Founder and CEO @Petabridge @getsdkbin @textforgeapp OSS: @AkkaDotNet and https://xitter.cf/t.co/i2Pc96fVg4 Catch me on YouTube: https://xitter.cf/t.co/gUNvkaNuAS
Houston, TX
Joined May 2008
- Tweets30.4K
- Following893
- Followers8.5K
- Likes18.4K
new release of Memorizer out courtesy of @dariogriffo - fixes an annoying error with older Ollama embedding APIs
I have a great name for this: ClawdBot
So @AkkaDotNET v1.6 is already an order-of-magnitude improvement for networking and serialization performance over v1.5, but the next beta of it is about to be even faster:
Looking at about a ~20% throughput improvement across the board and significant reductions in CPU / mem resources per message.
All from simple mechanical fixes plus some experimental IActorRef compression on the receiving side of the wire.
ALT Akka.NET v1.6.0-beta4 vs 1.6.0-beta3 throughput numbers
ALT Akka.NET v1.6.0-beta3 vs. v1.6.0-beta4 CPU per message
And join us and the other 1000+ Akka .NET developers active in our Discord if you have questions: discord.gg/GSCfPwhbWP
The problems this architecture will solve for agent-using organizations:
1. Fine-grained / scoped / short-lived credentials built per-task rather than credentials-per-box
2. Deterministic agent pipelines: plan / build / verify + review -> goes back to plan if things are bad. This will resemble something more like a GitHub Actions pipeline than skill files.
3. File storage and growth problems - all agent workspace is ephemeral and on-demand, with agent-assigned source control identities + output artifacts being the source of truth.
4. ACTUAL COLLABORATION BETWEEN MULTIPLE PEOPLE IN THE ORGANIZATION ON A SINGLE AGENT WORKFLOW - everything everyone else has been designing is really single-player work. This gets back into the type of collaborative mode that GitHub was originally built for.
5. Measurable - if your reviewer (human or AI) regularly sends back code that was authored by a combination of Opus 5.5 / Sonnet 5.5, you can start to measure per-model fault rates when it comes to code authoring. If someone in your organization is merging pull requests generated by this system at a rate higher than that, congratulations you now have a way of finding slop cannons. This will vary by size and complexity of task.
The next move up from ssh + tmux + agents on dev boxes is going to be self-hosted cloud-ish environments using agents in K8s with fortified containers and everyone is going to be mad about it
For real though
Kafka consumers must coordinate in-flight work, offset commits, and partition revocation. Akka.Streams.Kafka layers stream backpressure and committable offsets over Confluent.Kafka so those concerns stay in one pipeline.
petabridge.com/blog/akka-str…
Doh, was trying to get this to show in the tweet petabridge.com/blog/akka-str…
Aaron Stannard retweeted
When multi-agent terminals for managing multiple TUI agents, like tmux, started gaining popularity, there wasn't a Windows version available.
So I built one just for myself.
Eventually, I went through the painful process of removing WPF and porting everything to Avalonia, and now I've even released a macOS version!
English-speaking users probably won't notice any issues, but getting multilingual IME input windows to work properly on macOS was quite a headache. 😅
Other than that, everything works great!
Under the hood, each terminal is controlled as an actor using Akka.NET's actor model.
An ADE (Agent Development Environment) powered by Akka.NET, running on both Windows and macOS.
link : psmon.github.io/AgentZeroLit…
Been away on vacation this week and it's amazing what local models like Qwen3.8-Flash can do. Had it create a VHS tape -> turn it into an animated GIF showing off some of the new independent scrolling / performance improvements in Termia 0.17.0 while I was out sightseeing
Bear in mind this is not me sitting in front of a computer actively monitoring stuff - this is me periodically checking Slack notifications hours apart.
Cheap local tokens + easy to invoke and collect results anywhere is a game changer
I made Netclaw for doing this and we should have a big release out this week with much improved sub-agent support and lots of other improvements aimed at allowing long, autonomous sessions to execute safely netclaw.dev/
