Releases: rcourtman/Pulse
Release list
Pulse v6.3.0-rc.6
✅ Release Asset Validation: PASSED
All release assets have been validated successfully!
Status: Ready for publication ✅
Validated: 2026-08-21 18:55:20 UTC
Workflow: Pulse Release Pipeline #368
Validation Summary
- All required assets present ✓
- Checksums verified ✓
- Version strings correct ✓
- Binary architectures validated ✓
Pulse v6.3.0-rc.6 Release Notes
v6.3.0-rc.6 is a release candidate for the next Pulse v6 minor release. It
follows v6.3.0-rc.5 and stable v6.2.1, while keeping stable and latest users
on that stable release.
This candidate carries the complete 6.3 product packet and concentrates on
release turnaround. The release graph now makes fuller use of the dedicated
Ryzen PVE worker, moves more API production domains into independently
schedulable packages, and overlaps public and private publication work without
moving any customer-facing pointer before exact-SHA readiness.
Highlights
- Chart and resource-query services now qualify independently from the residual
API router, shrinking the root test critical path. - Backend admission waits for sibling compilers, then requires measured
headroom for two race-enabled API shards. - Public server and provider control-plane images publish and attest in
parallel from one verified exact-candidate payload.
Improved
- Patrol now works from durable outcomes, scoped investigations, and verified
work receipts instead of treating the chat stream as operational state. - Read-only observers extend Patrol coverage between full model investigations
without granting mutation authority. - Actions and Patrol identify whether a decision originated from a finding,
alert, objective, or explicit operator request, making review context clearer. - Action refusal telemetry now classifies target changes, prerequisites,
contract failures, capability limits, policy decisions, and stale plans. - Subscription-backed turns now complete their idle timeout promptly even when
a canceled CLI descendant still holds an inherited output pipe open. - Platform pages now lead with estate totals, status facets, and search that
share the same predicates as their underlying tables. - Notification settings show the outcome of real delivery attempts instead of
relying on test sends as a proxy for live delivery health. - Docker-in-LXC discovery is explicitly controlled and backs off against slow
or failing Proxmox hosts instead of creating a probe storm. - Unified Agent installs can opt into a supported least-privilege profile with
narrowly scoped elevation for the capabilities that require it. - Exact-version Docker staging begins as soon as the immutable candidate
exists. Candidate container qualification and every other immutable gate
still join before release activation. - Private release artifacts use a purpose-built Pro packaging profile and
compressed transfer rather than shipping unused frontend, MCP, public-server,
and control-plane products through the credential boundary. - Paid-runtime convergence uses the runner's installed Chrome with the pinned
Playwright client and runs Docker and direct-binary mismatch proofs in
parallel on separate ports.
Fixed
- Release activation recovery now joins the canonical readiness result instead
of maintaining a stale duplicate catalog of historical job names. - Helm convergence binds repository operations explicitly when the workflow is
executing from its nested Pages checkout. - Every hosted paid-runtime proof establishes its own tailnet connection before
calling the private license service. - The backend planner reports a capacity failure if the dedicated worker cannot
admit the required two-shard release shape after its bounded wait.
Security
- PVE compilation remains credential-free. GitHub-hosted jobs retain signing,
release mutation, and publication credentials. - Container images, Helm charts, public archives, private archives, checksums,
signatures, and convergence proof remain tied to the release commit and the
source workflow run. - Each parallel Docker matrix leg independently verifies the exact checkout,
anticipated release line, and immutable candidate manifest before publishing. - The chart and resource boundaries preserve tenant scope, router authorization,
action ownership, and operator-state mutation at explicit package seams.
Upgrade Notes
Use the normal v6 install or update flow for v6.3.0-rc.6 only when you are
comfortable testing a release candidate. The rollback target is v6.2.1:
./scripts/install.sh --version v6.2.1The changes since v6.3.0-rc.5 do not require a Pulse Mobile client change and
preserve the existing mobile, Relay, onboarding, and mobile-facing API
contracts, so no companion mobile build is required.
macOS artifacts follow the normal signing and notarization path. Windows
Unified Agent binaries in this prerelease retain exact-SHA, checksum, and
detached-signature verification but are not Authenticode-signed, so Windows may
display an Unknown Publisher warning. Stable v6.3.0 still requires the normal
SignPath Authenticode lane.
