Automaton Durable State’s maintainer claims externalized durable state can preserve persistent-agent continuity while materially reducing repeated context consumption and inference cost.

state: expiredheat: lowuncertainty: highknownscott: lowagent-memory durable-state inference-economicsCary Palmer

What is this?

Automaton Durable State is presented as a project maintained by Cary Palmer that externalizes an agent’s durable state so it can preserve continuity without repeatedly carrying a large context window. The supplied web snippets support the underlying architecture: agents can checkpoint progress and important findings to external storage, resume after resets or failures, and avoid repeating completed inference work. However, the snippets do not independently document this repository, Palmer’s role, or quantify the claimed reduction in Grok Bot consumption and cost.

Why it matters to Scott

Scott’s `long-running-agents` framework already states the same core architecture: stateless workers resume from durable, compressed external state rather than rereading full transcripts. The claimed cost reduction is unquantified and not independently established, so this is currently another implementation example rather than evidence that extends or changes his position.

ip:framework.long-running-agentsip:framework.context-engineeringdev:concept.agent-authored-context-compactionradar:concept.persistent-agentsradar:concept.long-running-agentsradar:concept.inference-economics

queries asked of Scott's wikis

externalized state versus prompt context for agent continuity
agent-maintained files and wikis as durable memory
context resets and selective state rehydration
checkpointing long-running coding agents
token economics of persistent agent workflows
durable execution versus semantic agent memory

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 (5) — ⭐ canonical anchor

source	object	author	score	comments
🟧 hn	Reducing Grok Bot consumption with durable state	carypalmer	4	0
🟧 echo.github ⭐	The repository’s first commit contains the original architecture note, titled “How to reduce GrokBot usage by changing the architecture.” It	Cary Palmer	—	—
🟧 hn	KHMS – a file-based long-term memory an LLM agent installs into itself	ksxcz	7	0
🟧 hn	AIPass – AI agents whose memory is small JSON files they own	AIOSAI	1	0
🟠 reddit	I made a skill and a hook so my 24/7 Claude Code agents pick up where they left off after a restart. One file each ClaudeAI	Any-Bobcat2370	4	7

Interpretation history

2026-09-02T15:46:18Z
The implementation wave has gone quiet without adoption, reliability evaluation, or quantified token and cost savings. Externalized durable state remains a corroborated architectural pattern, but this Automaton-centered episode no longer has active evidence expected within its horizon.
2026-08-31T15:41:40Z
The refreshed comments add only lightweight user interest in the restart-hook implementation, not adoption, evaluation, or quantified savings. The durable-state pattern remains corroborated, while the material inference-cost claim remains unvalidated.
2026-08-31T13:37:32Z
The restart-hook plugin adds practical evidence that small external handoff files can preserve coding-agent continuity across resets, further consolidating the implementation pattern. It still provides no measured token savings, cost comparison, or reliability evaluation, so Automaton’s material inference-economics claim remains unvalidated.
2026-08-31T13:24:20Z
evidence attached: reddit.post.1w3dfaj — A practical independent artifact uses milestone handoff files and restart hooks to preserve coding-agent continuity across sessions.
2026-08-30T07:27:45Z
AIPass becomes a third independent implementation converging on small, agent-owned external state, so the architectural pattern is now corroborated rather than anecdotal. It still adds no continuity evaluation, adoption signal, or measured inference savings, leaving Automaton’s material cost-reduction claim unvalidated.
2026-08-30T07:23:17Z
evidence attached: hn.story.49496317 — First-party artifact supports the open hypothesis that simple externalized, agent-owned state can provide durable persistence for agents.
2026-08-28T16:32:22Z
KHMS adds independent implementation-level convergence around file-based durable agent memory, moving the architecture from a lone artifact to a pattern worth watching. It still provides no continuity evaluation, adoption evidence, or measured inference savings, so Automaton’s material cost-reduction claim remains unvalidated.
2026-08-28T14:25:17Z
evidence attached: hn.story.49478170 — This hunted first-party artifact is an independent file-based implementation of durable agent memory, relevant to whether externalized state can preserve continuity.
2026-08-27T15:44:28Z
The case remains a lone implementation of an already-known architecture, with no adoption, independent validation, or measured cost savings. The unchanged response lowers its near-term attention temperature without disproving the approach.
2026-08-27T15:37:54Z
grounded: known/low — Scott’s `long-running-agents` framework already states the same core architecture: stateless workers resume from durable, compressed external state rather than
2026-08-27T15:36:25Z
origin walked (codex/luna, conf 0.98): anchor hn.story.49465534 -> echo.github.7400f10517 by Cary Palmer
2026-08-27T15:35:22Z
case created — The released implementation is a concrete agent-memory artifact with a specific cost-control claim distinct from existing decision-log and retrieval-memory cases.

