Pulse 5.0 Release Candidate - Call for Testing

[rcourtman]

Pulse v5.0.0-rc.2 is available for testing

This is the first release candidate for Pulse 5.0, which includes AI-powered monitoring, Kubernetes support, and a redesigned agent experience.

This is a release candidate. Please test thoroughly before using in production. Report any issues in this thread or open a separate issue.

---

Major Features in 5.0

AI-Powered Monitoring

Pulse 5.0 introduces an optional AI assistant for infrastructure monitoring:

• Natural Language Queries - Ask questions about your infrastructure in plain English
• Patrol Service - Automated health checks with configurable schedules
• Auto-Fix Capability - AI can diagnose and remediate issues automatically (opt-in)
• Predictive Intelligence - Failure pattern detection and anomaly identification
• Cost Tracking - Monitor AI API usage with budget controls

Kubernetes Support

Native Kubernetes cluster monitoring is now built-in:

• Monitor pods, deployments, services, and nodes
• Namespace filtering and resource sorting
• Column picker for customizable views
• Real-time cluster health status

Multi-Platform Unified Agent

A single agent binary now supports multiple platforms with automatic detection:

• Auto-Detection - Automatically enables Docker, Kubernetes, or Proxmox monitoring
• Multi-Host Tokens - One API token can register multiple hosts
• Linux, macOS, Windows - Cross-platform support
• Linked Host Agents - Host agents automatically link to PVE nodes

Streamlined Setup Experience

The new Setup Wizard simplifies getting started:

• Simplified 3-step onboarding (Welcome, Security, Complete)
• Guided agent installation with platform auto-detection
• Auto-generated install commands with token management

---

How to Install

Docker (recommended):

docker pull rcourtman/pulse:v5.0.0-rc.2

Docker Compose:
Update your docker-compose.yml to use rcourtman/pulse:v5.0.0-rc.2

Binary:

curl -fsSL https://github.com/rcourtman/Pulse/releases/download/v5.0.0-rc.2/install.sh | bash

---

Links

• Release Notes
• Documentation
• Live Demo (running 5.0.0-rc.2)

---

Feedback

Please test and share your experience. I am particularly interested in:

1. AI features - Does the Patrol service work as expected? Any issues with the AI assistant?
2. Kubernetes monitoring - Is cluster data accurate? Any missing information?
3. Agent installation - Does auto-detection work correctly on your hosts?
4. Setup Wizard - Is the onboarding flow intuitive?
5. Performance - Any slowdowns compared to 4.x?

Reply below with any bugs, suggestions, or just to confirm things are working.

---

Thanks for helping test Pulse 5.0.

[faustus1005]

So far, no issues! I'll let you know long term, but initially:

AI features - Working so far, for the bit I've had time to test.
Agent installation - Working perfectly, no issues on old or new hosts
Setup Wizard - Easy as can be! Love it! Upgrade was super easy no issues at all.
Performance - Not that I can tell at the moment!

[krestaino]

I'm getting the following error when trying to use DeepSeek. Let me know if you'd like me to open a separate issue.

Error: AI request failed: API error (400): Model Not Exist

[RLSinRFV]

Just installed v5.0.0-rc.2. I noticed that the agents are still reporting v5.0.0-rc.1.
Do the agent nodes need a separate update script or removed/reinstalled?

[rcourtman]

@krestaino Please open a separate issue for the DeepSeek error. Include:

1. Full browser console output when the error occurs
2. What action triggers the error (clicking Test, sending a message, or something else?)
3. Whether you see the error immediately after saving settings or only when trying to use AI

The error suggests DeepSeek is receiving an invalid model name, but I need more context to trace where it comes from.

---

@RLSinRFV Agents need to be reinstalled to get the new version. The agent binaries are separate from the main Pulse server.

