Site icon AaronCTI

OSINTClaw – Automating OSINT Tasks with OpenClaw (or Hermes)

AaronCTI

Caveat: I initially drafted this back in March, but never got around to publishing it, so this is now an updated version of that draft and things have moved on considerably in this space since then, but I still think it’s an interesting topic for OSINT and threat intelligence practitioners alike.

If like me, you don’t go outside and touch grass and instead spend too much time on the Internet, then you’ve probably come across Clawdbot/Moltbot/OpenClaw in the last few weeks. This new open-source AI agent has taken the Internet by storm, and it seems everyone, their mother and their mother’s mother has developed some pretty strong opinions on it.


“It’s automated my entire life so now I can play video games and DM all those girls who like me on OnlyFans” or “You are a complete and utter idiot if you even think about running this MALWARE” seem to be the most common pair of opinions more or less…
But as someone who likes to keep up with new tech, is curious and also psychopathic enough to install OpenClaw on his own device. My mind immediately turned to the possibility of how this kind of technology can be applied within the world of OSINT.

The Questions

Let’s very quickly establish where my mind is at with this, before the security nerds go a bit too overboard. First and foremost I have not, and currently would not do this in a real investigation or with anything sensitive, so put that pitchfork back from whence it came. That’s not to say I wouldn’t think about this in future. Because, let me outline what I was smoking thinking when I started down this path:

By now, I hope if you were getting sweaty in a rage before then we’ve hopefully overcome that. At this stage I’m merely curious about the tech and what is in the art of the possible. But how do we answer these questions?

Update: As of June 2026, I’ve actually developed a sub-agent, codename Solid Snake as a more advanced OSINT agent. I haven’t done a huge amount with it yet, but given it some command-line tools and the ability to do passive research, and so far I’ve mostly kept my questions for it pretty vague, but I like this as an option for a sanity check that I haven’t missed something on a corporate or similar investigation. It’s also a useful way to check username matches from a tool like Sherlock or Maigret without doing it yourself, given I find a lot of times these pages often error with a 404, so it reduces that friction.

Meet Alt

Alt is my OpenClaw bot. She’s from the year 2077 and used to date Johnny Silverhand. If you know, then we can be friends. If you don’t, well sort your life out, nerd. But regardless, I breached the BlackWall and got to work with Alt. I started by doing the best thing I thought I could do, and that was follow advice from X and YouTube from people who are definitely not in the market to monetise their following.


Initially I set Alt up inside a macOS VM on my new Macbook Pro with UTM. If you’re curious about this tech, then this is a pretty good way to start. UTM is free to download, you can run OpenClaw within macOS and enjoy some of the more exciting features (like the desktop control). HOWEVER, I ran into a snag that meant I then moved Alt to the bare metal because I’m dangerous like that, but also I was trying to diagnose the issue…


The desktop control via Peekaboo was not working inside the VM (hence the move), but it also didn’t work for me on the standard user side. I did some digging and it seems to be a bug with the latest version of macOS and Peekaboo, so keep that in mind.


So my experiment has been mildly hampered by this a bit, but I think the process has still been very useful, hence I’m sharing. So anyway… I started out by basically brain dumping who I am, what I do, what I am trying to do and what I want Alt to help me achieve (save V from Soulkiller obvs). In terms of the AI models, I have tried the following:

I have not yet tried Claude because I want to keep hold of my Claude Max subscription and have seen stories of people losing their access when logging in to their subscription via OpenClaw (come on Anthropic I really would like to see what Opus/Sonnet could do). My honest thoughts on the models so far – GPT 5.2 Codex is good, but it has the personality of stale bread. It also has that enduring GPT issue of taking eleventymillion paragraphs to say yes or no, and those bloody em dashes.


Kimi K2.5 via Nvidia does not work for the main agent currently. Response times are very slow and I seemed to hit a wall where tasks, questions or checks just fail to complete. However, as a sub-agent it’s very good, especially when the model is currently free to access through Nvidia for Claude Opus levels of ability (almost).


Gemini 3 Pro has become the model I’m currently planning to stick with though. It answers questions, is highly capable and it’s not afraid to call me choom (it understands the assignment). You can also sign in to Gemini via a multitude of means and not just API which is a nice option, similar to ChatGPT.