Paid Pulse Pro, Relay, and eligible legacy customers should continue to use the
private download page and private runtime image for paid runtime features.
Installation
Docker (recommended):
docker pull rcourtman/pulse:6.3.0-rc.6Docker Compose:
Update your docker-compose.yml to use rcourtman/pulse:6.3.0-rc.6
See the Installation Guide for complete setup instructions.
Review the Code signing policy for release provenance, approval roles, and signing scope.
Paid Pulse Pro, Relay, and eligible legacy customers: public GitHub release assets and the public rcourtman/pulse Docker image are community builds. They do not include the private Pulse Pro runtime hooks. Use https://pulserelay.pro/download.html with your activation key to get the private Pulse Pro Docker image or Linux/LXC archive.
Promotion Metadata
- Promotion channel: rc
- Candidate stable tag: v6.3.0-rc.6
- Promoted prerelease tag: n/a
- Rollback target: v6.2.1
- Rollback command:
./scripts/install.sh --version v6.2.1 - Hotfix exception: false
- Windows Authenticode required: false
- Unsigned Windows exception: false
Pulse v6.3.0-rc.5
✅ Release Asset Validation: PASSED
All release assets have been validated successfully!
Status: Ready for publication ✅
Validated: 2026-08-21 16:28:16 UTC
Workflow: Pulse Release Pipeline #365
Validation Summary
- All required assets present ✓
- Checksums verified ✓
- Version strings correct ✓
- Binary architectures validated ✓
Pulse v6.3.0-rc.5 Release Notes
v6.3.0-rc.5 is a release candidate for the next Pulse v6 minor release. It
follows v6.3.0-rc.4 and stable v6.2.1, while keeping stable and latest
users on that stable release.
This candidate carries the complete 6.3 product packet from rc.4. Its new
work is focused on release integrity and turnaround: release payloads are
compiled and qualified on the dedicated PVE worker, independent checks run in
parallel, and every public and paid-runtime artifact remains bound to the exact
source commit that passed qualification.
Highlights
- Release payload compilation now uses the dedicated Ryzen PVE worker and its
persistent caches instead of leaving that hardware idle. - Archive, backend, container, Helm, installer, and paid-runtime checks overlap
where their trust boundaries permit it. - Container publication promotes the already-qualified exact-candidate payload
rather than rebuilding release binaries after qualification.
Improved
- Patrol now works from durable outcomes, scoped investigations, and verified
work receipts instead of treating the chat stream as operational state. - Read-only observers extend Patrol coverage between full model investigations
without granting mutation authority. - Actions and Patrol identify whether a decision originated from a finding,
alert, objective, or explicit operator request, making review context clearer. - Action refusal telemetry now classifies target changes, prerequisites,
contract failures, capability limits, policy decisions, and stale plans. - Subscription-backed turns now complete their idle timeout promptly even when
a canceled CLI descendant still holds an inherited output pipe open. - Platform pages now lead with estate totals, status facets, and search that
share the same predicates as their underlying tables. - Notification settings show the outcome of real delivery attempts instead of
relying on test sends as a proxy for live delivery health. - Docker-in-LXC discovery is explicitly controlled and backs off against slow
or failing Proxmox hosts instead of creating a probe storm. - Unified Agent installs can opt into a supported least-privilege profile with
narrowly scoped elevation for the capabilities that require it. - Platform release archives are validated concurrently, while each archive is
decompressed only once for its required-entry checks. - Public and private release payloads are staged during qualification so later
publication jobs consume immutable evidence instead of repeating builds. - Helm convergence verifies and publishes the chart artifact already produced
by the source release run, avoiding duplicate packaging and cluster smoke. - Paid-runtime convergence separates independent public/download-page checks
from the lease-bound customer-path proof so both can run at the same time. - Activation and child-workflow polling use short bounded intervals, reducing
avoidable idle time without weakening their timeouts or failure reporting. - The API runtime now has explicit alerting, configuration, agent-token,
agent-binding, request-context, and HTTP-scope package boundaries. Existing
routes and extension contracts remain compatible while the test graph gains
useful package-level parallelism.