Decision trace

09-03 01:46expireThe implementation wave has gone quiet without adoption, reliability evaluation, or quantified token and cost savings. Externalized durable state remains a corroborated architectural pattern, but this
09-03 01:46alert_silentThe staleness check found no consequential delta; a minor score change does not alter the case or justify attention before routine briefing.
09-03 01:46alert_routeThe staleness check found no consequential delta; a minor score change does not alter the case or justify attention before routine briefing.
09-01 01:41repriceThe refreshed comments add only lightweight user interest in the restart-hook implementation, not adoption, evaluation, or quantified savings. The durable-state pattern remains corroborated, while the
09-01 01:41alert_silentThe new discussion is repetitive appreciation rather than a consequential technical or adoption delta, so it does not warrant interrupting Scott before the next briefing.
09-01 01:41alert_routeThe new discussion is repetitive appreciation rather than a consequential technical or adoption delta, so it does not warrant interrupting Scott before the next briefing.
09-01 01:21sensor_dirtycomment_update
09-01 00:21sensor_dirtycomment_update
08-31 23:37repriceThe restart-hook plugin adds practical evidence that small external handoff files can preserve coding-agent continuity across resets, further consolidating the implementation pattern. It still provide
08-31 23:37alert_silentThis is another low-adoption implementation of an architecture Scott already endorses; without quantified savings, comparative reliability, or a novel mechanism, it can wait for routine briefing.
08-31 23:37alert_routeThis is another low-adoption implementation of an architecture Scott already endorses; without quantified savings, comparative reliability, or a novel mechanism, it can wait for routine briefing.
08-31 23:24alert_silentThis is another small implementation of the already-known durable-handoff pattern: one rewritten file is injected after Claude Code restarts or context resets. It adds a concrete plugin design but no
08-31 23:24alert_routeThis is another small implementation of the already-known durable-handoff pattern: one rewritten file is injected after Claude Code restarts or context resets. It adds a concrete plugin design but no
08-31 23:24attachA practical independent artifact uses milestone handoff files and restart hooks to preserve coding-agent continuity across sessions.
08-31 23:22propose_attachA practical independent artifact uses milestone handoff files and restart hooks to preserve coding-agent continuity across sessions.
08-30 17:27repriceAIPass becomes a third independent implementation converging on small, agent-owned external state, so the architectural pattern is now corroborated rather than anecdotal. It still adds no continuity e
08-30 17:27alert_silentThe new artifact strengthens an architecture Scott already endorses but supplies no benchmark, operational result, or novel mechanism that would make the next briefing too late.
08-30 17:27alert_routeThe new artifact strengthens an architecture Scott already endorses but supplies no benchmark, operational result, or novel mechanism that would make the next briefing too late.
08-30 17:23alert_silentAIPass is another lightly evidenced implementation of agents owning small external state files, an architecture Scott already covers. The submission provides no technical detail, quantified context or
08-30 17:23alert_routeAIPass is another lightly evidenced implementation of agents owning small external state files, an architecture Scott already covers. The submission provides no technical detail, quantified context or
08-30 17:23attachFirst-party artifact supports the open hypothesis that simple externalized, agent-owned state can provide durable persistence for agents.
08-30 17:23propose_attachFirst-party artifact supports the open hypothesis that simple externalized, agent-owned state can provide durable persistence for agents.
08-29 03:21sensor_dirtyengagement_update
08-29 02:32repriceKHMS adds independent implementation-level convergence around file-based durable agent memory, moving the architecture from a lone artifact to a pattern worth watching. It still provides no continuity
08-29 02:32alert_silentThe new implementation strengthens the pattern but reports no benchmark, operational result, or novel mechanism that would change Scott’s existing durable-state position; it can wait for routine revie
08-29 02:32alert_routeThe new implementation strengthens the pattern but reports no benchmark, operational result, or novel mechanism that would change Scott’s existing durable-state position; it can wait for routine revie
08-29 00:25alert_silentKHMS appears to be another implementation of file-based external agent memory, but the supplied evidence contains no capability, evaluation, cost measurement, or architectural result that extends Scot
08-29 00:25alert_routeKHMS appears to be another implementation of file-based external agent memory, but the supplied evidence contains no capability, evaluation, cost measurement, or architectural result that extends Scot
08-29 00:25attachThis hunted first-party artifact is an independent file-based implementation of durable agent memory, relevant to whether externalized state can preserve continuity.
08-29 00:24propose_attachThis hunted first-party artifact is an independent file-based implementation of durable agent memory, relevant to whether externalized state can preserve continuity.
08-28 01:44repriceThe case remains a lone implementation of an already-known architecture, with no adoption, independent validation, or measured cost savings. The unchanged response lowers its near-term attention tempe
08-28 01:44alert_silentNo new consequential delta occurred; the artifact still lacks benchmarks, independent users, or technical novelty beyond Scott’s existing long-running-agent framework, so another briefing would add li
08-28 01:44alert_routeNo new consequential delta occurred; the artifact still lacks benchmarks, independent users, or technical novelty beyond Scott’s existing long-running-agent framework, so another briefing would add li
08-28 01:42alert_silentThis is a small implementation example of an architecture Scott already advocates: external durable state, bounded context, selective retrieval, and artifact reuse. The maintainer provides no quantifi
08-28 01:42alert_routeThis is a small implementation example of an architecture Scott already advocates: external durable state, bounded context, selective retrieval, and artifact reuse. The maintainer provides no quantifi
08-28 01:37groundScott’s `long-running-agents` framework already states the same core architecture: stateless workers resume from durable, compressed external state rather than rereading full transcripts. The claimed
08-28 01:36promote_anchororigin walk conf 0.98
08-28 01:35createThe released implementation is a concrete agent-memory artifact with a specific cost-control claim distinct from existing decision-log and retrieval-memory cases.