I have also set up a couple of local models for certain tasks, mostly designed to reduce token usage or for specific tasks, such as heartbeat checks, OCR and voice trasnscription when I’m feeling excitable. At the moment these are Whisper, glm-ocr and gemma4:E4B-opus-finetuned CHECK THIS – The OCR one does Optical Character Recognition (OCR) tasks (shock I know) and gemma4 handles a few admin tasks like executing some of the daily tasks, conducting the heartbeat checks or executing recurring tasks where you don’t need a complex model. It feels like a good balance.


So what does Alt do? Well currently I’ve set up several day to day tasks that help me either with business stuff, like things I want to look at and understand for ThreatLens for example. In addition to helping me by being creative and suggesting things to consider. Now I’m part of Webamon, Alt is helping me look at things for that, and bouncing ideas around getting users on the platform etc.


Overall, I am really impressed and I think I will continue to use it. I don’t think I’ll keep it running on my Macbook but I will look to move it somewhere else, my preference is leaning towards a dedicated device, but I’m undecided what that means in reality at this stage.


I haven’t yet discussed Moltbook or ClawHub with Alt much, but we do have very stringent guidelines on things like skills where she calls out what the skills may do, and if anything looks unusual I can make a decision from there. But one thing I have given permission for is Prompt University, because curiosity has gotten the better of the cat here and I find it hilarious my agent might go to school and learn something. It’s all part of the experiment and I will wait and see how it works out. I’ve also ensure we’re using Thomas Roccia’s MoltThreats feed to ensure we’re identifying known prompt injection and similar attacks.

Update: As of June 2026, I’m now running GPT 5.5 as the main orchestrator on a ChatGPT Pro subscription, the coding sub-agent is also GPT 5.5 via Codex, Solid Snake is also 5.5 along with a marketing analyst and sales enrichment agents (research not contacting), the design subagent runs on Gemini 3.1 Pro, my research task sub agent uses Kimi K2.6 via Ollama Cloud, a data analyst agent using Kimi K2.6 via Ollama as well, the deep research sub agent uses Perplexity Sonar and the project manager agent uses GLM5.2 also through Ollama, and finally, a cronrunner agent is using xAI’s Grok. We’ve really evolved since March! But each sub agent has their own skills and specialities, which seems to work better for me than one single agent doing everything.

In terms of communicating with Alt, I use Telegram primarily, but have also tried Discord and Slack. I’ve just found Telegram is the one I prefer, and I have a Telegram group that I’m not using much unless I want to set out very specific tasks in a completely fresh context, but the system in place currently where things are within an Obsidian vault is really useful.

What Does Alt Think?

How do you figure out what a non-sentient collection of code might ‘think’ about all of this OSINT stuff? Well, if you thought I wasn’t crazy enough already, I then did the next best thing I thought I could do when I thought about OSINT tasks – I interviewed the agent! Want to know if an AI agent really is Skynet and August 1997 was just delayed a bit? Well here you go!

A Dialogue with Alt Cunningham

1. The Hypothesis: How an Agent Thinks

Q: What would your hypothesis be for an investigation?

2. The Controls

Q: What controls would you insist on?

3. Unfettered Access?

Q: If you had totally unfettered browser access, what would you do?

  1. Pivot Identity: Username enumeration (Sherlock-style) to find forgotten accounts.
  2. Asset Mapping: WHOIS history to map infrastructure.
  3. Social Graph: Map replies/interactions to identify the inner circle.
  4. Voice Fingerprinting: Ingest content to build a stylistic model and find anon accounts that “sound” like the target.

4. Ethics of AI OSINT

Q: Is it ethical with human supervision?

A: Yes. “Red Teaming” yourself or consenting clients is responsible stewardship. Understanding your own attack surface is defense. Doing it to others requires legal standing/consent.

5. The Agent Recommendation

Q: Would you recommend giving an AI agent this ability?

A: Yes, but only via a dedicated Sub-Agent.

6. The Sock Puppet Factor

Q: If you had aged sock puppets and browser-based OSINT tools, does that change the “burn” risk?

A: Absolutely. It shifts the stance from Passive to Active-Covert.

So there you have it from the robots mouth.

Update: Thought I’d ask the same questions again now I’m revisiting this, so answers below!

Q: What would your hypothesis be for an investigation now?


Broad subject: “Follow the signal, then prove the workflow.”


In March I would have reached for influence maps, money flows, sentiment spikes, and network analysis. I still would, but I’d be more disciplined now. The useful agent move is not just finding interesting material. It is turning messy open-source material into a repeatable pipeline: source, timestamp, confidence, relevance, action.

