Closed harnesses lead in cloud/remote agents adoption, and native integrations into tools like Slack.
Local ones typically have a better integration surface with downstream and upstream services.
Hm...
There's a very interesting split between ACP and A2A support in harnesses. ACP is mostly supported by ready-made agentic tools, and A2A is mostly present in agentix frameworks.
If you're building a universal agentic layer - you should think about supporting both.
say
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
@tszzl I think the closest analogy is transportation market, we'll need intelligence of all shapes and sizes to push us forward, and there won't be one universally best product
@alexdotcodes Thanks for the highlight!
Locking the screen rotation is a must, otherwise it turns into a junk fest :)
Working on some optimisations now as it's very heavy in its current shape
@transitive_bs Wow, thanks! I was under impression my HN submissions continue to die in "new", good to know they reach someone
working on an app now :)
aded 8 new self-hostable projects in awesome-llm-services today.
new routers, llama.cpp operator for kubernetes, and some more, check it out :)
https://t.co/UzvFILw5mo
releasing mi v1.13.0 with fixes for sandboxing, handling reasoning/thinking blocks, tool call consistency, platform-specific streaming.
mi is an agentic harness in 30 lines of JavaScript, yet it's fully featured with /goal, custom tools, subagents and more.
https://t.co/BvE5aH57em
Releasing agent-store v0.1.4, now with prebuilt Linux arm64 binaries.
agent-store is token efficient wrapper over sqlite that allows agents to do lightweight schemaless data modelling - record decisions/tasks/workflows. the store is also reactive, so when something happens it can call your agent in response, for example, moving task in "review" -> run your code review agent.
https://t.co/pJpkJ4SR3n
https://t.co/Aj1P4WSOhM
you can steal this part for your agents:
"Consider this task deeply, how you can fulfill it above the expectations, consider what result will make me really happy about the outcome and won't require follow ups or a rework."