Its really fantastic. I can't imagine why you'd go through the effort using Claude Code with other models when pi is a much better harness. There's tons of extensions already available, and its trivial to prompt an LLM to create your any new extension you want. Lacking creativity and want something from another harness?
> Run <other harness> in tmux and interrogate it how feature X works, then build me the equivalent as a pi extension.
Maybe in a few years there will be obvious patterns with harnesses having built really optimal flows, but right now it works so much better to experiment and try new approaches and prompts and flows, and pi is the easiest one to tweak and make it your own.
If you ever want to use other models, pi can do that. In the middle of a session I might switch from gpt-5.2 to opus and get it to do something or review something and then switch back to gpt. Since models are being released every few weeks this is interesting to compare models without having to switch to a different harness.
And if there’s any feature codex has that you want, just have pi run codex in a tmux session and interrogate it how said feature works, and recreate it in pi.
Interesting point. And from the agents point of view, it’s always joining at the last minute, and doesn’t stick around longer than its context window. There’s a lesson in there maybe…
Unifi makes a doorbell and consumer (and commerical) security cameras which run and store data on a local device, but still reachable online with their app connecting directly to your device. I used their dream machine pro with a big HDD, but they're released a few other devices in the last few years which might be cheaper and use SSDs. And I think you could run the stack in docker. But if you want to hack it yourself, there's probably easier projects. If you want to spend a bit more but have everything more or less just work with nice hardware and apps, Ubiquity's Unifi system is really great for home security. Not to mention the wifi and other networking solutions they have.
Honestly very confused by the people happy or agreeing with Anthropic here. You can use their API on a pay-per-use basis, or (as I interpreted the agreement) you can prepay as a subscription and use their service with hourly & weekly session limits.
What's changed is that I thought I was subscribing to use their API services, claude code as a service. They are now pushing it more as using only their specific CLI tool.
As a user, I am surprised, because why should it matter to them whether I open my terminal and start up using `claude code`, `opencode`, `pi`, or any other local client I want to send bits to their server.
Now, having done some work with other clients, I can kind of see the point of this change (to play devils' advocate): their subscription limits likely assume aggregate usage among all users doing X amount of coding, which when used with their own cli tool for coding works especially well with client side and service caching and tool-calls log filtering— something 3rd party clients also do to varying effectivness.
So I can imagine a reason why they might make this change, but again, I thought I was subscribing to a prepaid account where I can use their service within certain session limits, and I see no reason why the cli tool on my laptop would matter then.
Seems like a great day for Pulsar / StreamNative and Redpanda who are going to get a lot of new customers in the coming years.
When looking for alternatives to Kafka these are the most promising options I found, not counting RabbitMQ which needs no introduction.
Pulsar seems to be 'kafka done better'. The version of Kafka that Confluent used internally (Kora) seems closer to Pular as well. Pulsar has a lot of features that Kafka doesn't have, like per-message acknowledgement, similar to RabbitMQ. And it has protocol support for RabbitMQ and Kafka, so can be a drop-in replacement.
Redpanda seems like a great re-implementation of Kafka.
I'm hoping this boosts Pulsar's status and helps get some traction for StreamNative. It seems like the best technical solution for events and messaging. It just needs a bit more market adoption to make an easier choice for enterprise, in my opinion. This might be that moment.
This and Materialize seemed like great tools. I met some of the team of Rising Wave at the Kafka conference last year in London and was impressed by their work. It may be great if you need such a tool.
In the end, I went with ClickHouse and it's materialized views feature. It might not be quite as powerful as what these other tools are doing, but it works for us, and it's really easy to set up. Before we were using Timescale's continuous aggregates, which had good performance, but require some domain knowledge to setup. ClickHouse materialized views are great because you don't need to be an expert to use them. And even so, performance is still very good.
Bluetooth and carplay do indeed work as you need to be connected to the vehicle. I used those features too. The ConnectedDrive feature discussed here is when you install the BMW app on your phone and register the VIN number of the car to your personal BMW account, verified by tapping some buttons in the vehicle while linking, and "owning" the car in your BMW app. This gives access to things like remote location tracking, starting the car from anywhere to get the airco etc working.
n.b. Doing this in a rental car probably violates some of the terms and conditions one would have to agree to when linking the car, like "I promise this is _my_ car and/or I have permission from the owner to link it to my personal BMW account"...
I rented a BMW from Sixt in the USA earlier this year. I wanted to use the ConnectedDrive features, but it was blocked by BMW because the vehicle VIN was (correctly) registered as a Fleet Vehicle (i.e. a rental car) and thus none of those features were allowed with that car.
I have rented BMWs in the Netherlands and don't recall being able to use these features either.
Thus you seem to have encountered a situation which BMW and Sixt know about and have procedures in place to prevent, but their Italian subsidiary seems to have missed it with a certain batch of fleet vehicles, or just this specific one. I'd report it Sixt and move on.
Amazon Q. Claude Code is great (the best imho, what everything else measures against right now), and Amazon Q seems almost as good and for the first week I've been using it I'm still on the free tier.
The flat pricing of Claude Code seems tempting, but it's probably still cheaper for me to go with usage pricing. I feel like loading my Anthropic account with the minimum of $5 each time would last me 2-3 days depending on usage. Some days it wouldn't last even a day.
I'll probably give Open AI's Codex a try soon, and also circle back to Aider after not using it for a few months.
I don't know if I misundersand something with Cursor or Copilot. It seems so much easier to use Claude Code than Cursor, as Claude Code has many more tools for figuring things out. Cursor also required me to add files to the context, which I thought it should 'figure out' on its own.
I've been down this path, and if my experience is more common, then it really boils down to the classic "Nobody gets fired for buying IBM", and here IBM -> Confluent.
StreamNative seems like an excellent team, and I hope they succeed. But as another comment has written, something (puslar) being better (than kafka) has to either be adopted from the start, or be a big enough improvement to change— and as difficult and feature-poor that Kafka is, it still gets the job done.
I can rant longer about this topic but Pulsar _should_ be more popular, but unfortunately Confluent has dominated here and rent-seeking this field into the ground.
I'm biased because I recently introduced ClickHouse at my company, but everything I've seen so far makes me think analytical and observability use cases like this "just work" in ClickHouse.
Just like Postgres became the default choice for operational/relational workloads, I think ClickHouse is (or should) quickly become the standard for analytical workloads. In both cases, they both "just work". Postgres even has columnar storage extensions, but I still think ClickHouse is a better choice if you don't need transactions.
A rule of thumb I think devs should follow would be: use Postgres for operational cases, and ClickHouse for analytical ones. That should cover most scenarios well, at least until you encounter something unique enough to justify deeper research.
> Run <other harness> in tmux and interrogate it how feature X works, then build me the equivalent as a pi extension.
Maybe in a few years there will be obvious patterns with harnesses having built really optimal flows, but right now it works so much better to experiment and try new approaches and prompts and flows, and pi is the easiest one to tweak and make it your own.