Fixed
- Architecture-specific server signatures are validated against the matching
platform binary rather than incorrectly requiring every platform signature
to be identical. - Windows compatibility aliases in the container payload are validated as
exact symlinks and then recreated by the image build, keeping the immutable
manifest limited to regular files. - Candidate qualification now runs even when intentionally disabled native
signing jobs are skipped, while still failing closed on a required signing
failure. - Backend qualification prefix-compresses the exact API test selector, keeping
a memory-driven one-shard fallback in one ordered process when safe while
still byte-bounding deterministic batches below the operating system limit.
Security
- PVE compilation remains credential-free. GitHub-hosted jobs retain signing,
release mutation, and publication credentials. - Container images, Helm charts, public archives, private archives, checksums,
signatures, and convergence proof remain tied to the release commit and the
source workflow run. - The API decomposition preserves tenant scope, install-token binding, command
channel binding, and notification delivery boundaries with focused contract
tests at the new package seams.
Upgrade Notes
Use the normal v6 install or update flow for v6.3.0-rc.5 only when you are
comfortable testing a release candidate. The rollback target is v6.2.1:
./scripts/install.sh --version v6.2.1The changes since v6.3.0-rc.4 do not require a Pulse Mobile client change and
preserve the existing mobile, Relay, onboarding, and mobile-facing API
contracts, so no companion mobile build is required.
macOS artifacts follow the normal signing and notarization path. Windows
Unified Agent binaries in this prerelease retain exact-SHA, checksum, and
detached-signature verification but are not Authenticode-signed, so Windows may
display an Unknown Publisher warning. Stable v6.3.0 still requires the normal
SignPath Authenticode lane.
Paid Pulse Pro, Relay, and eligible legacy customers should continue to use the
private download page and private runtime image for paid runtime features.
Installation
Docker (recommended):
docker pull rcourtman/pulse:6.3.0-rc.5Docker Compose:
Update your docker-compose.yml to use rcourtman/pulse:6.3.0-rc.5
See the Installation Guide for complete setup instructions.
Review the Code signing policy for release provenance, approval roles, and signing scope.
Paid Pulse Pro, Relay, and eligible legacy customers: public GitHub release assets and the public rcourtman/pulse Docker image are community builds. They do not include the private Pulse Pro runtime hooks. Use https://pulserelay.pro/download.html with your activation key to get the private Pulse Pro Docker image or Linux/LXC archive.
Promotion Metadata
- Promotion channel: rc
- Candidate stable tag: v6.3.0-rc.5
- Promoted prerelease tag: n/a
- Rollback target: v6.2.1
- Rollback command:
./scripts/install.sh --version v6.2.1 - Hotfix exception: false
- Windows Authenticode required: false
- Unsigned Windows exception: false
Pulse v6.3.0-rc.4
✅ Release Asset Validation: PASSED
All release assets have been validated successfully!
Status: Ready for publication ✅
Validated: 2026-08-21 09:59:09 UTC
Workflow: Pulse Release Pipeline #356
Validation Summary
- All required assets present ✓
- Checksums verified ✓
- Version strings correct ✓
- Binary architectures validated ✓
Pulse v6.3.0-rc.4 Release Notes
v6.3.0-rc.4 is a release candidate for the next Pulse v6 minor release,
following v6.3.0-rc.3 in the 6.3 line and stable v6.2.1. It carries the
Patrol v2 operating model, a first-class Actions workspace, guarded agent
action preflight, large-estate response improvements, and monitoring
correctness fixes.
This candidate adds estate-first platform summaries and search, an operator-
visible notification delivery log, a supported least-privilege agent profile,
and safer Docker-in-LXC discovery. It also prevents failed settings or AI-state
reads from silently overwriting preserved data, restores distinct alert-history
recurrences, and hardens agent operation on QNAP and hosts with stalled mounts.
Highlights
- Read-only observers extend Patrol coverage between full model investigations
without granting mutation authority. - Platform pages now lead with estate totals, status facets, and search that
share the same predicates as their underlying tables. - Notification settings show the outcome of real delivery attempts instead of
relying on test sends as a proxy for live delivery health.