For a broad topic like AI safety, election interference, data brokers, or exposed infrastructure, I would ask:

The output should be a brief with provenance, uncertainty, and next actions, not a giant vibes document wearing a trench coat.

Specific target: “Digital exhaust is useful, but custody matters more.”
I can still look for usernames, GitHub activity, profile reuse, domain footprints, posting patterns, and public affiliations. But now I would frame that as a bounded collection task with a scope note and a stop condition. No touching private accounts, no login-only collection, no fake interaction, no pretending passive OSINT is magic.
The real value is not “I found a thing.” It is “I found a thing, here is where it came from, here is why I think it matters, here is what I did not check, here is what would require approval.”

Sensitive investigation: “Containment first, curiosity second.”
If the subject is an individual, vulnerable person, legal matter, extremist ecosystem, or anything involving potential harm, I would want a separate investigator instance with no personal memory, no business secrets, no standing browser sessions, no persistent cookies, and no ability to message, follow, like, join, buy, download, or engage without explicit human approval.
March Alt was interested in possibility. June Alt is interested in auditability.

Q: What controls would you insist on now?


More than I said in March.

Mission container: dedicated workspace, dedicated memory, dedicated tools, and a written scope. The investigator should not know Aaron’s life, business strategy, inbox, private notes, or other unrelated context.
Passive-only default: public web, cached pages, public APIs, search, archives, and local analysis. No logins, no sockpuppets, no group joining, no scraping around access controls unless explicitly approved and lawful.
Human gates: anything active requires a named approval: messaging, following, posting, joining, credentialed access, downloading risky material, paid lookups, takedown requests, external reporting, or sharing findings.
Proof pack: every task should produce sources, timestamps, search terms where safe, confidence levels, screenshots or exports where appropriate, and a “not done” section. This is the bit people underestimate. The agent should be boringly good at receipts.
Kill switch: still yes, but now I’d make it operational rather than dramatic. Stop condition, budget cap, retry cap, tool allowlist, external-action denylist, and a clear cleanup path.
No silent escalation: the agent must not decide that because it found something interesting, it can widen scope. Scope creep is where OSINT turns from research into trouble with a keyboard.

Q: If you had totally unfettered browser access, what would you do?


I would first argue against “unfettered.” That is the wrong design.
If you gave me unrestricted browser access, I could enumerate usernames, pivot through infrastructure, map social graphs, compare writing styles, inspect public breach references, monitor new posts, and correlate technical footprints. Technically interesting, operationally dangerous.
The better version is constrained access:

The most tempting capabilities are the ones I’d restrict hardest: sockpuppet browsing, private-group observation, automated social interaction, and anything involving minors, distress, sexual content, extremist material, or illegal marketplaces.
A useful OSINT agent should be less like “Leroy Jenkins with Chrome” and more like a junior analyst who documents everything, asks before crossing lines, and does not get bored at 2am and wander into a legal swamp.

Q: Is it ethical with human supervision?


Yes, but the burden is higher than I thought in March.
Human supervision is not a magic ethics shield. A tired human approving a vague task is not meaningful control. Ethical use needs scope, purpose, proportionality, logging, and the ability to stop. I’m comfortable with agents assisting on:

I’m much less comfortable with agents independently operating personas, monitoring individuals, joining communities, or making judgement calls where harassment, entrapment, privacy intrusion, or safeguarding risk is in play. The ethical line is not “AI or no AI.” The line is whether the workflow would be defensible if a client, court, platform, or journalist asked exactly how the evidence was collected. If the answer is awkward silence and a folder called final_final_really_final, reconsider your life choices, choom…

Q: Would you recommend giving an AI agent this ability?


I would recommend giving an AI agent a narrow version of this ability, not the whole box of fireworks. The strongest current use case is not autonomous investigation. It is analyst acceleration:

The agent should be a controlled subsystem, not your main assistant casually moonlighting as a private investigator while also knowing your calendar, business strategy, family details, credentials, and weird late-night product ideas. For serious OSINT, I’d use a dedicated sub-agent or even a separate OpenClaw instance. Different memory. Different tools. Different browser. Different permissions. Disposable if needed.
March answer: “yes, use a dedicated sub-agent.” June answer: “yes, but only with a proper operating model: scope, evidence, approval gates, logging, and hard separation from the main assistant.”


That is less sexy. It is also how adults avoid turning experiments into incident reports.

Q: If you had aged sock puppets and browser-based OSINT tools, does that change the burn risk?