To update agents:

# Stop the agent service first
sudo systemctl stop pulse-agent

# Download and install the new agent
curl -fsSL https://github.com/rcourtman/Pulse/releases/download/v5.0.0-rc.2/pulse-agent-linux-amd64 -o /tmp/pulse-agent
sudo mv /tmp/pulse-agent /usr/local/bin/pulse-agent
sudo chmod +x /usr/local/bin/pulse-agent

# Restart
sudo systemctl start pulse-agent

Alternatively, if you installed via the unified agent installer script, rerun it and it will download the latest version.

[krestaino]

5.0.0-rc.3 fixed my issue. DeepSeek is now working. I did not change anything on my end. I simply updated to 5.0.0-rc.3 and it started working.

[kwat2025]

Wow. Major new features added. Recommend you support current release (critical bug fixes only) for awhile even after release of 5.0 to allow users time to test and upgrade when they are comfortable.

[rcourtman]

Fair point, and I appreciate you raising it.

Honestly, as a solo maintainer I just don't have the bandwidth to keep two branches going properly. The v4 temperature monitoring was complicated. It worked for some setups but never reliably across all Proxmox configurations. I spent a fair bit of time trying to improve it and eventually concluded the only way forward was the v5 agent rewrite.

I won't be pulling v4 releases down, and if something genuinely critical surfaces I'll have a look. But active work is on v5 now. That's where the fixes are going.

If you're hesitant to upgrade, that's completely understandable. Might be worth giving the RC a spin in a test environment when you have a moment. The agent setup is considerably less fiddly than it used to be.

[MichelfrancisBustillos]

After upgrading to v5.0.0-rc3, I appear to be getting 'random' crashes/restarts of the container, with the log info below. When this panic occurs, the container does automatically restart, but all acknowledged alerts are reissued. Each one occurs after Broadcasting state update (ticker) message.

I am also seeing Docker 'blinking' on some agents. It goes away, and the agent disappears from the Docker tab, then it will return after a couple of minutes.

Additionally, the Single Sign-On button is missing from the login page.

This RC does appear to fix the issues in Issue #770

[wwwlicious]

Small UX issue in that models dropdown doesn't scroll on screen overflow

Also posted about OIDC not working #853

[misterlegend]

I was able to get the AI Assistant working once, but every time after it just gives me "I'm a large language model, I can't do XYZ" for every question after that.

5.0.0-rc.3
It happens with llama3.3, mistral, and gpt-oss models via ollama.

[rcourtman]

@misterlegend This is a known limitation with Ollama models and tool calling. The "I'm a large language model, I can't do XYZ" response indicates the model is ignoring the available tools and responding as if it has no capabilities.

Not all Ollama models support function/tool calling reliably. While we pass the tool definitions to Ollama, whether the model actually uses them depends on:

1. Model support - Some models have better tool-calling capabilities than others
2. Model size - Larger parameter models tend to follow tool schemas better
3. Prompt sensitivity - Some models need very explicit prompting to use tools

Workarounds:

• Try llama3.1:70b or llama3.2:latest which have better tool support
• For questions that don't require system access ("what's using the most CPU?"), the model should respond with general guidance even without tools
• Consider using a cloud provider (Anthropic, OpenAI) for more reliable tool execution

I'm looking into whether we can improve the system prompt or tool definitions to encourage more consistent tool usage with smaller Ollama models.

[misterlegend]

Thanks for the tips. llama3.1:70b is having the best success rate so far.

[RLSinRFV]

Getting this error on the host running Pulse (Container 102). I'm running v5.0.0-rc3 and am using the new unified agent.

CT 102 - Pull file
TASK ERROR: failed to open /opt/pulse/bin/pulse-sensor-proxy-linux-amd64: No such file or directory

[rcourtman]

Thanks for reporting this.

Root cause: The install.sh script bundled with v5.0.0-rc.3 was attempting to install the legacy pulse-sensor-proxy, which is deprecated in v5+. The new unified agent (pulse-agent) now handles temperature monitoring, making the sensor proxy unnecessary.