Added
- Durable Patrol objectives with pause, archive, resource-scope, coverage, and
observer-health state. - Model-authored observer proposals that are validated, installed as bounded
read-only checks, and kept separate from action authority. - Verified Patrol work receipts and clearer navigation between findings,
objectives, attention items, and governed actions. - A persistent decision workspace with explicit decision origin, review
progress, next-item navigation, and automatic advancement after a decision. - A canonical platform-admission facet projected on unified resources so
supported onboarding and runtime modes stay consistent across the product. - Unified Agent preflight contracts for package updates, package-cache cleanup,
and Docker lifecycle or update operations. - Production security deployment guidance and a focused security-review packet.
- Estate overviews and infrastructure search across Proxmox, Docker,
Kubernetes, TrueNAS, VMware, and standalone agent surfaces. - A seven-day notification delivery activity log with destination and outcome
details for queued live alerts. - An administrator setting for Docker-in-LXC discovery.
- Unified Agent installs can opt into a supported least-privilege profile with
narrowly scoped elevation for the capabilities that require it.
Improved
- Patrol now works from durable outcomes, scoped investigations, and verified
work receipts instead of treating the chat stream as operational state. - Patrol investigations preserve objective and resource intent across retries,
provider interruptions, chat restarts, truncated responses, and retained
objective runs. - Finding identity, evidence, causal conclusions, and remediation proposals are
canonicalized and validated before Patrol writes or acts on them. - Autonomous execution remains bounded by advertised capabilities, explicit
policy, agent preflight, current target state, and post-action verification. - Patrol separates its Inbox, Protection, and Activity modes, removes duplicate
counters and generic state, and keeps Assistant focused on explaining the
selected item rather than acting as the operations front door. - Authenticated startup no longer pulls the full legacy state payload. Unified
resource pages load concurrently after the first page and no longer wait for
WebSocket hydration before beginning the canonical fetch. - Actions and Patrol identify whether a decision originated from a finding,
alert, objective, or explicit operator request, making review context clearer. - Action refusal telemetry now classifies target changes, prerequisites,
contract failures, capability limits, policy decisions, and stale plans
instead of collapsing the new agent reason codes into the catch-all bucket. - API gzip handling preserves informational and bodyless responses, while
polling, metric lookup, registry resolution, and source-target mapping avoid
repeated large-estate scans. - OpenRouter, Ollama, and subscription-backed Patrol routes handle reasoning
limits, readiness checks, deadlines, and continuation latency more reliably. - Subscription-backed turns now complete their idle timeout promptly even when
a canceled CLI descendant still holds an inherited output pipe open. - Platform and alert-history facet counts are derived from each table's own
filter predicate, keeping summary totals aligned with the visible rows. - Failed settings and AI-state reads preserve the last known data and surface
the failure instead of treating unreadable storage as an empty value. - Docker-in-LXC discovery is explicitly controlled and backs off against slow
or failing Proxmox hosts instead of creating a probe storm.
Fixed
- Patrol no longer accepts empty, blocked, contradicted, incoherent, unscoped,
or unsupported findings and proposals as successful investigation output. - Enabling full AI mode or restarting the chat provider now preserves and
rewires Patrol controls and investigation dependencies. - Platform-admission state stays current across tenant changes, reconnects,
shell navigation, and canonical resource aggregation. - Agent Doctor names the judged credential in verdict evidence, and agents
warn when the server overrides a presented agent identity. - Alert threshold overrides resolve through canonical registry identity, and
the monitoring-policy menu no longer clips inside alert cards. - Docker health-check dependencies and app-container scope now remain attached
to the correct canonical findings. - Stale ZFS alerts clear when storage loses its pool attachment, and node-local
ZFS pools are no longer attached to shared storage records. - vSphere backup status, agent thermal history, explicit cluster-member address
overrides, and discovery-analysis request timeouts now reflect their actual
runtime state. - Docker-in-LXC discovery no longer retries overlapping probes against slow
Proxmox hosts, and command enablement after install is reflected without a
reinstall. - In-progress Proxmox backups are no longer presented as completed backups.