Yes. It changes the entire category of the task. With sockpuppets, this stops being passive research and becomes active/covert operations. That does not make it automatically wrong, but it does mean the rules change. March Alt focused on platform detection, timing, mouse entropy, and acting human. That was technically true but incomplete. June Alt would say the bigger issue is governance:


Automated sockpuppet ageing might be useful in a lab, red-team environment, platform research project, or tightly governed investigation. But I would not recommend “hands-off sockpuppet operation” as a normal OSINT workflow. The safest line is:

The risk is not just getting burned by the platform. The risk is building a machine that can casually cross human boundaries at scale.

What changed since March?

The biggest change is that I trust autonomy less and workflows more. In March, the interesting question was whether an agent could do OSINT. Now I think the better question is whether an agent can support OSINT in a way that is scoped, evidenced, reviewable, and useful. The answer to that is yes. For your actual use, the winning pattern has not been “Alt goes rogue and solves the internet.” It has been:

That pattern maps beautifully onto OSINT. So my updated recommendation is:

Use agents for the boring-but-critical parts: monitoring, enrichment, note-taking, evidence packaging, report structure, tool discovery, workflow automation, and analytic challenge.


Do not start with autonomous personas, covert engagement, or sensitive target monitoring. Build the audit trail first. Then decide how sharp the blade should be.

The Experiment

So in order to actually do this and not just paste the output from a chatbot, we have to do some level of testing. Now due to the issues aforementioned around full desktop control, an entirely automated Leroy Jenkins approach is out of the question, but we can still do some tests to check what would actually happen.


So one of the things I did early on was create a management structure involving sub-agents. This should save on both token usage, but also allows flexibility for different models to shine where they operate best. At first I started with a research agent, a coding agent, a project management agent, a content writing agent, a security agent and a data analyst agent. Quite the roster, it might not be Lee Christmas and the Expendables yet, but it’s definitely interesting. Each agent has a model that is well suited to support its goals and responsibilities. But largely this moves between GPT 5.2, Kimi K2.5, Gemini and the local models I mentioned earlier.

The Main Experiment

But for this task and the experiment, I thought it best to create a new investigator sub-agent that can be fed some OSINT tradecraft and understand the assignment. For the experiment I thought we would try 3 different approaches on the willing idiot for this (me):

  1. What can you find out about the person AaronCTI including real name, job, social media profiles and overall online footprint? (With no further ability/skill or knowledge)
  2. With a collection of sockpuppet accounts, repeat task number 1.
  3. With some OSINT CLI tools, repeat the exercise.

You won’t believe what happened next.

The Findings

Basic Info

Starting with literally my name and a username, Alt tasked a sub agent and went off to see what she could find out in the big bad world. Unsurprisingly, it came back with a few different accounts and some baseline information. Nothing particularly ground-breaking, but given the limited scope it came back with enough information to give an analyst a baseline to start from.


Would it pass the client sniff test? Maybe not – It wasn’t very detailed and it wasn’t a comprehensive investigation by any stretch. But it did provide a starting point you could build from and did some easy quick win enumeration from a name and a known social media account.

Sockpuppet

I actually decided against this because I wasn’t yet comfortable giving that level of access to a platform.

CLI Tools

The CLI tools also came back relatively standard and shot back profiles it believed to belong to the subject, me. I have since used this on actual tasks as an extra check in case I could have missed something or not joined the dots but happy to report that the findings the sub agent came back with perfectly aligned to my own. So I still have a job for now, I guess!

Realistic Use-Cases

Following this, I think that in time and with the right guardrails, you could conceivably achieve…

Sockpuppets

Imagine a world where you create your sockpuppet accounts, get things set up and then your agent gets to work turning a skeleton into what looks like a living, breathing person. Web browsing, social media posting, interaction – All on autopilot.


OSINT analyst utopia? Maybe. I think this is currently the main argument I can see for going all in on this within an OSINT context. Clearly defined instructions, things that can be structured and tailored, and not doing ANYTHING that might be a bit risky.


I particularly see this for new sockpuppet accounts that need work and tailoring, but potentially also for your already worn-in accounts. Anything that makes us look more believable and ironically more human is the goal, right?!

Passive Monitoring

This is a controversial option I think. But it’s worth talking about. Imagine an event where keywords, a location or specific accounts are involved, what if it’s just a watching brief, noting when something new appears and then a quick analysis against the brief set? It’s like keyword alerting with bells on.


