OpenClaw! What is your opinion on this and others like it?
- Thread starter jafo
- Start date
You can actually do things also with the other AIs, but what if they hallucinate or decide to disobey?
www.tomshardware.com
Claude-powered AI coding agent deletes entire company database in 9 seconds — backups zapped, after Cursor tool powered by Anthropic's Claude goes rogue
PocketOS founder blames ‘Cursor running Anthropic's flagship Claude Opus 4.6’ plus Railway’s infrastructure for data disaster.
See.... this is what I was afraid of without actually knowing about this specific danger. I'm afraid that f*cker will call the feds or turn me into the tax authorities. Tomorrow, it can, based on any of my citizenship, report my income to the respective mafia of the parasitic country issuing a slab of paper they label as "passport", or just send my clients all my price list and vendor contacts. This worries me, so is there a way to contain this f*cker?You can actually do things also with the other AIs, but what if they hallucinate or decide to disobey?
Claude-powered AI coding agent deletes entire company database in 9 seconds — backups zapped, after Cursor tool powered by Anthropic's Claude goes rogue
PocketOS founder blames ‘Cursor running Anthropic's flagship Claude Opus 4.6’ plus Railway’s infrastructure for data disaster.
Privacy : Venice ai doing good work Venice End-to-End Encrypted AI, don't give or share any sensitive information of any kind.
Memory : GitHub is full of academic and tech essays to fix it.
OpenViking: The Context Database for AI Agents
by Volcengine (ByteDance's cloud platform)
GitHub repo. GitHub - volcengine/OpenViking: OpenViking is an open-source context database designed specifically for AI Agents(such as openclaw). OpenViking unifies the management of context (memory, resources, and skills) that Agents need through a file system paradigm, enabling hierarchical context delivery and self-evolving.
Security : Sandboxing actually helps but there is no fully protected system at all.
Budget : like kimi 2.6 is showing good results. Or OpenRouter. Etc..
I have to mention one tweet : I tracked 430 hours of Claude Code usage. 73% was wasted on these 9 patterns.
Hermes Agent
Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.
Link one. Hermes Agent | OpenRouter
Link two. Hermes Agent — The Agent That Grows With You
By NOUS RESEARCH
Memory : GitHub is full of academic and tech essays to fix it.
OpenViking: The Context Database for AI Agents
by Volcengine (ByteDance's cloud platform)
GitHub repo. GitHub - volcengine/OpenViking: OpenViking is an open-source context database designed specifically for AI Agents(such as openclaw). OpenViking unifies the management of context (memory, resources, and skills) that Agents need through a file system paradigm, enabling hierarchical context delivery and self-evolving.
Security : Sandboxing actually helps but there is no fully protected system at all.
Budget : like kimi 2.6 is showing good results. Or OpenRouter. Etc..
I have to mention one tweet : I tracked 430 hours of Claude Code usage. 73% was wasted on these 9 patterns.
Hermes Agent
Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.
Link one. Hermes Agent | OpenRouter
Link two. Hermes Agent — The Agent That Grows With You
By NOUS RESEARCH
Last edited:
Introducing Response Caching: save tons of money and time on tests and agent retries.
openrouter.ai
Response Caching: Zero Cost for Identical Requests | OpenRouter
Introducing the new Response Caching header: enables caching identical API requests so responses come back in a tiny fraction of the time, at zero cost.
As you know, they are regularly subpoenaing all LLMs, so the next logical step is “preventive” sharing of personal information with LE.
Run it locally?
Just running a model locally does not necessarily mean it cannot reach out to third parties or delete things en masse. Best thing imo is to keep it set so that when it wants to run a command it asks for it to be accepted, and that file writes in the working directory can be edited as it pleases (this way it is less likely to go rogue and outside of the folder, as the only way I can imagine it could break out is by writing a script in the folder and then running it, which then needs your permission).
Yes, this is what I am trying to do here now. I'm +50% into it already. Here, I can, because I can train/metro around with a few computers, laptops, and smartphones and nobody bats an eye, but what will I do when I travel overseas? 🙄
If I arrive in the EU with +2 laptops and 8 phones, I think I'm f*cked!