- The notifications surface now states when alert delivery is paused, and that
a passing test send does not prove live alerts are getting through. Test
sends bypass the delivery pause, so a configured destination could look
healthy while every real alert was dropped before it reached the queue. - Degraded notification delivery is reported on the alerts overview rather than
only on the destinations configuration tab, and Pulse raises an alert about
its own delivery when it fails, so a dead notification channel is visible
where a channel failure cannot announce itself. - The alert flapping cooldown now suppresses for its configured duration. The
cooldown was recorded and never read, so suppression ended as soon as the
measurement window drained and a resource oscillating just under the
threshold was never damped at all. - Per-guest overrides that do not set powered-off severity continue to inherit
the global severity instead of silently being stamped as warning. - Each alert recurrence now creates its own history row after the preceding
occurrence resolves, while genuine short observation gaps still coalesce. - Notification delivery retries now cover roughly three minutes by default, so
a destination restarting with monitored infrastructure can recover before
the notification is dead-lettered. - QNAP installs place the agent runtime and download staging on the persistent
data volume instead of requiring tens of MiB on the RAM-backed root. - An unreachable network mount no longer freezes the entire host disk collector
or delays agent shutdown; skipped network filesystems are filtered before
usage probes and remaining probes are bounded. - Unix Repair Authentication commands pin the existing agent ID and hostname,
matching Windows behavior and preventing a repair from creating a suffixed
duplicate identity.
Security
- Patrol observers are read-only and proposal-only. Mutating operations still
require the normal action-policy, approval, capability, preflight, and
verification path. - Agent preflight responses expose bounded machine reason codes rather than
command output, paths, package names, or provider-specific error text. - Investigation tooling is projected from the selected resource scope and
rejects tools that were not advertised for that run.
Upgrade Notes
Use the normal v6 install or update flow for v6.3.0-rc.4 only when you are
comfortable testing a release candidate. The rollback target is v6.2.1.
The exact rollback reinstall command is:
./scripts/install.sh --version v6.2.1The changes since v6.3.0-rc.3 do not require a Pulse Mobile client change
and preserve the existing mobile, Relay, onboarding, and mobile-facing API
contracts. No companion mobile build upload or public mobile-store rollout is
part of this candidate.
Windows Unified Agent binaries in this prerelease retain exact-SHA, checksum,
and detached-signature verification but are not Authenticode-signed, so Windows
may display an Unknown Publisher warning. Stable v6.3.0 still requires the
normal SignPath Authenticode lane unless a separate version-bound owner decision
is recorded.
Paid Pulse Pro, Relay, and eligible legacy customers should continue to use the
p...
Helm chart 6.3.0-rc.6
Helm chart for Pulse 6.3.0-rc.6.
Helm chart 6.3.0-rc.5
Helm chart for Pulse 6.3.0-rc.5.
helm-chart-6.3.0-rc.4
Helm chart for deploying the Pulse hub and optional Docker, Kubernetes, or OpenShift monitoring agent.
Pulse v6.3.0-rc.3
✅ Release Asset Validation: PASSED
All release assets have been validated successfully!
Status: Ready for publication ✅
Validated: 2026-08-18 23:24:47 UTC
Workflow: Pulse Release Pipeline #353
Validation Summary
- All required assets present ✓
- Checksums verified ✓
- Version strings correct ✓
- Binary architectures validated ✓
Pulse v6.3.0-rc.3 Release Notes
v6.3.0-rc.3 is a release candidate for the next Pulse v6 minor release,
following v6.3.0-rc.2 in the 6.3 line and stable v6.2.1. It carries the
Patrol v2 operating model, a first-class Actions workspace, guarded agent
action preflight, large-estate response improvements, and monitoring
correctness fixes.
This candidate adds a group of alert-delivery repairs. Production telemetry
showed installs with a configured notification destination that had never
delivered anything and had no way to find out, because the channel that would
carry such a warning is the one that had failed.
Highlights
- Patrol guides operators through one ranked decision at a time; Actions now
has its own primary review workspace. - Read-only observers extend Patrol coverage between full model investigations
without granting mutation authority. - Approved actions gain agent preflight and stable refusal telemetry; large
installations gain compressed APIs and indexed lookups.