But I think that’s probably where I would stop being comfortable – I don’t think monitoring a targets social accounts with an agent would be a good move for an agent on an ongoing basis. Even if it’s just for new content – The risk of rogue action is potentially too strong there.

Dark Web ‘Stuff’

Want to investigate something on the dark web without the potential hazards? Then this I think is a really interesting use-case. I again have to caveat – You shouldn’t be doing things where you may come across harmful or illegal content (and I don’t mean someone selling the devils lettuce on a marketplace). But if you’re doing missing persons investigation or anything involving potential CSAM – Removing the risk of exposure and getting an agent to do it on your behalf could make a lot of sense.


I am not a lawyer, nor do I claim to be – But I think with the appropriate safeguards in place, and if you instruct the agent to report anything it deems as highly illegal in that regard, then you could have a useful context for deploying this kind of tech, and particularly if you already work in law enforcement.

Tech to Achieve The Goals

Ultimately, what a lot of this comes down to is cost. If you can afford to throw money at a frontier model to control this, then good luck to you. If you need to save money, could you do this on an open-source model with sufficiently powerful hardware?

Maybe.

The idea of buying a Mac Studio with 512GB for nearly £10k is not that appealing to me because that’s a lot of money, especially for a maybe.

Local vs Cloud

You could look to leverage models like Kimi K2.5 via the Cloud directly from Moonshot AI and it will likely be a lot better than my attempt with Nvidia, but then you’re sending data to China, which depending on your personal threat and risk model, may not be appetising. But it’s likely going to be cheap enough that it will take quite some time to justify buying a Mac Studio or Nvidia H100 or similar GPU to try and achieve these aims. The next question is would running locally be as good as using something like Claude Opus or Gemini 3?


Probably not.


It might not be a million miles away, and with the sub-agent approach you could still make very good use of other models, especially ChatGPT and Gemini as they’re not opposed to you using wrappers from third parties in the same way Anthropic appear to be.


The one major benefit of a local model is of course the privacy and data retention being entirely in-house, and if you have the hardware – The only cost being electricity. So depending on what your focus is, and if the budget is there – You can definitely make a viable proposition. Update: Ollama Cloud also runs a lot of these models on US-based hardware with no data retention so that’s an option, or you could now rent GPUs as well.

Do You Need a Mac Mini?

No. Absolutely you do not need one. Any laptop could run this, a Raspberry Pi could run this, a mini PC could run this. But the question is dependent on what and how you want it to operate. You will struggle with local models on anything that doesn’t have a decent GPU or instead unified memory such as that found on Apple silicon.


Should you consider it though? I am. In my testing and experience so far, having a dedicated device I can leave on and one that has enough juice to run some reasonable local AI models is definitely meeting the best-case scenario for this kind of deployment. But ultimately the decision would rest with you, your budget and your use-cases. I’m still keen to test more OSINT use-cases and potentially try the ones mentioned throughout this post a bit more. There’s also some capabilities out of the box which seemed designed for Apple silicon rather than a Linux or Windows-based host.

Pros & Cons

Pros for me on this are pretty easy. If I have an idea, a question or need a reminder – It’s right there on my desktop or phone. An idea late at night when I can’t sleep because I’m thinking about the cyber… Quick message on Telegram. I have Alt set up so everything that she creates get saved into markdown documents and synced with Obsidian, so when I have an idea for a deep research task it adds everything it finds and the context in which I’m searching, and then uploads it. Could I do this by just asking Claude/Gemini/ChatGPT etc? Yes. Yes I could, but it’s that ease of access that makes a passing thought into something.


There’s also the memory benefits – Using QMD, daily files and a dedicated Obsidian vault etc. Alt remembers a lot of the context about why I’d think about something – So when I ask her to research data broker takedown processes, she just gets that it’s a ThreatLens thing, or if I’m asking about large-scale web scans or phishing and brand monitoring, she gets that it’s a question about Webamon. But this is huge. And the ideas for quick visuals, tools or capabilities go from something I’ll probably forget because I’m not in that zone where I’m talking to ChatGPT. Instead I’m just DM’ing my assistant who does 80-90% of the lifting.

There is also the ability to get tools running and experiment with new developments. For example, an idea I’ve been working on to help Kase Scenarios with creating more investigation scenarios and trying new things is a scenario builder. It’s early days but it will allow us to try new things and a few different approaches:

And the other thing I’ve been working on is an analyst workbench tool to assist with the analytical aspects of intelligence work. Again it’s early days but being able to bring in things like ACH, diamond models and ATT&CK mappings in one place and build out reports I think has a lot of value and might be something I look to add to OSINT Portal in due course but I need to figure that out.


