[Discussion] Does this node need separate install of OpenClawbot to function on backend?

[AKDigitalAgency]

It's not clear to me, for this node to work on frontend, do I need to install 1st OpenClawBot separately to run it on backend?
Or is it installed inside your node as part of package?

[rookiestar28]

The short answer is: No, you do NOT need to install OpenClawBot separately. Everything is included right inside this custom node package.

To clarify how it works right now:

1. Core Package: When you install this repository in your custom_nodes folder, it automatically sets up the frontend UI (sidebar/settings) and the backend APIs directly within your ComfyUI server. The LLM client, security layers, and scheduler run natively inside ComfyUI.
2. The Chat Connector: If you want to use the ChatBot features (like Telegram, Discord, or LINE), the connector script is already bundled inside this exact same folder.
3. How to run the bot: You don't need to download another project. You simply open a terminal inside this node's folder and run python -m connector alongside ComfyUI.

So, you only need to clone this one repository to get both the ComfyUI nodes and the ChatBot functionalities working.

[AKDigitalAgency]

1st thanks for this node package & explanation!

2nd When you say it uses API on backend does it mean it uses external models (ie Claude etc) instead of local?
Can I set it up to only use local models?

3rd Is there another of your packages or someone else you trust with high security enabled functions, stand alone package that I can install?

Alll my LLMs I run through LM Studio & I want to use OpenClawBot with LM Studio as backend.
I run it on Win 11 Enterprise.

[rookiestar28]

1st thanks for this node package & explanation!

Thanks! Really happy to hear it’s useful.

2nd When you say it uses API on backend does it mean it uses external models (ie Claude etc) instead of local?

Not necessarily. “Backend API” means OpenClaw sends requests to an HTTP endpoint (OpenAI-compatible or provider-native, depending on provider).
That endpoint can be:

• Cloud (OpenAI / Anthropic / etc.), OR
• Local (LM Studio / Ollama / vLLM / llama.cpp server) running on your own machine.

So API-based ≠ external-by-default.

Can I set it up to only use local models?

Yes.
For LM Studio with OpenClaw, set:

• OPENCLAW_LLM_PROVIDER=lmstudio
• OPENCLAW_LLM_BASE_URL=http://127.0.0.1:/v1
• OPENCLAW_LLM_MODEL=<exact_model_id>

Then inference stays local to your machine.

Security note:
“Local-only” should also include runtime/network controls.
In this project:

• local provider lmstudio does not require an API key by default
• external tools are opt-in (OPENCLAW_ENABLE_EXTERNAL_TOOLS)
• fallback candidates are explicitly configurable (OPENCLAW_FALLBACK_MODELS, OPENCLAW_FALLBACK_PROVIDERS)
• Strongest guarantee is still OS-level outbound restrictions (firewall), so unexpected egress is blocked even if misconfigured.

3rd Is there another of your packages or someone else you trust with high security enabled functions, stand alone package that I can install?

In security terms, safest outcomes come from boundary controls more than package branding:

• Keep model endpoint local (127.0.0.1)
• Deny outbound network by default (except loopback)
• Run non-admin
• Keep write access scoped to required folders only
• Keep external tools disabled unless required (OPENCLAW_ENABLE_EXTERNAL_TOOLS=false)
• If tools are enabled, use strict allowlists and bounded execution
• Keep logs/audit available for incident review

All my LLMs I run through LM Studio & I want to use OpenClawBot with LM Studio as backend. I run it on Win 11 Enterprise.**

That’s a good fit. Suggested “secure-by-default” setup on Win11:

A) LM Studio

1. Start LM Studio local server
2. Bind to localhost only:
   • Host: 127.0.0.1
   • Port: (commonly 1234)
3. Confirm endpoint is reachable:
   • /v1/models
   • /v1/chat/completions)

B) OpenClaw config (actual keys used by this project)