Added
- Durable Patrol objectives with pause, archive, resource-scope, coverage, and
observer-health state. - Model-authored observer proposals that are validated, installed as bounded
read-only checks, and kept separate from action authority. - Verified Patrol work receipts and clearer navigation between findings,
objectives, attention items, and governed actions. - A persistent decision workspace with explicit decision origin, review
progress, next-item navigation, and automatic advancement after a decision. - A canonical platform-admission facet projected on unified resources so
supported onboarding and runtime modes stay consistent across the product. - Unified Agent preflight contracts for package updates, package-cache cleanup,
and Docker lifecycle or update operations. - Production security deployment guidance and a focused security-review packet.
Improved
- Patrol now works from durable outcomes, scoped investigations, and verified
work receipts instead of treating the chat stream as operational state. - Patrol investigations preserve objective and resource intent across retries,
provider interruptions, chat restarts, truncated responses, and retained
objective runs. - Finding identity, evidence, causal conclusions, and remediation proposals are
canonicalized and validated before Patrol writes or acts on them. - Autonomous execution remains bounded by advertised capabilities, explicit
policy, agent preflight, current target state, and post-action verification. - Patrol separates its Inbox, Protection, and Activity modes, removes duplicate
counters and generic state, and keeps Assistant focused on explaining the
selected item rather than acting as the operations front door. - Authenticated startup no longer pulls the full legacy state payload. Unified
resource pages load concurrently after the first page and no longer wait for
WebSocket hydration before beginning the canonical fetch. - Actions and Patrol identify whether a decision originated from a finding,
alert, objective, or explicit operator request, making review context clearer. - Action refusal telemetry now classifies target changes, prerequisites,
contract failures, capability limits, policy decisions, and stale plans
instead of collapsing the new agent reason codes into the catch-all bucket. - API gzip handling preserves informational and bodyless responses, while
polling, metric lookup, registry resolution, and source-target mapping avoid
repeated large-estate scans. - OpenRouter, Ollama, and subscription-backed Patrol routes handle reasoning
limits, readiness checks, deadlines, and continuation latency more reliably. - Subscription-backed turns now complete their idle timeout promptly even when
a canceled CLI descendant still holds an inherited output pipe open.
Fixed
- Patrol no longer accepts empty, blocked, contradicted, incoherent, unscoped,
or unsupported findings and proposals as successful investigation output. - Enabling full AI mode or restarting the chat provider now preserves and
rewires Patrol controls and investigation dependencies. - Platform-admission state stays current across tenant changes, reconnects,
shell navigation, and canonical resource aggregation. - Agent Doctor names the judged credential in verdict evidence, and agents
warn when the server overrides a presented agent identity. - Alert threshold overrides resolve through canonical registry identity, and
the monitoring-policy menu no longer clips inside alert cards. - Docker health-check dependencies and app-container scope now remain attached
to the correct canonical findings. - Stale ZFS alerts clear when storage loses its pool attachment, and node-local
ZFS pools are no longer attached to shared storage records. - vSphere backup status, agent thermal history, explicit cluster-member address
overrides, and discovery-analysis request timeouts now reflect their actual
runtime state. - The notifications surface now states when alert delivery is paused, and that
a passing test send does not prove live alerts are getting through. Test
sends bypass the delivery pause, so a configured destination could look
healthy while every real alert was dropped before it reached the queue. - Degraded notification delivery is reported on the alerts overview rather than
only on the destinations configuration tab, and Pulse raises an alert about
its own delivery when it fails, so a dead notification channel is visible
where a channel failure cannot announce itself. - The alert flapping cooldown now suppresses for its configured duration. The
cooldown was recorded and never read, so suppression ended as soon as the
measurement window drained and a resource oscillating just under the
threshold was never damped at all.
Security
- Patrol observers are read-only and proposal-only. Mutating operations still
require the normal action-policy, approval, capability, preflight, and
verification path. - Agent preflight responses expose bounded machine reason codes rather than
command output, paths, package names, or provider-specific error text. - Investigation tooling is projected from the selected resource scope and
rejects tools that were not advertised for that run.
Upgrade Notes
Use the normal v6 install or update flow for v6.3.0-rc.3 only when you are
comfortable testing a release candidate. The rollback target is v6.2.1.
