Since watch/info findings are filtered from the UI and never shown
to users, don't include them in the patrol run status summary.
This makes the summary consistent with what users actually see.
The LLM was confusing VMIDs because they weren't included in the
context. Now the formatted context shows:
### Container: ollama (VMID 200) on minipc
This prevents the AI from referencing the wrong VMID when generating
findings and recommendations.
When the service restarts, it now checks if a patrol ran within the
last hour. If so, it skips the initial patrol to avoid wasting API
tokens during development/maintenance when the service is restarted
frequently.
The scheduled patrol runs (every 6 hours) are not affected.
100 samples was causing 326k+ input tokens which is expensive.
24 samples (hourly resolution) still provides good pattern visibility
while significantly reducing token cost.
Estimated reduction: ~75% fewer metric tokens.
When AI patrol fails due to API issues like insufficient balance, invalid
API key, or rate limiting, we now create a finding that appears in the
AI Insights tab. This makes the issue visible to users rather than hidden
in logs.
The finding includes:
- Clear description of the issue (e.g., 'Insufficient API credits')
- Recommendation for how to fix it
- Evidence showing the actual error message
When a patrol run encounters errors (e.g., LLM call failed), don't
display 'All healthy' in the summary as that's misleading - the
analysis didn't complete properly.
Now shows 'Analysis incomplete (N errors)' instead, which correctly
explains why the status badge shows red/error.
Modern LLMs have 100k+ token contexts. 100 samples over 24h gives
~15 minute resolution while adding minimal token overhead.
This lets the LLM see fine-grained patterns, short spikes, and
accurately distinguish anomalies from normal behavior.
The in-memory MetricsHistory only retains 24 hours of data, not 7 days.
Changed computeGuestMetricSamples to use trendWindow24h instead of
trendWindow7d, and reduced sample count from 24 to 12 points.
This ensures the LLM actually receives metric samples in the context,
which wasn't happening before because the 7-day query returned empty data.
Instead of relying on pre-computed trend heuristics (which can be misleading
for edge cases like step changes vs continuous growth), we now pass downsampled
raw data points to the LLM so it can interpret patterns directly.
Changes:
- Add MetricSamples field to ResourceContext
- Add DownsampleMetrics() to reduce data points for LLM consumption
- Add formatMetricSamples() to format data compactly (e.g., 'Disk: 26→26→31%')
- Add computeGuestMetricSamples() to gather 7-day sampled history
- Populate MetricSamples for VMs and containers during context build
- Add History section to formatted context output
The LLM now sees actual patterns like 'stable for 6 days then jumped' rather
than just '45.8%/day growth rate' - allowing for much more nuanced interpretation.
This approach:
- Leverages LLM's pattern recognition instead of hard-coded heuristics
- Provides 7 days of data (~24 samples) for context on normal behavior
- Uses minimal tokens due to compact formatting with deduplication
- Is more future-proof as LLMs improve
Example output:
**History (7d sampled, oldest→newest)**: Disk: 26→26→26→26→26→31%
Refs: Frigate disk usage false positive investigation
Filter out 'watch' and 'info' severity findings from the API response.
These lower-severity findings were mostly noise:
- 'watch': CPU is 35% instead of 11% (who cares)
- 'info': Stopped container exists (knew that)
Now only showing actionable findings:
- critical: Something is broken NOW
- warning: Something needs attention soon
Users prefer silence to noise.
Runbooks were a half-built feature that provided no value:
- Only 3 runbooks existed
- AI dynamic remediation already covers the same ground
- Added UI complexity without benefit
Removed:
- runbooks.go and runbooks_test.go
- Handler functions in ai_handlers.go
- Routes in router.go
- Test cases in ai_handlers_test.go
- Auto-fix call in patrol.go
Kept (dead code but harmless):
- Frontend types/API calls (will 404)
- RecordIncidentRunbook function (unused)
Less code = easier to maintain.
