N
NatorOS
Sign inBook a demo
← Resources
EngineeringApr 28, 202610 min read

Replayable agents: why every step has a fingerprint

By Devon Park · Engineering, NatorOS
Replayable agents

An agent run on NatorOS is a directed graph of steps. Each step has an input, a tool or model call, and an output. We record all three for every step of every run, then make them queryable, replayable, and diffable. This post explains why.

The problem with stateless logs

Most agent frameworks log a flat list of LLM calls and tool calls. That is enough to debug a single failure if you happen to catch it. It is not enough to answer the questions a CIO actually asks:

  • Why did this agent approve this invoice last Tuesday?
  • Has it ever made a different decision on the same kind of input?
  • If we changed the prompt yesterday, what would that change have done to the last 200 runs?

Those questions require state, structure, and the ability to replay.

What we record

Each step writes a row to an append-only event log:

run_id, step_id, parent_step_id, agent, kind, input_hash, output_hash, model, model_version, started_at, finished_at, latency_ms, cost_cents, approver_id, status

The input and output are stored as content-addressed blobs. Identical inputs yield identical hashes; we deduplicate. The fingerprint of any run is the concatenation of its step hashes.

Replay, not just inspect

Because every input is captured, we can take any past run and re-execute it against a different agent version. This is how every workflow change is validated before it ships. We pick a window of recent runs, replay them against the new agent, diff the outputs, and surface the ones that changed. A human reviews the diffs before the new version is promoted.

When something does go wrong in production, the same machinery lets you bisect: replay the same input against the last twenty agent versions and find the one where the behavior changed.

The cost

Recording every step costs us roughly four percent of total runtime CPU and a few cents of storage per thousand runs. It is the highest-value four percent we spend.

Every customer we have ever onboarded has, at some point, asked us to explain a specific past decision. Replayable runs have answered every one of those questions in under a minute. That is the only thing that lets a CIO sleep at night.

Talk to us
Want to see what an AI worker would look like for your team?
Book a demo

Keep reading

Long-running agents

Long-running agents on a stateless runtime

EngineeringApr 7, 2026
The runtime owns this

Memory, tools, permissions: the three things the runtime owns

EngineeringMar 3, 2026
Approvals as product surface

Approvals are product surface area: designing human-in-the-loop agent work

PracticeJul 11, 2026

Hire your AI workforce.

Book a demo Talk to founders
NatorOS, Inc. · 2026PrivacyTerms of servicellms.txt