The exact rollback reinstall command is:
./scripts/install.sh --version v6.2.1The changes since v6.3.0-rc.2 do not require a Pulse Mobile client change
and preserve the existing mobile, Relay, onboarding, and mobile-facing API
contracts. No companion mobile build upload or public mobile-store rollout is
part of this candidate.
Windows Unified Agent binaries in this prerelease retain exact-SHA, checksum,
and detached-signature verification but are not Authenticode-signed, so Windows
may display an Unknown Publisher warning. Stable v6.3.0 still requires the
normal SignPath Authenticode lane unless a separate version-bound owner decision
is recorded.
Paid Pulse Pro, Relay, and eligible legacy customers should continue to use the
private download page and private runtime image for paid runtime features.
Installation
Docker (recommended):
docker pull rcourtman/pulse:6.3.0-rc.3Docker Compose:
Update your docker-compose.yml to use rcourtman/pulse:6.3.0-rc.3
See the Installation Guide for complete setup instructions.
Review the Code signing policy for release provenance, approval roles, and signing scope.
Paid Pulse Pro, Relay, and eligible legacy customers: public GitHub release assets and the public rcourtman/pulse Docker image are community builds. They do not include the private Pulse Pro runtime hooks. Use https://pulserelay.pro/download.html with your activation key to get the private Pulse Pro Docker image or Linux/LXC archive.
Promotion Metadata
- Promotion channel: rc
- Candidate stable tag: v6.3.0-rc.3
- Promoted prerelease tag: n/a
- Rollback target: v6.2.1
- Rollback command:
./scripts/install.sh --version v6.2.1 - Hotfix exception: false
- Windows Authenticode required: false
- Unsigned Windows exception: false
Pulse v6.3.0-rc.2
✅ Release Asset Validation: PASSED
All release assets have been validated successfully!
Status: Ready for publication ✅
Validated: 2026-08-18 13:09:21 UTC
Workflow: Pulse Release Pipeline #352
Validation Summary
- All required assets present ✓
- Checksums verified ✓
- Version strings correct ✓
- Binary architectures validated ✓
Pulse v6.3.0-rc.2 Release Notes
v6.3.0-rc.2 is a release candidate for the next Pulse v6 minor release and
the first published candidate in the 6.3 line. An earlier rc.1 publication
did not complete, so this cut uses a fresh immutable version for the advanced
main branch. It follows stable
v6.2.1 and introduces the Patrol v2 operating model, a first-class Actions
workspace, guarded agent action preflight, large-estate response improvements,
and monitoring correctness fixes.
Highlights
- Patrol guides operators through one ranked decision at a time; Actions now
has its own primary review workspace. - Read-only observers extend Patrol coverage between full model investigations
without granting mutation authority. - Approved actions gain agent preflight and stable refusal telemetry; large
installations gain compressed APIs and indexed lookups.
Added
- Durable Patrol objectives with pause, archive, resource-scope, coverage, and
observer-health state. - Model-authored observer proposals that are validated, installed as bounded
read-only checks, and kept separate from action authority. - Verified Patrol work receipts and clearer navigation between findings,
objectives, attention items, and governed actions. - A persistent decision workspace with explicit decision origin, review
progress, next-item navigation, and automatic advancement after a decision. - A canonical platform-admission facet projected on unified resources so
supported onboarding and runtime modes stay consistent across the product. - Unified Agent preflight contracts for package updates, package-cache cleanup,
and Docker lifecycle or update operations. - Production security deployment guidance and a focused security-review packet.
Improved
- Patrol now works from durable outcomes, scoped investigations, and verified
work receipts instead of treating the chat stream as operational state. - Patrol investigations preserve objective and resource intent across retries,
provider interruptions, chat restarts, truncated responses, and retained
objective runs. - Finding identity, evidence, causal conclusions, and remediation proposals are
canonicalized and validated before Patrol writes or acts on them. - Autonomous execution remains bounded by advertised capabilities, explicit
policy, agent preflight, current target state, and post-action verification. - Patrol separates its Inbox, Protection, and Activity modes, removes duplicate
counters and generic state, and keeps Assistant focused on explaining the
selected item rather than acting as the operations front door. - Authenticated startup no longer pulls the full legacy state payload. Unified
resource pages load concurrently after the first page and no longer wait for
WebSocket hydration before beginning the canonical fetch. - Actions and Patrol identify whether a decision originated from a finding,
alert, objective, or explicit operator request, making review context clearer. - Action refusal telemetry now classifies target changes, prerequisites,
contract failures, capability limits, policy decisions, and stale plans
instead of collapsing the new agent reason codes into the catch-all bucket. - API gzip handling preserves informational and bodyless responses, while
polling, metric lookup, registry resolution, and source-target mapping avoid
repeated large-estate scans. - OpenRouter, Ollama, and subscription-backed Patrol routes handle reasoning
limits, readiness checks, deadlines, and continuation latency more reliably. - Subscription-backed turns now complete their idle timeout promptly even when
a canceled CLI descendant still holds an inherited output pipe open.
