Pro sells Patrol investigation, but every install starts in Watch only
and stays there unless the operator finds the mode control. In the paid
telemetry read on 12 October, 20 of 35 Pro installs with Patrol on ran
Watch only, and none of those 20 ran an investigation in 30 days, so
the feature they paid for never ran.
An install that never saved a Patrol mode now runs Ask first when it
holds a positive investigation licence. Patrol investigates new and
re-detected warning and critical findings and proposes a fix; nothing
changes without approval. The default only applies to a positive
licence answer: a missing licence checker counts as unlicensed, so a
startup window cannot spend provider budget on a free install. Free
installs and explicit choices are unchanged.
The Patrol header says so once, with the cost (the operator's AI
provider and 30-day budget) and two actions that save Ask first or
Watch only. The notice never renders on a plan-locked install. The
header also stops claiming "Watching your infrastructure" while Patrol
cannot run yet and reads "Patrol is not running yet" instead.
Reverts #2863 before any release ships it. The live Pro pricing and
Patrol pages, in English, German and Spanish, sell both modes ("applies
safe fixes in the mode you choose: Ask first, Safe auto-fix, or
Autopilot"), and paying installs chose them. Shipping the retirement
would have silently downgraded those installs while the sales page still
promised the modes.
If the modes are retired, that needs a dated notice: stop selling them
first, warn installs on those modes in-app, and then re-ship the change.
It should not arrive through an engineering cleanup.
The receipt #2863 landed stays in place; this commit adds the receipt
for the restored content.
Patrol could run some fixes on its own in two paid modes: assisted (Safe
auto-fix) and full (Autopilot, behind a server-owned acknowledgement).
Schema-v17 telemetry for the week to 2026-10-11 counts 1 assisted and 2
full installs out of 612 reporting Patrol, 4 executed Patrol actions in
30 days, all on one install, and no GitHub request for more autonomy,
while the automatic path carried its own policy leases, acknowledgement
store, per-resource allowlist, release gate and Colima lab. Investigation
is what paid installs use, so it stays.
Patrol now has two modes. Watch only reports findings. Ask first
investigates and proposes fixes, and every fix waits in Actions for an
operator to approve and run it.
- A saved assisted, full or autonomous level reads as approval, and the
update routes refuse those levels with 400 invalid_autonomy_level.
- The core Patrol action broker only plans: Plan and Submit persist the
plan and return. ExecuteUnderPolicy, the policy authorization lease,
the policy admission store methods, the Autopilot acknowledgement
persistence, routes and dialog, and the per-resource automatic-action
allowlist (store column now written as NULL) are gone.
GetPatrolFullModeUnlocked stays on the extension contract and reports
false.
- Historical action records keep their policy lease and tenant/resource
reason codes decodable, labelled as retired modes.
- The Patrol inbox no longer splits issues into "continuing without a
decision"; every current issue is an operator decision.
- Release gate RG-06 now proves Patrol never executes without an operator
decision (integration tier) instead of limited unattended autonomy,
and RG-02/RG-04 point at the surviving human-admission proofs.
- Pro plan copy, AI docs, privacy telemetry wording and the subsystem
contracts describe the two modes.
Audited the customer-facing docs against the code: about 5,200 concrete
claims (env vars, endpoints, flags, defaults, ports, UI labels), plus every
environment variable the code reads. Each discrepancy was re-verified
against the source before editing, and every edit was reviewed again.
What users would have hit following the old text:
- UI paths and labels that no longer exist (Settings > Relay, Platform
connections, Access Control, SSO Providers, Create Organization) and
Patrol mode names (now Ask first, Safe auto-fix, Autopilot).
- Wrong values and behaviour: LOG_MAX_AGE=0 does not disable cleanup,
Patrol evidence budget is 10 calls and 12 turns, ALLOW_UNPROTECTED_EXPORT
covers private networks too, PULSE_AUDIT_ASYNC defaults on.
- Agent install: install.sh ignores PULSE_ENABLE_KUBERNETES and
PULSE_ENABLE_PROXMOX, rejects --disable-auto-update, and the token
fallback lives in the state directory.
- API: audit endpoints answer 402 on Community, onboarding QR is GET,
type=container is rejected, several routes need admin plus a scope.
- HTTP_REDIRECT_PORT was not documented anywhere.
Docs are mirrored under frontend-modern/public/docs, so each change is
applied to both copies. The tests that pinned the old wording (including
three that asserted claims the code contradicts) are updated alongside.
Contract-Neutral: Customer docs corrected to match shipped behaviour; no runtime, API or contract change.
Settings > Assistant offered a Protected guests list (protected_guests)
that was saved and pushed into the Assistant tool executor, but its only
reader was executeControlGuest, which nothing calls. pulse_control plans
through the shared action lifecycle and never consulted the list, so a
guest named in it (by VMID and by name) was planned like any other. The
"legacy" label added in #2765 states that limit; this removes the
setting instead of keeping a field that cannot do its job.
