially contextualising the durable-runtime hypothesis.
2026-08-17T12:45:51Z
No independent deployment, recovery test, or architectural evidence has appeared; the unchanged observation adds nothing beyond the vendor’s release claims. The implementation remains worth passively tracking, but the case has cooled pending real-world validation.
2026-08-17T12:41:36Z
grounded: known/low — Known via ip:framework.long-running-agents and dev:concept.resumable-agent-job-control-plane: Scott already treats durable external state, restartability, and r
2026-08-17T12:39:20Z
origin walked (codex/luna, conf 0.99): anchor hn.story.49329451 -> echo.blog.25088a7012 by Blankline Research
2026-08-17T12:38:38Z
case created — The released SDK is a concrete persistent-agent runtime artifact directly relevant to long-running orchestration and memory.
Decision trace
- 08-21 14:32expireNo independent deployment, recovery test, or architectural evidence appeared within the observation window; Relay remains only adjacent evidence. The episode has faded and should be reopened only if r
- 08-21 14:32alert_silentThe only delta is elapsed time with effectively unchanged engagement and no new technical evidence, so there is nothing consequential to surface.
- 08-21 14:32alert_routeThe only delta is elapsed time with effectively unchanged engagement and no new technical evidence, so there is nothing consequential to surface.
- 08-19 14:29repriceRelay adds adjacent evidence that developers want cross-device agent continuity, but its live-session remote-control model neither validates Dropstone nor demonstrates checkpointing, restart recovery,
- 08-19 14:29alert_silentThe new evidence concerns a separate implementation and does not establish Dropstone’s reliability or change Scott’s durable-runtime decisions; it can wait for independent deployment or recovery evide
- 08-19 14:29alert_routeThe new evidence concerns a separate implementation and does not establish Dropstone’s reliability or change Scott’s durable-runtime decisions; it can wait for independent deployment or recovery evide
- 08-19 14:23alert_silentRelay is a separate remote-control interface whose claimed continuity appears to come from keeping sessions alive on one backend machine. It provides no independent evidence for Dropstone and no demon
- 08-19 14:23alert_routeRelay is a separate remote-control interface whose claimed continuity appears to come from keeping sessions alive on one backend machine. It provides no independent evidence for Dropstone and no demon
- 08-19 14:22attachRelay is a concrete open-source implementation of persistent, remotely controlled coding-agent sessions across devices, materially contextualising the durable-runtime hypothesis.
- 08-19 14:22propose_attachRelay is a concrete open-source implementation of persistent, remotely controlled coding-agent sessions across devices, materially contextualising the durable-runtime hypothesis.
- 08-17 22:45repriceNo independent deployment, recovery test, or architectural evidence has appeared; the unchanged observation adds nothing beyond the vendor’s release claims. The implementation remains worth passively
- 08-17 22:45alert_silentThere is no new consequential delta: engagement is unchanged and the core reliability and continuity claims remain unverified. It can wait until an independent deployment or failure-recovery test appe
- 08-17 22:45alert_routeThere is no new consequential delta: engagement is unchanged and the core reliability and continuity claims remain unverified. It can wait until an independent deployment or failure-recovery test appe
- 08-17 22:43alert_silentDropstone SDK 1.0 appears to be a real first-party product release, but the only consequential claims are vendor assertions about persistent cross-interface memory. With no independent deployment, rec
- 08-17 22:43surface_candidateDropstone SDK 1.0 appears to be a real first-party product release, but the only consequential claims are vendor assertions about persistent cross-interface memory. With no independent deployment, rec
- 08-17 22:43alert_routeDropstone SDK 1.0 appears to be a real first-party product release, but the only consequential claims are vendor assertions about persistent cross-interface memory. With no independent deployment, rec
- 08-17 22:41groundKnown via ip:framework.long-running-agents and dev:concept.resumable-agent-job-control-plane: Scott already treats durable external state, restartability, and recovery evidence as the basis of multi-s
- 08-17 22:39promote_anchororigin walk conf 0.99
- 08-17 22:38createThe released SDK is a concrete persistent-agent runtime artifact directly relevant to long-running orchestration and memory.