# Spark OpenClaw + Nemotron 3 Super

We will use ollama to download and install nemotron 3 super and openclaw. At the end you will be able to message openclaw directly on the terminal, the control UI, or via whatsapp.

Step 0 is to ensure docker got the right permissions.

sudo usermod -aG docker $USER
newgrp docker

Then, we install the latest Ollama (verified up to 0.20.7)

#ollama
curl -fsSL https://ollama.com/install.sh | sh


#for quick testing the model
ollama pull nemotron-3-super

#load and run 
ollama run nemotron-3-super --verbose

ollama launch openclaw --model nemotron-3-super


Then, here are a few update (4/6/2026)

#run this if gateway is not approved (4/6/2026 bug)
openclaw devices approve

#also update the .json to avoid timeout issues.
#https://www.reddit.com/r/LocalLLaMA/comments/1scdcqe/openclaw_llm_timeout_solved/

add timeoutseconds to .openclaw/openclaw.json
{
  "agents": {
    "defaults": {
      "timeoutSeconds": 300,
      "llm": {
        "idleTimeoutSeconds": 300
      }
    }
  }
}
#type this to show all the substeps when running with openclaw tui
/verbose full 

Once it is ready, you can enter a prompt on the terminal to interact with openclaw, or you can click on the URL (e.g., http://localhost:18789/ ...) to open the control UI, and you can start messaging the Nemotron 3 Super models, and see all the other features.


Onboarding new user to setup channels like whatsapp

openclaw configure --section channels

Follow the instruction by selecting Whatsapp, and get the QR code ready on terminal.

Then, on your phone, go to ["Linked Device"](https://faq.whatsapp.com/1317564962315842/?cms_platform=android) and scan the QR. Then wait.

At the end message 'Yourself' (yes see below) in Whatsapp app to talk to the openclaw.


Note: Make sure you remove any existing openclaw configs before starting the above scripts or it can conflict (which made me confused why whatsapp did not reply)

## Setup a cron job and newsletter

One quick way to setup openclaw is to ask it to give you news from RSS feed, and present you back with a summary and links. You simply speak to it and it will handle the complexity for you.

The ideal prompt for triggering this will be:

read rss news from techcrunch about nvidia gtc

Then, to create a cron job

please repeat the news every 12 hours


After that you can check your 'cron job' tab in the WebUI, and where you will be able to see the new jobs. (See screenshot below)

## Some demo tricks
1. Use ```/verbose full``` to see mid-results and improve responsiveness.

Before:
After:

2. Try demos that's complex up front, and share the results to the audience.


For example, ou can try a research example, and save up the results to show to the audience.


And here is the actual output file (txt):

OpenClaw Demo Showcase in Korea - Research Report
Date: 2026-04-06

Overview:
OpenClaw is an open-source personal AI assistant framework that has gained traction in South Korea. The community is active in Seoul and other major cities, with events and collaborations highlighting its use cases.

Community Activities:
1. OpenClaw Builder Meetup in Seoul
   - Date: Saturday, 14 March 2026, 12:00 UTC
   - Location: 사이오닉에이아이 지하 1층, 강남구 남부순환로359길 29, Seoul
   - Hosted by: Instruct.KR
   - Details: This meetup focused on building with OpenClaw, showcasing demos and networking.
   - Source: CreateWith event page, Luma page.

2. AI Builders - Clawdbot? Openclaw.. Why now ?
   - Date: 2026년 2월 28일 (토) AM 10:00 (Feb 28, 2026)
   - Hosted by: Allan A. via Meetup (AI Builders Seoul)
   - Details: Deep dive into OpenClaw as an AI agent framework, comparing with alternatives.
   - Source: Meetup event page.

3. OpenClaw Korea Community
   - Organization: OpenClaw Korea (@OpenClaw-Korea) on GitHub
   - Location: Seoul, Korea
   - Website: https://github.com/OpenClaw-Korea
   - Public Repos: 7, Followers: 10 (as of search)
   - Source: GitHub organization page.