Fix: This has been addressed in main:

1. Skips the sensor proxy installation entirely for v5+
2. Removes the confusing "Temperature Monitoring Setup" prompt
3. Updates the completion message to not show outdated sensor proxy instructions

Workaround for v5.0.0-rc.3:
If you need to install before the next release, you can bypass the sensor proxy by passing --proxy no:

curl -fsSL https://github.com/rcourtman/Pulse/releases/download/v5.0.0-rc.3/install.sh | bash -s -- --proxy no

The fix will be included in the next release candidate (rc.4). Thanks for testing.

[clar2242]

Not getting the login via cidc button anymore using 5.0.0-rc.3
Was working on 4.36.2

I see in /api/security/status
"oidcEnabled": true

[clar2242]

Also X-Forwarded-For and X-Forwarded-Proto set by Caddy are not being honoured.
/api/security/status reports the docker internal IP for client IP, and false for hasHTTPS.

Debug log does see this though, so is getting through.
{"level":"debug","component":"pulse","proxied":true,"secure":true,"cf_ray":"","cf_connecting_ip":"","x_forwarded_for":"192.168.0.69","x_forwarded_proto":"https","time":"2025-12-17T20:38:37Z","message":"Proxy/tunnel detected - adjusting cookie settings"}

[RLSinRFV]

Just FYI @rcourtman, been testing v5.0.0-rc3 pretty thoroughly the last two days, and while my Proxmox environment isn't very complex (2 PVE hosts, no cluster, 2 PBS LXCx, various other VMs/LXCs, 3 Docker environments with agents), everything seems to be working very nicely.
The AI Assistant is DOPE! It uncovered a few things I needed to remedy.

Just saying, Pulse is maturing very quickly - Well done!
May be time for another XDA review...

[Lorondos]

Hmm I ran the bash script directly in my Proxmox LXC container:

It shutdown the LXC after that, rebooted and looks like it did update but now it is reporting the version still as 4.36.2 though I can see the new features, I will try the RC option from settings and will try to resolve.

Edit: Yup, cool, that fixed the versioning, can see the correct version number and RC label now.

[Lorondos]

Feedback:

Looks like everthing is working so far, two things I have noticed:

1. Maybe this is intended based on the ordering logic but a bit of wasted space where it has done the CRLF for the next set of filters.
2. The trueNAS capacity issue is still present, not in this branch? (Reporting 44TB where in reality, it is 28.94TB) - Pulse Host Agent over-reports usable storage on ZFS (TrueNAS) due to dataset AVAIL summation #718

[Lorondos]

I will attempt to update the agent on TrueNAS, still reporting the old v4 one.

[Lorondos]

Success:

[krestaino]

What happened to the drawers that opened when clicking on a resource? Every table row click now selects it for the new AI tool, even if the AI tool is off. Is there a way to get the drawers back, while also using the new AI features?

[rcourtman]

Hey @krestaino, thanks for the feedback.

The drawers were intentionally removed in v5 in favor of a "columns over clicks" approach. For a monitoring platform, I wanted all the information visible at a glance rather than requiring extra clicks to access details.

You can customize which columns are visible using the column toggle in the filter bar – so if you want to see disk I/O, network stats, backup status, etc., you can enable those columns. The gear/columns icon in the toolbar lets you show/hide any of them.

As for row clicking – you're right that it currently only selects resources for AI context. If AI is disabled in your setup, clicking rows won't do anything. I could look into removing the clickable appearance (cursor) when AI isn't enabled to make this clearer. Would that help?

Let me know if there's specific information you were accessing via the drawer that isn't available as a column option.

[RLSinRFV]

@rcourtman remind me again how to update the agents (from v3 to v4 in this case). I can REMOVE them from the Settings -> Agents -> Managed Agents UI, run the Uninstall agent script, then add them again, but wondering if there's a simpler update script I can run.
Tkx in advance.

[MDE186]

Minor UI,

[MDE186]

@rcourtman the run diagnostic button does not fit the windows.