Updated LLM prompt with explicit guidance on what NOT to report:
- Small baseline deviations (7% vs 4% is normal variance)
- Low utilization (under 50% CPU or 60% memory is fine)
- Stopped containers that aren't autostart
- 'Elevated' metrics still well under limits
Severity guidelines made more specific:
- CRITICAL: disk >95%, service down, data loss
- WARNING: disk >85%, memory >90%, failures
- WATCH: Only for trends projected to hit critical in <7 days
- INFO: Context/observations
Key message to LLM: 'Users prefer silence to noise'
Only flag things that require operator action.
Smarter anomaly detection to reduce false positives:
**Learning Window:** 7 days → 14 days
- Captures weekly patterns (weekday vs weekend)
**Metric-Specific Thresholds:**
CPU:
- Only report if usage >70% AND >2x baseline
- Low CPU variance (5% vs 10%) is not actionable
Memory:
- Report if >80% OR (>1.5x baseline AND >60%)
- Memory is more stable, lower threshold makes sense
Disk:
- Report if >85% usage OR +15 percentage points growth
- Disk problems are critical, use absolute thresholds
Other metrics:
- Use 2x threshold as default
This dramatically reduces 'noise' anomalies while catching
actual problems that need operator attention.
More aggressive noise filtering:
1. Anomaly threshold raised from 1.5x to 2x
- 1.5x is too borderline to be actionable
- Now requires genuinely significant deviation
2. Filter out 'Ran diagnostic' and 'Executed command' fallback items
- These are generic summaries that provide no value
- Only show remediations with specific, meaningful descriptions
Goal: If something shows in AI Intelligence, it should demand attention.
Critical changes to surface only actionable insights:
1. Anomalies now require at least 50% deviation from baseline
- '1.0x baseline' values filtered out (statistically significant but not actionable)
- Must be >1.5x above OR <0.5x below baseline to report
2. Status changes filter out startup noise
- 'unknown → running' is just system starting, not a real state change
- Backups removed from main list (they have dedicated section)
3. Only show genuinely interesting changes:
- Config changes, migrations, restarts, deletions
- Things that require operator attention
This massively reduces noise while keeping high-signal alerts.
AIStatusIndicator:
- Now shows BOTH patrol findings AND baseline anomalies
- Displays even when only anomaly detection is active (no patrol)
- Badge count includes both findings + anomalies
- Tooltip provides detailed breakdown by severity
Trend Prediction (backend):
- Add TrendPrediction struct for resource exhaustion forecasting
- CalculateTrend() uses linear regression on sample history
- Predicts days until resource is full (or if declining/stable)
- Severity: critical (<7 days), warning (<30 days), info (>30 days)
- Human-readable descriptions like 'full in ~2 weeks (+0.5% per day)'
This creates a more cohesive intelligence experience where anomaly
detection works independently of the pro/patrol features, making
value visible immediately to all users.
Add /api/ai/intelligence/anomalies endpoint that compares live metrics
against learned baselines to surface deviations - all deterministic
(no LLM required).
Backend:
- Add AnomalyReport struct with severity classification
- Add CheckResourceAnomalies method to baseline store
- Add HandleGetAnomalies API handler
- Add GetStateProvider getter to AI service
Frontend:
- Add AnomalyReport and AnomaliesResponse types
- Add getAnomalies API function
- Add AnomalySeverity type
This is the first step toward surfacing deterministic intelligence
directly in the UI without requiring LLM interaction.
- Create Intelligence struct that aggregates all AI subsystems
- Add /api/ai/intelligence endpoint for system-wide and per-resource insights
- Wire Intelligence into PatrolService as a facade (not replacement)
- Add TypeScript types and API client for frontend
- Add unit tests for Intelligence orchestrator
- Fix pre-existing test failures using diagnostic commands instead of actionable ones
The Intelligence orchestrator provides:
- System-wide health scoring (A-F grades)
- Aggregated findings, predictions, correlations
- Per-resource context generation for AI prompts
- Learning progress tracking
This unifies access to AI subsystems without replacing existing code paths.