The canonical way to keep Pulse off a resource is its operator lock
(Never auto-remediate, or a Retired lifecycle). It is keyed by canonical
resource id, applies to every platform, and the action lifecycle refuses
dispatch for it even for approved plans. The one gap was planning: the
lifecycle plans without reading operator state, so the Assistant could
leave a plan in Actions that can never run. assistantTypedActionPlanner
now reads the state through the tenant-scoped lifecycle store (the same
read Patrol's broker uses) and declines before Plan, and pulse_control
reports the block with how to clear it. A state that cannot be read does
not refuse planning, because dispatch re-checks.
Removed: AIConfig.ProtectedGuests and its getter, the settings API
field, the executor field/setter/config, the chat service plumbing, the
Settings field and form state, and the legacy copy, pins and prose from
#2765. A stored value still loads (tolerant decode) and is dropped the
next time the settings are written; older PUT bodies that send the key
are ignored. The Settings panel points at the guest's Manage tab.
The current canonical resource planner does not consult the retained guest list. Label that compatibility data honestly and link its accessible limitation without discarding saved policy. Keep the review-only mode menu within the composer on phones, with geometry and resize controls.
Reconcile the old autonomous-mode test with the explicit core-owned investigation flag, preserving saved preferences and legacy approvals. Mounted settings, parent-discriminating copy/menu checks, help mirrors and final-byte browser receipts cover the existing flow.
Change-source: pulse-maintainer
Project interactive Assistant controls to Read-only and Ask first without migrating legacy stored preferences or relaxing entitlement, legacy command approval, scoped execution or Patrol policy. Preserve stored policy even when an older client echoes the projected mode on an unrelated save. Explain saved Actions plans and operator review in existing Settings, Chat and help.
Add saved-client and encrypted-config compatibility, authenticated plan-only access, active-stream and scope regression controls with parent comparisons. Record inspected final-byte desktop/phone browser proof and the affected subsystem contracts.
Change-source: pulse-maintainer
Reuse the existing API helper for saved Patrol settings and deliberate Assistant control changes. Preserve methods, payloads and access scopes while keeping private response bodies out of terminal recordings and making lost-write readback explicit.
Contract-Neutral: Documentation and recipe regression tests only; existing AI settings methods, payloads and access controls are unchanged.
Change-source: pulse-maintainer
Use private header files rather than shared passwords or proxy secrets in process arguments. Separate intentional AI settings writes from diagnostics, distinguish proxy denials from authentication/transport failures, and preserve the real proxy trust boundary and 5.x limitation. Exercise copied commands, private-file preparation and rendered help.
Contract-Neutral: Documentation and executable help controls only; no authentication, authorization, AI autonomy or runtime contract change.
Change-source: pulse-maintainer
AI.md, AI_AUTONOMY.md and PULSE_PRO.md linked nine times into
docs/architecture/, which .gitignore marks as not for public release. The
targets were never missing, they were deliberately unpublished, so every one
of those links was dead for anybody but the maintainer.
Write the three promised documents against the code and publish them under
docs/ where the rest of the shipped set lives.
PATROL_ARCHITECTURE.md covers a run end to end. The interesting part is that
deterministic signal detection runs after the model, not before, so unmatched
signals catch what the model failed to file. Signal types, the thresholds
those signals derive from your own alert settings, and every condition in
Finding.ShouldInvestigate are documented from internal/ai/patrol_signals.go
and internal/ai/findings.go.
ASSISTANT_SAFETY.md documents the session state machine in
internal/ai/chat/fsm.go, its four states, the tool classification it runs on,
and its invariants. No write without a validated target, no second write
before the first is verified, no final answer about an unverified change, and
no attempt count that wears the gate down.
ASSISTANT_ARCHITECTURE.md covers the agentic loop around that machine, the
three-phase pipeline, why only execution parallelises and at what cap, the
read-before-write batch that must stay ordered, the look-before-asking gate
and its bound, and the stable error codes.
Also removed two older pointers into the same private directory, in API.md
and UPGRADE_v6.md, and the two references to ENTITLEMENT_MATRIX.md in
PULSE_PRO.md. That file exists locally and is a monetization document, so
publishing it is not a documentation decision.
Unresolvable intra-doc links are now 2 of 218, both internal release-control
documents deliberately withheld from the shipped set.
Contract-Neutral: documentation only
Manifest-backed MCP tools, prompts, and resources with surface affordance contracts; agent capability manifest and governance projection; API contract tests and capability route projection; operations-loop and intelligence-funnel telemetry; release-control subsystem documentation, registry, and tooling; licensing and configuration.
rcourtman