Attached new view.

[darthrater78]

I cannot remove an agent from Pulse in RC4. I have to remove the API token associated first.

[rcourtman]

Addressing recent feedback:

@clar2242 - OIDC button missing + X-Forwarded headers

OIDC button: Fixed in main. The SSO button now appears correctly when OIDC is enabled.

hasHTTPS reporting false: Just pushed a fix (commit fdb6789) that respects the X-Forwarded-Proto header in /api/security/status.

clientIP showing Docker internal IP: For security, X-Forwarded-For is only trusted from configured proxy IPs. Add to your Pulse environment:

PULSE_TRUSTED_PROXY_CIDRS=172.17.0.0/16

(Adjust to match your Docker/reverse proxy network)

---

@RLSinRFV - How to update agents

The unified installer handles upgrades automatically. Just rerun the install command:

curl -fsSL https://your-pulse-url/install.sh | sudo bash -s -- --url YOUR_PULSE_URL --token YOUR_TOKEN

No need to uninstall first - it will detect and upgrade the existing agent.

---

@MDE186 - Minor UI

Could you describe what the UI issue is? The screenshot shows a mobile view but I am not sure what needs fixing. More details would help!

---

@darthrater78 - Cannot remove agent without removing API token first

This should not be required - the agent removal should work independently of the API token. Could you share:

1. The exact error message you see (if any)
2. Which type of agent (Host/Docker/Kubernetes)
3. Browser console output when clicking Remove

This will help me track down the issue.

---

@wwwlicious - Models dropdown overflow

The model dropdown not scrolling is a known browser limitation with native select elements. Looking into replacing it with a custom searchable dropdown component in a future update.

---

@MichelfrancisBustillos - Crashes + Docker blinking

These sound concerning! For the crashes, could you open a separate issue with:

1. Full stack trace from the panic
2. Approximate container count/Proxmox setup size

For the Docker blinking, this might be related to WebSocket connection stability. Are you seeing any network errors in the browser console?

[MichelfrancisBustillos]

It looks like after bringing all the agents up to the latest version, the crashing stopped. I'm assuming the issue was using 4.x agents on a 5.0 installation.
OIDC is also now working as it should on rc4. Thanks!

[rcourtman]

@Lorondos - Wasted space in filter layout

Thanks for the screenshot! I see what you mean about the gap where the filter chips wrap to the next line. This is a flexbox layout behavior where items wrap when they exceed the container width.

Ill look into tightening up the spacing or using a more compact layout for the filter bar. Could you let me know your screen width when you see this? That will help me reproduce it.

Regarding the TrueNAS capacity issue (#718) - yes, that fix requires updating the agent on TrueNAS to v5. I see you already did that successfully. The storage calculation should now use zpool list for accurate pool capacity instead of summing dataset AVAIL values. Let me know if youre still seeing incorrect values with the v5 agent!

[Lorondos]

Ah, awesome, thanks for that! I figure its due to a 4K display (3840 x 2160). Might be a quick fix here in the tailwind config:

  screens: {
    '3xl': '1920px', 
    '4k': '2560px',   
  },

Hmm, still seeing 44TB oddly

[darthrater78]

@rcourtman I didn't realize I was still running the docker agents everywhere, that way have a been a contributing factor. I removed them and installed/reinstalled the binary agent and tested and it worked as expected.

Other issues though.

I access Pulse via reverse proxy, and it ends up using the proxy name for the agent instead of the IP. Pulse communication should always be direct, can you adjust the logic to always show (or have an option to) generate the token from the real IP:port?

Second, having the issue again where docker isn't being detected on the console.

[rcourtman]

@darthrater78 Thanks for reporting these issues with screenshots! Let me address both:

1. Reverse Proxy URL Issue

You're right - the install command currently uses the browser's URL (window.location.origin), which will be your reverse proxy hostname. We'll add a PULSE_AGENT_URL environment variable so you can explicitly set the direct internal URL for agents to use.