• OPENCLAW_LLM_PROVIDER=lmstudio

• OPENCLAW_LLM_BASE_URL=http://127.0.0.1:1234/v1

• OPENCLAW_LLM_MODEL=<exact model id from /v1/models>

• OPENCLAW_ENABLE_EXTERNAL_TOOLS=false

• Optional hardening

  • OPENCLAW_ADMIN_TOKEN=
  • leave fallback vars empty unless intentionally used:
    • OPENCLAW_FALLBACK_MODELS
    • OPENCLAW_FALLBACK_PROVIDERS

C) Hardening (Win 11 Enterprise)

• Windows Firewall outbound rules:

  • block outbound for the ComfyUI/OpenClaw host process(es)
  • allow loopback (127.0.0.1) only

• Run as non-admin

• Restrict workspace folder permissions

• If tools must be enabled:

  • allowlist commands only
  • keep filesystem scope tight
  • keep network blocked unless explicitly required

If useful, I can also provide a copy-paste .env example for your exact LM Studio port/model and whether you want OPENCLAW_ADMIN_TOKEN enabled.

[AKDigitalAgency]

Thanks for this detailed explanation!
Yes, Please provide an exact minimal config snippet tailored to LM Studio!

Just to clarify from your 1st post, do I use command: "You simply open a terminal inside this node's folder and run python -m connector alongside ComfyUI." to activate OCB or just to activate chat ability using Telegram?

Right now, working directly in Comfy canvas I don't need telegram chat function.

[rookiestar28]

Thanks for this detailed explanation! Yes, Please provide an exact minimal config snippet tailored to LM Studio!

Just to clarify from your 1st post, do I use command: "You simply open a terminal inside this node's folder and run python -m connector alongside ComfyUI." to activate OCB or just to activate chat ability using Telegram?

Right now, working directly in Comfy canvas I don't need telegram chat function.

Yes, your understanding is correct.

python -m connector is for the optional Connector process (Telegram/Discord/LINE/WhatsApp/WeChat/Kakao chat control).
It is not required to use OpenClaw directly inside the ComfyUI canvas.

If you only use OpenClaw in Comfy canvas (no Telegram chat), just run ComfyUI normally with these env vars:

Minimal LM Studio config (canvas-only)

OPENCLAW_LLM_PROVIDER=lmstudio
OPENCLAW_LLM_BASE_URL=http://127.0.0.1:1234/v1
OPENCLAW_LLM_MODEL=<your_lmstudio_model_id>
OPENCLAW_ENABLE_EXTERNAL_TOOLS=false

Optional but recommended for stricter admin boundary:

OPENCLAW_ADMIN_TOKEN=<strong_random_token>

Notes:

• Replace <your_lmstudio_model_id> with the exact model ID from LM Studio /v1/models.
• LM Studio provider in this project does not require an API key by default.
• Do not run python -m connector unless you want remote chat-platform control.

[AKDigitalAgency]

Remind me in which file in OCB do I add LMS snippet ie config.json?

I seen too many YT videos about OCB and my brain overloaded with info now!

[AKDigitalAgency]

1. Is OCB full version personal assistant or limited in function to Comfy image generation & workflow fixing?
2. Can it move files ie custom-nodes from broken Comfy-old instance to working Comfy-new instance?
3. Try re-importing FAILED custom-nodes & move them to Comfy-new instance?
4. Can it create custom nodes on the fly, just describe in prompt what I want to happen? Similiar to what "Any Node" custom node did using ChatGPT.
5. Where is OCB files located ie MEMORY, SOUL, etc can't find it your node folders?
6. Can I run it as Personal assistant for example "create annual report for 2025" & point it to folder with data files?
7. If using cloud based API free model like GLM 4.7 etc, instead of using local model through LMS, what do I use as WebUI this http://127.0.0.1:1234/v1 address?
8. What is built-in security scan command? Is it --doctor ?
9. OCB server automatically launches in Comfy but can I manually start it by going into your custom-node folder and running CMD --launch command & use http://127.0.0.1:1234/v1 as WebUI to chat with it if I'm using API cloud model only?