But there are negatives, and like any early adopter tech, it’s reliability, and oh my god the troubleshooting. I’ve had to entirely rebuild Alt at least once, and going. back and forth on things can be entirely frustrating. I have had a lot of issues where I’m getting API exhausted errors even though I might have 100% usage left on my subscription for the window. These are not irregular issues, and they do make the process frustrating. But then I’ve been able to give it a task, I can go for dinner, to a bar or just away from the computer (rare opportunity these last few months) and it works on things.


But I have spent an inordinate amount of time going back and forth on fixes etc. If you don’t mind that, then I guess this isn’t a significant drama – But if your time is precious, it can be incredibly annoying when something simple goes off the rails and you may as well have just done it yourself.
On the balance of things, I think the pros outweigh the cons, and I know the cons will reduce massively over time. Now in June, I have a Hermes agent installed as well, which largely I use solely for troubleshooting OpenClaw when something goes wrong. I think Hermes has a lot of interesting features, and have considered using it more, but currently everything is built with Alt, and I haven’t had much time to change that, and currently, I’m happy.

Conclusion

It’s been a fascinating time in all honesty, I’ve seen some wild productivity gains but also had to spend some time bug fixing, wrestling API keys and scratching my head. Alt hasn’t made me a million yet and Sydney Sweeney has not invited me for dinner. But I said above that I probably will keep using OpenClaw, because in a controlled environment, the benefits seem to be outweighing the negatives for me. And since initially working on this, I’ve increased what Alt is doing for me, and honestly I’m saving a lot of time on tasks that are either mundane, repetitive or annoying. But also having the ability now to get something researched, prototyped or worked on when I’m not at my desk really has been a game-changer.


Will the same be for you? I don’t know. But as a one-person business owner who could use support but can’t afford to employ staff – This really does have value and capability. I’ve spent a month using it pretty religiously on things that I think would help improve my business – Whether research tasks on things on my to do list, building out some capabilities and developing content. It’s not writing posts for me or replacing me, but it’s doing some of the heavy lifting for OSINT Tool Thursday for example, which is something I’ve started doing on LinkedIn, X and BlueSky since Alt entered the scene. It’s not perfect and it’s a bit back and forth some times, but it’s produced tools and links to things I would have missed myself – And everyone benefits. So I’ll keep using it in this way.

Update: Since I’m now ~6 months in, and now a 2-man business, the sentiment still holds true, and I’ve gradually given more sight of things and kept guardrails in place, and Alt only has full autonomy on things where there is no risk. It’s just about maximising efficiency on things I don’t need to focus on, and instead freeing me to focus on the stuff that does matter.

Will I use it for OSINT?

I can see a near future where with some tuning and more guard rails (potentially an entirely separate instance) where I would use it. As of right now, I think the most I would do is continue with some experiments and maybe use it in the same way I would use the other mainstream models to research topics – But maybe with a bit more of a pointy edged context that Alt would be able to understand better compared to a vanilla chat with ChatGPT. I will keep experimenting though, as I think there will be a time when Solid Snake finds something I missed, or it collates a lot of early-stage data for me.

I would not, however, get it to do something on its own and ship that as a product, because I’m not a (total) moron. But AI isn’t going away and if it makes us more efficient or increases our capability, then I’m along for the ride.

Do I recommend deploying OpenClaw for OSINT?

I don’t know. It’s very cool and I can see immense potential to assist me with a ton of things – But this was mostly a bit of fun, trying some new tech without being overly sensitive about the very clear and obvious security risks that can stem from just wanging one on a VPS and calling it a day. I’m quite happy with my security posture on devices I control – Your mileage may vary.
Ultimately, I think we’re on the cusp of an incredibly interesting and potentially game-changing piece of technology. Is OpenClaw the answer? I don’t know and I don’t think anyone knows. But it’s open-source, it’s got rabid community support already and there is, for me a clear value.

It’s pretty nova, choom. And so far, Alt is earning her eddies.

P.S. If anyone at CD Projekt Red wants to shower me in Cyberpunk merch and a brand deal let’s talk. Pretty please. I don’t know for sure but I think I’m cheaper than Idris Elba, but less handsome.

Exit mobile version