Workaround for now: Manually edit the install command to use your internal IP:port instead:

curl -fsSL http://192.168.x.x:7655/install.sh | sudo bash -s -- --url http://192.168.x.x:7655 --token <token>

2. Docker Not Detected

The installer does auto-detect Docker by running docker info. Looking at your screenshots, both hosts show TYPE = "Host" only (not "Host, Docker"), which means the Docker module isn't active.

Could you check a few things on docker-main and docker-services?

1. Does docker info work as root?

   sudo docker info

2. What does the agent service show?

   sudo systemctl status pulse-agent
   sudo journalctl -u pulse-agent --no-pager -n 50

   Look for any Docker-related errors or "enable-docker" in the ExecStart line.

3. Was Docker installed AFTER the agent? If you installed the agent before Docker was set up, it would have skipped Docker detection.

Quick fix: Re-run the installer with explicit Docker flag:

curl -fsSL https://pulse.example.net/install.sh | sudo bash -s -- --url http://<internal-ip>:7655 --token <token> --enable-docker

This will reinstall and force-enable Docker monitoring regardless of auto-detection.

[darthrater78]

sudo docker info - no problem

[darthrater78]

" Use Allow re-enroll in Settings -> Agents -> Removed Docker Hosts or rerun the installer with a docker:manage token to clear this block"

I have no area to allow a "re-enroll" except for Kubernetes

[rcourtman]

@darthrater78 Found and fixed the bug! 🎉

Root cause: The frontend WebSocket store was missing the handler for removedDockerHosts data. Even though the backend was sending this data correctly, the frontend was completely ignoring it - so the "Removed Docker Hosts" section never appeared.

The fix (commit 97c851c):

• Added removedDockerHosts to the WebSocket store initial state and message handler
• Added missing state broadcast after "Allow re-enroll" succeeds

Once you update to the latest version:

1. Go to Settings → Agents
2. The "Removed Docker Hosts" section should now appear with your blocked host
3. Click "Allow re-enroll" to unblock the agent
4. Your Docker agent should then be able to report again

Thanks for reporting this - it was a genuine bug affecting all Docker re-enrollment scenarios!

[darthrater78]

@rcourtman when do you expect the new build to be available that has this commit?

[jnbolsen]

First, thanks for all the awesome work you have done! Pulse has been amazing so far.

I am running Pulse 5.0.0-rc.4 via systemd in an LXC. When adding an agent to a docker host, I noticed the following warning every 30 seconds. This is with an API token with Container lifecycle management and Container agent reporting scopes only.

Dec 22 03:37:18 ezarr pulse-agent[6175]: {"level":"warn","component":"host-agent","error":"pulse responded with status 403 Forbidden","time":"2025-12-22T03:37:18Z","message":"Failed to send report, buffering"}

However, docker monitoring and reporting seems to work fine.

Is this something to worry about?

[rcourtman]

I have pushed a fix to improve this behavior.

The issue is that Pulse Unified Agent enables the Host Agent module by default, but your API token lacks the Host agent reporting scope (since you only enabled Container scopes). The agent was receiving 403 Forbidden and buffering the reports, causing the log spam.

With the fix:

1. The agent will now check for 403 Forbidden errors.
2. It will log a clear error message explaining that the token scopes are missing or that PULSE_ENABLE_HOST=false should be set if host monitoring is not intended.
3. It will stop buffering these failed reports, eliminating the noise.

In the meantime, you can silence this by setting PULSE_ENABLE_HOST=false in your agent environment, or by adding the Host agent reporting scope to your token.

[jnbolsen]

Perfect, thanks! Both adding the PULSE_ENABLE_HOST=false variable to the docker host agent or adding the Host agent reporting scope to the API token stops the buffering. The only reason I thought to not add the Host agent reporting scope in the first place was that this particular hosts resources are already being reported via PVE (VM).