BetterClaw Reviews

I just created two agents up to now, but already that made me happy with my purchase!

Agent1: Connect to the webstudio.is MCP by simply importing the skill from codeberg.org/visioncraft/webstudio-builder.git. Big shout out to visioncraft.studio for this awesome work!
That is all it needs to build pages in webstudio by just explaining what you want.

Agent2: Pulls weekly data from an REST API (requests auth token with the stored credentials and use it), checks the differences to the already stored data in my database (in NoCodeBackend connected via MCP) and update the data and status/change log and then send me a summary email. It took me maybe an hour to set it up while with other tools (ottokit, pabbly, ...) I spend days and was not able to even fill the database with the initial data because of various constraints/errors..

As LLM I connected my Kilo.ai key and it works nicely.

This tool has a lot of potential, and I’m currently rolling it out across my team. However, we keep running into this error:

Agent disconnected during chat response — the agent process may have crashed

For a team that relies on agents every day, reliability isn’t optional: when agents crash mid-response, we lose our workflow and our trust in the tool.

I’m giving it 3 tacos because I understand this is likely a temporary issue and I believe in the product. But if it isn’t fixed soon, we’d have to consider going back to self-managed setups like OpenClaw. I hope you can prioritize this, because with stable agents, this tool would be outstanding.

I've never had to use an AI API key before, but i was having a really hard time setting this up to get it to work.

I wanted to use a free/low cost API key as I tested this out, so i figured google gemini. I encountered error after error, and was even having the gemini web app attempt to help me in another tab. The chat also didn't work well with the free model through openrouter (eventually gemini 3.5 worked alright).

In my process of trying to get gemini to work through openrouter, imagine my surprise when i saw that my short chat messages just testing whether things were working were eating away at my small credit balance on Openrouter (and were calling claude 4.5 through Amazon Bedrock and pi).

I can fully recognize that perhaps someone well-versed in AI api keys and doesn't mind paying up for their LLMs might have had an easier time than me setting up, but this was not a simple ready to build stuff in 90 seconds for me.
I spent nearly 100 credits trying to make sure that everything was going through the LLM that I wanted through chat, but also a lot of the chats would fail for one reason or another.

Eventually, I finally, finally was ready to build something and my agent told me it built a feature/automation/crm thing based on my prompt but it wasn't under apps anywhere. The agent was trying to help me find where it put the app, but it was possibly referring from an older gui design? Idk, for something that claims to be really simple, this made me feel quite dumb.

I did put a ticket in so we'll see what happens, but since everyone is claiming how easy and seamless everything seems to be I feel like I should add a bit of contrast.

To be fair, if this does work as intended, this sounds like a cool tool to have. I feel like Tier 2 (for small use-cases) and Tier 3 make the most sense, and I might upgrade to Tier 3, but I'm still not sure about it overall.

So far I am really enjoying setting up BetterClaw.

The team is fast with their support. Fixes, connectors and requests are being worked on very quickly.

My AI Agents are slowly learning all my processes, etc. It is a great first experience into AI Agents.

Thank you BetterClaw for allowing me this opportunity.

I started on Tier 3 and moved up to Tier 5 after a few tests. In one evening I built an MVP: a custom skill with Python scripts that generates sales proposals as PowerPoint files. You request it via Telegram, the agent asks for missing data, builds the deck, sends it back to the chat, and emails me a control copy with the file attached. It worked better than I expected.

Things worth knowing:

Skill uploads have a size limit (a 7 MB ZIP was rejected, 2.7 MB was fine), so keep large files compressed.
In my tests, Telegram didn’t pass files I sent to the agent, so I forward them to the agent’s Gmail instead. Sending files from the agent works using the Bot API.
Secrets must be declared in the skill’s Credentials, or the scripts won’t receive them.

No servers, no Docker, and support answers fast. Excited to see what comes next.