Skip to main content
Intramedial
AS

Read-only demo account · Enterprise

The Awesome Company, Inc.

All capabilities enabled · unlimited quotas · 42 active agents · 4 workspaces

Preview · Tops Configuration

Tops control plane (demo)

Marketing mirror of /app/tops/config. Cascade · effective config · presets with diff · actor testbed · change history from the config change log.

Adversary-safety posture

Why Tops records hashes + verdict buckets only by default.

Safe
Default · adversary-readable: low

Payload hashes + verdict buckets + reason codes. Auditors can confirm a decision happened and verify against your own data store, without exposing thresholds or raw content.

Extended
On request · adversary-readable: medium

Adds threshold values + reason strings + check parameters. Useful during active investigation but exposes detection signals an attacker could probe. Upgrade writes a tops_config_change meta-entry.

Forensic
Use sparingly · adversary-readable: high

Adds raw payload references (pointers to your data store; never content). For active incident response only. Cool-down banner appears after 7 days; revert promptly when the incident closes.

Detail-level changes are themselves audited — turning Tops up (or off) writes a tops_config_change entry into the Tops chain. Reverts write the same kind of entry.

Effective config (resolved)

Each field's value with the scope it came from. This is what the actor runtime gates against.

FieldValueFrom scope
EnabledonWorkspace · EU Production
actor_observeonWorkspace · EU Production
actor_decideonWorkspace · EU Production
actor_skiponWorkspace · EU Production
actor_stateonWorkspace · EU Production
actor_driftonWorkspace · EU Production
Environmentsproduction, stagingWorkspace · EU Production
Sampling rate100%Workspace · EU Production
Detail levelextendedWorkspace · EU Production

Editor · Workspace · EU Production

Field-level configuration for this scope. Save writes a tops_config_change entry.

Tops enabled at this scope

When off, no spot_* entries emit at this scope or any inheriting scope (unless that scope has its own override).

Record kinds

Which of the five Spot kinds emit entries when enabled.

actor_observe· Observe

What the actor saw entering its turn — the inputs, the tool outputs, the context window. Recorded as payload hashes + type tags by default.

actor_decide· Decide

Which check fired with which verdict on the workspace's outputs. The compliance heart of the ledger — recorded with reason codes + threshold buckets at safe detail.

actor_skip· Skip

Checks that intentionally did not fire for this turn — typically because the input type was not in scope. Important for the completeness defence: 'we ran every applicable check'.

actor_state· State

State transitions inside the actor — model swaps, tool changes, memory writes that change the actor's posture turn-over-turn.

actor_drift· Drift

Self-monitoring findings — when the system itself thinks behaviour has shifted. These are typically rare events but high-value for the auditor.

Environments

Only emit entries when the actor's runtime environment matches one of these.

Sampling rate

Fraction of events that emit. Deterministic per (actorId, eventId), so the same event consistently emits or consistently doesn't across retries.

100%

At this rate ≈ 12,500 entries/day for this workspace (based on the last 7 days at 100% sampling: ~12,500 entries/day).

Detail level

Adversary-safety tradeoff. Upgrading to extended or forensic writes a separate tops_config_change meta-entry recording the decision itself.

Apply preset

One-click configurations covering the 90% case. You can tweak any field after applying.

Actor testbed

Pick an actor or type any ID to preview the effective config the runtime would gate against.

Suggested:

Configuration history

The append-only config change log. Every change is here — including the one that turned Tops off.

Detail-level activity (last 8 changes)
Safe Extended Forensic
9 entries
  1. Workspace · eu-prodDetail: safe → extendedMay 18, 2026, 09:42 AM

    “Bumped EU prod to extended detail during incident #4127 (Council drift investigation). Revert to safe after RCA closes.”

    by acct_priya_devesh · change tops_cfg_chg_026
  2. Actor · assistant-prodConfig createdMay 19, 2026, 02:00 PM

    “Pinned assistant-prod actor override to record sandbox sessions alongside production while the new beta cohort onboards.”

    by acct_priya_devesh · change tops_cfg_chg_025
  3. Actor · billing-botMay 10, 2026, 08:22 AM

    “Dropped billing-bot sampling to 10% — high-volume cron actor, no compliance value at full sampling.”

    by acct_julien_carda · change tops_cfg_chg_024
  4. Actor · analytics-workerMay 5, 2026, 05:51 PM

    “Disabled analytics-worker actor recording — pure read-only analytics, no decisions to record.”

    by acct_julien_carda · change tops_cfg_chg_023
  5. Workspace · us-prodConfig createdApr 29, 2026, 03:08 PM

    “Applied Minimal preset to US prod. Cost-cutting pass after the Q1 spend review surfaced actor_observe as 71% of US prod's Tops volume.”

    by acct_julien_carda · change tops_cfg_chg_022
  6. Org · org_theawesomecompanyConfig createdApr 12, 2026, 11:14 AM

    “Initial org default applied via Recommended preset at org signup.”

    by acct_michele_brogli · change tops_cfg_chg_021
  7. Workspace · eu-prodDetail: extended → safeApr 3, 2026, 10:11 AM

    “Reverted to safe after incident #4012 closed.”

    by acct_priya_devesh · change tops_cfg_chg_020
  8. Workspace · eu-prodDetail: safe → extendedMar 31, 2026, 04:44 PM

    “Extended detail during incident #4012 (Council drift spike).”

    author not recorded · change tops_cfg_chg_019
  9. Actor · fraud-watchConfig deletedMar 22, 2026, 08:05 AM

    “Removed the fraud-watch override so it follows EU prod again — the reduced sampling was only for the migration window.”

    by acct_julien_carda · change tops_cfg_chg_018