Fixed
- Patrol no longer accepts empty, blocked, contradicted, incoherent, unscoped,
or unsupported findings and proposals as successful investigation output. - Enabling full AI mode or restarting the chat provider now preserves and
rewires Patrol controls and investigation dependencies. - Platform-admission state stays current across tenant changes, reconnects,
shell navigation, and canonical resource aggregation. - Agent Doctor names the judged credential in verdict evidence, and agents
warn when the server overrides a presented agent identity. - Alert threshold overrides resolve through canonical registry identity, and
the monitoring-policy menu no longer clips inside alert cards. - Docker health-check dependencies and app-container scope now remain attached
to the correct canonical findings. - Stale ZFS alerts clear when storage loses its pool attachment, and node-local
ZFS pools are no longer attached to shared storage records. - vSphere backup status, agent thermal history, explicit cluster-member address
overrides, and discovery-analysis request timeouts now reflect their actual
runtime state.
Security
- Patrol observers are read-only and proposal-only. Mutating operations still
require the normal action-policy, approval, capability, preflight, and
verification path. - Agent preflight responses expose bounded machine reason codes rather than
command output, paths, package names, or provider-specific error text. - Investigation tooling is projected from the selected resource scope and
rejects tools that were not advertised for that run.
Upgrade Notes
Use the normal v6 install or update flow for v6.3.0-rc.2 only when you are
comfortable testing a release candidate. The rollback target is v6.2.1.
The exact rollback reinstall command is:
./scripts/install.sh --version v6.2.1The changes since v6.3.0-rc.1 do not require a Pulse Mobile client change
and preserve the existing mobile, Relay, onboarding, and mobile-facing API
contracts. No companion mobile build upload or public mobile-store rollout is
part of this candidate.
Windows Unified Agent binaries in this prerelease retain exact-SHA, checksum,
and detached-signature verification but are not Authenticode-signed, so Windows
may display an Unknown Publisher warning. Stable v6.3.0 still requires the
normal SignPath Authenticode lane unless a separate version-bound owner decision
is recorded.
Paid Pulse Pro, Relay, and eligible legacy customers should continue to use the
private download page and private runtime image for paid runtime features.
Installation
Docker (recommended):
docker pull rcourtman/pulse:6.3.0-rc.2Docker Compose:
Update your docker-compose.yml to use rcourtman/pulse:6.3.0-rc.2
See the Installation Guide for complete setup instructions.
Review the Code signing policy for release provenance, approval roles, and signing scope.
Paid Pulse Pro, Relay, and eligible legacy customers: public GitHub release assets and the public rcourtman/pulse Docker image are community builds. They do not include the private Pulse Pro runtime hooks. Use https://pulserelay.pro/download.html with your activation key to get the private Pulse Pro Docker image or Linux/LXC archive.
Promotion Metadata
- Promotion channel: rc
- Candidate stable tag: v6.3.0-rc.2
- Promoted prerelease tag: n/a
- Rollback target: v6.2.1
- Rollback command:
./scripts/install.sh --version v6.2.1 - Hotfix exception: false
- Windows Authenticode required: false
- Unsigned Windows exception: false
helm-chart-6.3.0-rc.3
Helm chart for deploying the Pulse hub and optional Docker, Kubernetes, or OpenShift monitoring agent.
helm-chart-6.3.0-rc.2
Helm chart for deploying the Pulse hub and optional Docker, Kubernetes, or OpenShift monitoring agent.