Case Studies:
4. Naver Cloud Explores OpenClaw-Based AI Agent Development with B2B Focus
   - Date: 2026-04-07T07:05:47+09:00 (April 7, 2026)
   - Source: Seoul Economic Daily (English version)
   - Summary: Naver Cloud is exploring the development of AI agents using OpenClaw for B2B applications, indicating enterprise interest.
   - Link: https://en.sedaily.com/finance/2026/04/07/naver-cloud-explores-openclaw-based-ai-agent-development

5. OpenClaw AI Agent for Ecommerce Operations | Nventory South Korea
   - Date: Feb 25, 2026
   - Source: Nventory.io blog
   - Summary: Details how OpenClaw is being used to reshape ecommerce operations in South Korea, showcasing a practical demo.
   - Link: https://nventory.io/kr/blog/openclaw-ai-agent-ecommerce-operations

Documentation & Resources:
- OpenClaw Showcase (쇼케이스) page (Korean): https://openclawdoc.org/ko-KR/start/showcase
- OpenClaw Docs (Korean): https://openclaws.io/ko/docs/start/showcase
- OpenClaw Korea Community Wiki: https://clawbot.ai/wiki/industry/openclaw-korea-community.html

Recommendations for Showcasing Demos in Korea:
1. Leverage Local Meetups: Partner with groups like Instruct.KR and AI Builders Seoul to host demo sessions at their events.
2. Showcase Real-World Use Cases: Highlight implementations like Nventory's ecommerce agent and Naver Cloud's B2B exploration.
3. Utilize Korean Language Resources: Provide demos and documentation in Korean to lower barrier for entry.
4. Engage with Developer Communities: Participate in GitHub repositories under OpenClaw-Korea and share code samples.
5. Target Specific Industries: Focus on ecommerce, cloud services, and automation where OpenClaw has shown traction.
6. Host Hackathons or Build Sessions: Encourage hands-on building to demonstrate ease of use and extensibility.
7. Showcase at Tech Events: Consider larger events like Developer Weeks or AI expos in Seoul, Busan, and Daegu.
8. Create Video Tutorials: Produce Korean-language video demos (YouTube, Naver TV) showing setup and customization.
9. Highlight Integration: Demonstrate how OpenClaw works with popular Korean apps (KakaoTalk, Naver, etc.) if possible.
10. Offer Localized Support: Provide Korean-speaking support channels (Discord, forums) for troubleshooting and guidance.

Conclusion:
OpenClaw has a growing presence in Korea with active community events, enterprise interest, and practical use cases. The best way to showcase demos is through localized, hands-on events that highlight real-world applications and provide Korean language resources.

Or ask about github issues and research painpoints regulary etc...

OpenClaw Developer Pain Points Analysis
Date: 2026-04-06

Overview:
This document summarizes the developer pain points identified through analysis of OpenClaw's GitHub repositories, issues, and community discussions. The pain points are categorized by severity and frequency based on community feedback.

Top Developer Pain Points:

1. Configuration and Authentication Issues (Blocker)
   - Single Point of Failure: Single openclaw.json config file - one bad key kills the gateway and all communication
   - Config/auth/state drift: "It works in CLI but not in Telegram" - Very high frequency (~12+)
   - Gateway token auth failures: Most-Googled error: WS 1008 - Very high frequency
   - Custom LLM provider auth: Ollama/LM Studio need dummy tokens - High frequency
   - OAuth token expiration and refresh complexity

2. Installation and Setup Problems (Blocker)
   - Installation failures: Median install time: 3.2 hours - Very high frequency (~9+)
   - Docker setup broken: Defaults don't work out of the box - Very high frequency
   - Dependency conflicts: Particularly with node-gyp and native modules
   - Platform-specific setup issues (Windows, Linux, macOS variations)

3. Memory and Data Management (Blocker)
   - Memory compaction data loss: `memoryFlush` OFF by default - Very high frequency
   - Workspace .openclaw/extensions/ auto-loads and executes arbitrary code from cloned repositories (Issue #11031)
   - Long-Term Memory & Knowledge Management challenges (Issue #50096)
   - Session state persistence across restarts

