A while back I asked about using TW5 with AI. I was discussing both Input and Output. Now a while later I can explain a little better what I meant. For the Output side I was creating an AI model from a TW5 file. Feed in theTW File and this helps either train an existing Model.or creates a model by itself. Then you can ask questions and it will give you details without being in the TW5. Then if you needed too you could jump into the TW5 from the answer you received. On the Input Side I was envisioning it as an AI Notebook of sorts. This is where Open Claw comes in. There were several ways I envisioned using Open Claw with TW5. First was as a āFront Endā or āUIā . You have Open Claw watch your TW file, when it changes Open Claw checks it and sees if it needs to do anything. If it does then it does some action you requested and reports the details back into that TW after archiving the original if you want to back it up. Another way was that you use whatever you set up for Open Claws Input, like Slack, and the replies go into TW5. The last way I thought of was that Open Claw builds a TW5 based on a āsomethingā. And something could be anything you want it to be. Has anyone used Open Claw for anything like this yet?
Another example would be for using TW5 as an Image repository. You have a folder of Images. You ask Open Claw to scan each image in the folder, create a thumbnail of each. Also scan each image for content. Now create Individual Tiddlers for each image using the thumbnail for viewing with the Image Description based on the scan as the details. You also have it create the tags for each image as well as a link to the full sized image. This makes the TW smaller as you used thumbnails. It also changes the camera bulk name from img001.jpg to āCow in field eating grassā.
Now you Train an AI Model on the TW5 you jist created and then ask questions or take actions using Open Claw. Open Claw show me the Images with Cows. When the response comes you could then say, use the third image or use the image name itself, and use the background from Image 5. The āanswerā which is a combined image would feed back to TW with the same Thumbnail scan concept from above and it would link to both original images as well as creating a new tag of combined imagesā¦
Edit to add - Since you have a variety of interaction methods you could talk into the system and it could respond back by voice. Or you could ask for a Slideshow on Screen 3 of all the cows.
Hi Rich
I am not qualified to talk about Ai, so Iām going to, and to be honest had never heard of OpenClaw until a few days ago despite it being āan open-source AI agent framework thatās the fastest-growing project in GitHub history.ā
However I did happen to listen to this, Peter Steinberger the man who built OpenClaw, talking about his creation.
OpenClaw: The Viral AI Agent that Broke the Internet ā Peter Steinberger (itās a Lex Fridman podcast, long.)
Most of it went over my head but the idea that ātraditional Aiā has been superceded (yes, they already talk about traditional Ai) by agentic ai is possibly an opportunity for TW as methods of communicating with those agents are required. Peter Steinberger seemed to be using whatsapp/signal or any messaging app to assign tasks to his agents. You might be right that TW could fit that role.
I expect you know that there are several users on the forum working on TW/Ai interaction. I can think of @linonetwo @buggyj and @well-noted off the top of my head but Iām sre there are others if you search.
There has also been work done on MCP servers - see this thread . Interestingly Peter Steinberger seemed to think that MCP was now no longer worth pursuing.
I didnāt have a very high opinion myself about Ai until I tried out https://dr.miromind.ai/ as recommended by @tomzheng , which is really good for my purposes. (Ask it to include references in its answers as well as its āthinking referencesā if you try it. Paste responses in to a wiki with the markdown plugin installed for tidy results. Save the āshareā link)
Maybe TW will help us ride the wave if it comes THE 2028 GLOBAL INTELLIGENCE CRISIS - A Thought Exercise in Financial History, from the Future
I have considered this 
It would take a lot of server configuration, but my understanding is that OpenClaw requires this too, so possible for advanced users.
EDIT: And would open you (and your wiki!) up to a lot of the same security problems of OpenClaw. If not for those issues, I might play with it more.
As @well-noted points out, there are many voices in the community who warn that OpenClaw is inherently dangerously insecure:
The vision of OpenClaw is very attractive, allowing everything in ones IT life to be tied together. But it is pretty much equivalent to giving access to your email accounts to a random piece of software in exchange for candy. The security vulnerabilities involving prompt injection are particularly insidious, and impossible to guard against.
No doubt the situation will improve if demand drives service operators to begin to more safely support this sort of thing.
Thanks that helps a lot. I have been looking into local only no outside network for security purposes. Also looking at a Mac Mini I have to be the Brain.I have seen several people setring them up and it just seems to āworkā. They also all warn about non local pay per token models. Some people have racked up crazy bills. Reminds me of old skool cell phone pay per minute overage charges.
And the security has been a nightmare for many people.
Oh one.of the crazy examples was someone asking Open Claw, āMake me a reservation at X restraunt for Y timeā. Open Claw tried an online.reservation portal and failed. So it installed a Voice AL, called the restraunt, and made the reservation.
100% agreed about security risks it is why I was asking here and I will be doing my best to sandbox the machine. I am hoping the genie doesnāt get.out.of the bottle.
Be safe, and document what youāre doing so you can keep us up to date I have several other projects Iām working on at the moment, or I might be exploring along the same lines as you as we speak.
I definitely will do that. I wanted to do a bit more reading and research and I know I cannot start until early next week at the earliest.
I know I am going to make sure to create Credentials for Apple for Open Claw Mac. I will be also getting sime models diwnloaded and tranfered to the Mac.
I will be opening.up to the internet but not using online model due to cost constraints.
I will be trying the simple methods I have seen people.do with Open Claw and asking it to do stuff for me. Like download an Empty Tiddlywiki and put it in a folder that you check so we can communicate with each other through TW5.
I will tell it to learn about TW5
I will also try to screen capture it all.
You should have a closer look at the security info: Security - OpenClaw ā It is intended to be used on a local network. Itās possible to communicate with it over different channels, that are connected to the internet. But the open claw agents should be sandboxed.
Also see the interview with Peter Steinberger the creator of open claw ā https://youtu.be/9jgcT0Fqt7U?t=1655
Peter says: āIt counts as CVSS 10.0ā at the moment. See: https://youtu.be/9jgcT0Fqt7U?t=1717
Just some thoughts.
-m
PS: If you have time, the whole interview is very interesting.