Backend:
- Enhanced buildEnrichedResourceContext to ALWAYS show learned baselines with
status indicators (normal/elevated/anomaly) instead of only when anomalous
- This makes Pulse Pro's 'moat' visible - users can see the AI understands
their infrastructure's normal behavior patterns
- Added baseline import to service.go
Frontend (user changes):
- Added incident event type filtering with toggle buttons
- Added resource incident panel to view all incidents for a resource
- Added timeline expand/collapse functionality in alert history
- Added incident note saving with proper incidentId tracking
- Added startedAt parameter for proper incident timeline loading
Multiple frontend components were using - as a fallback
when guest.id was falsy. This format drops the node component, which is
critical for clustered setups where the same VMID can exist on different
nodes.
Changes:
- GuestDrawer.tsx: Updated guestId() and handleAskAI() to use canonical format
- GuestRow.tsx: Updated buildGuestId() to use canonical format
- Dashboard.tsx: Updated handleGuestRowClick() and guest rendering loop,
also fixed legacy metadata fallback to use consistent keying
- ThresholdsTable.tsx: Updated guestsGroupedByNode() to use canonical format
Backend changes:
- Removed temporary debug logging added during investigation
- Added alert history section to AI buildEnrichedResourceContext() function
The backend generates VM/Container IDs in instance:node:vmid format (e.g.,
delly:delly:101) via makeGuestID(). This format is now consistently used
across all frontend fallbacks to prevent AI context, metadata, overrides,
and metrics from colliding or desyncing in clustered environments.
- Fixed normalizeStorageDefaults to allow Trigger=0
- Fixed normalizeNodeDefaults (Temperature) to allow Trigger=0
- Added comprehensive tests for all threshold normalization patterns
- Updated existing test that expected old behavior
Related to #864
- Add HandleLicenseFeatures handler that was missing from license_handlers.go
- Add /api/license/features route to router
- Update AI service and metadata provider
- Update frontend license API and components
- Fix CI build failure caused by tests referencing unimplemented method
Fixes#858
The patrol interval setting was not being properly applied due to:
1. ReconfigurePatrol() was setting the deprecated QuickCheckInterval field
instead of the preferred Interval field
2. SetConfig() was comparing raw field values instead of using GetInterval()
to compare effective intervals, causing change detection to fail
3. The API response was missing interval_ms, preventing the frontend from
displaying the correct interval
Changes:
- Update StartPatrol() and ReconfigurePatrol() to use the Interval field
- Fix SetConfig() to use GetInterval() for interval comparison
- Add IntervalMs to PatrolStatusResponse and include it in the API response
- Add CollapsibleSection component with animated expand/collapse
- Wrap all 6 resource sections (Nodes, VMs, PBS, Storage, Backups, Snapshots) with accordion UI
- Add section icons and resource counts in headers
- Add expand all / collapse all buttons for quick navigation
- Make help banner dismissible with localStorage persistence
- Add Ctrl/Cmd+F keyboard shortcut to focus search
- Add keyboard shortcut hint badge on search input
- Add icons to tab navigation for quick identification
- Improve mobile tab labels with shorter text on small screens
- Create reusable components: ThresholdBadge, ResourceCard, GlobalDefaultsRow
- Create useCollapsedSections hook with localStorage persistence
- Default less-used sections (Storage, Backups, Snapshots, PBS) to collapsed
The issue was a SolidJS reactivity problem in the Dashboard component.
When guestMetadata signal was accessed inside a For loop callback and
assigned to a plain variable, SolidJS lost reactive tracking.
Changed from:
const metadata = guestMetadata()[guestId] || ...
customUrl={metadata?.customUrl}
To:
const getMetadata = () => guestMetadata()[guestId] || ...
customUrl={getMetadata()?.customUrl}
This ensures SolidJS properly tracks the signal dependency when the
getter function is called directly in JSX props.