4. Channel Plugin Issues (Blocker)
   - WhatsApp bans & linking: Baileys violates Meta ToS - Very high frequency
   - Telegram bot silent: "Online but receives no messages" - Very high frequency
   - Channel plugins disabled: Regression in v2026.2.22 - Very high frequency
   - Platform-specific channel issues (Feishu, Slack, Discord quirks)

5. Browser Control Problems (Blocker - Ubuntu specific)
   - Browser control fails: Snap Chromium AppArmor blocks CDP - Very high frequency on Ubuntu
   - Headless browser setup complexity
   - Cross-browser compatibility issues

6. Performance and Reliability Issues
   - Slow Ollama qwen3:14b prompt ingestion in long-context runs
   - Gateway startup slowness and restart loops
   - Memory leaks in long-running sessions
   - CPU usage spikes during intensive operations

7. Developer Experience Friction
   - Documentation fragmentation and gaps
   - Complex debugging due to distributed architecture
   - Limited local development tooling
   - Inconsistent error messaging and logging
   - Plugin development complexity and documentation

8. Security Concerns
   - Exposed gateways without proper hardening
   - API key leakage in logs and error messages
   - Authentication bypass vulnerabilities
   - Path traversal risks in file operations

9. Integration Challenges
   - Custom LLM provider integration complexity
   - Third-party service API changes breaking integrations
   - Webhook setup and verification difficulties
   - Proxy and firewall configuration complexity

10. Community and Contribution Barriers
    - High barrier to entry for new contributors
    - Complex PR review process
    - Limited guidance on contributing to core vs plugins
    - Inconsistent coding standards enforcement

Specific GitHub Issues Highlighting Pain Points:
- Issue #28561: Exec Secret Provider ignores timeoutMs and lacks flexible schema validation
- Issue #11499: OpenClaw Model Retry Implementation problems
- Issue #25007: Single openclaw.json config file is a single point of failure
- Issue #11031: Workspace .openclaw/extensions/ auto-loads and executes arbitrary code
- Issue #50096: Long-Term Memory & Knowledge Management challenges
- Issue #32392: Native GitHub-as-Infrastructure Agent Deployment feature request

Community Research Findings:
The BlueBirdBack/openclaw-pain-points repository provides a community-driven analysis ranking pain points by frequency × severity. Key findings include:
- Config/auth/state drift and installation failures are the top blockers
- Memory compaction data loss causes significant user frustration
- Channel-specific issues (Telegram, WhatsApp) are particularly pervasive
- Docker setup problems affect a large portion of users attempting containerized deployment

Recommended Focus Areas for Improvement:
1. Configuration System: Move away from single point of failure config, implement better validation
2. Installation Experience: Create streamlined, cross-platform installation methods
3. Memory Management: Improve memory compaction reliability and add better safeguards
4. Channel Stability: Focus on making core channels (Telegram, WhatsApp) more reliable
5. Developer Tooling: Improve local development experience, debugging, and testing
6. Documentation: Consolidate and improve documentation, especially for troubleshooting
7. Security: Implement better security practices and vulnerability scanning
8. Plugin System: Simplify plugin development and improve stability

Conclusion:
OpenClaw developers face significant challenges primarily around configuration management, installation complexity, and channel reliability. Addressing these core issues would substantially improve the developer experience and reduce friction for both new and experienced users. The community has actively identified and begun working on solutions for many of these pain points, indicating strong engagement and commitment to improvement.

Sources Analyzed:
- OpenClaw GitHub repository issues and discussions
- BlueBirdBack/openclaw-pain-points community research
- Various blog posts and community forums
- Specific GitHub issues referenced in research

Known Issues or tricks or tips:

Gemma 4 26b issues:
https://www.reddit.com/r/LocalLLaMA/comments/1sconnk/gemma_4_26b_a4b_just_doesnt_want_to_finish_the/