Backend:
- Add smart provider fallback when selected model's provider isn't configured
- Automatically switch to a model from a configured provider instead of failing
- Log warning when fallback occurs for visibility
Frontend (AISettings.tsx):
- Add helper functions to check if model's provider is configured
- Group model dropdown: configured providers first, unconfigured marked with ⚠️
- Add inline warning when selecting model from unconfigured provider
- Validate on save that model's provider is configured (or being added)
- Warn before clearing last configured provider (would disable AI)
- Warn before clearing provider that current model uses
- Add patrol interval validation (must be 0 or >= 10 minutes)
- Show red border + inline error for invalid patrol intervals 1-9
- Update patrol interval hint: '(0=off, 10+ to enable)'
These changes prevent confusing '500 Internal Server Error' and
'AI is not enabled or configured' errors when model/provider mismatch.
Fixes issue where Ollama users get 'I'm a large language model, I can't do XYZ'
responses when trying to use the AI assistant. The problem was that the
Ollama provider was not passing tool definitions to the API.
Changes:
- Add Tools field to ollamaRequest struct
- Add ollamaTool, ollamaToolFunction, ollamaToolCall structs
- Convert tools from ChatRequest to Ollama format in Chat()
- Parse tool_calls from Ollama response
- Set StopReason to 'tool_use' when model requests tool execution
- Handle tool results in multi-turn conversations
Requires Ollama v0.3.0+ and a tool-capable model (llama3.1+, mistral-nemo, etc.)
Closes: Discussion #845 comment by misterlegend
- Add integration tests for Ollama provider (17 tests against real API)
- Add unit tests for baseline, correlation, patterns, memory, knowledge, cost packages
- Add context formatter and builder tests
- Add factory tests for provider initialization
- Add Makefile targets: test-integration, test-all
- Clean up test theatre (removed struct field tests)
Integration tests require Ollama at OLLAMA_URL (default: 192.168.0.124:11434)
Run with: make test-integration
- Replace verbose info banner with streamlined layout
- Add collapsible 'Advanced Model Selection' accordion for Chat/Patrol models
- Make AI Patrol Settings section collapsible with inline summary badges
- Compact Cost Controls into single-row inline layout
- Reduce form spacing for tighter presentation
- Remove unused formHelpText import
Also includes:
- OpenAI provider fixes for max_tokens parameters
- Security setup CSRF and 401 fixes
- Minor UI tweaks
Backend fixes:
- Strip provider prefix (anthropic:, openai:, deepseek:, ollama:) in all
provider Chat methods and constructors for robust handling
- Models are now correctly parsed regardless of caller format
Frontend fixes:
- Tool cards now persist in AI chat after approval execution by adding
to streamEvents array
- Dashboard now listens for pulse:metadata-changed custom event
- AI Chat emits this event when set_resource_url tool completes
- Guest URL icons now update instantly when AI sets them
- Add setup modal that appears when enabling AI without configured provider
- Modal allows selecting provider (Anthropic, OpenAI, DeepSeek, Ollama)
- Enter API key/URL and enable AI in one smooth flow
- Reorder backend to apply API keys before enabled check
- Fix Ollama to strip 'ollama:' prefix from model names
- Simplify backend error message for unconfigured providers
1. resources/store.go: Implement sorting in Query.Execute()
- Added sortResources function with support for common fields
- Supports: name, type, status, cpu, memory, disk, last_seen
- Both ascending and descending order supported
2. ai/service.go: Implement hasAgentForTarget properly
- Now maps target to specific agent based on hostname/node
- Uses ResourceProvider lookup for container→host mapping
- Supports cluster peer routing for Proxmox clusters
- Properly handles single-agent vs multi-agent scenarios
The set_resource_url tool had an incorrect example ID format ('pve1-delly-101')
which caused the AI to save URLs with wrong IDs that didn't match the actual
guest IDs used by Pulse ('instance-VMID' format like 'delly-150').
This fix updates the tool description to clearly document the correct format,
so URLs saved by the AI will now properly appear in the dashboard.
