The big obvious central smoking gun that you'll get to in computer science 200 level classes is Amdahl's Law, which states:
> the overall performance improvement gained by optimizing a single part of a system is limited by the fraction of time that the improved part is actually used
You queue up some work for an agent. The LLM is going to do a bunch of work over time, and spend 20 minutes crunching on a task. Let's generously say it takes your PC 2 minute of it's CPU time for it to do the tool calls, to run the build, to run tests. If we expand this to 10 minutes to run it on a phone, that's indeed starting to be a big enough difference to notice. But in 99.9999% of cases, I don't think the harness consumes that much CPU and I don't think the growth factor is 5x to move to phone, and even if it did, it's still only an increase from 22 to 30 minutes: it's an async job either way, and the time budget is not dominated by the phone or PC running the harness.
Ideally yes, there's some intelligence to see: oh, we are about do to a build. Send the build to the build server, that's a 384 core 1U with terabytes of memory bandwidth and let it do that. But most work is not like running builds and tests. The harness doesn't need that. We need some small local computers cheap that we can have lots of running.
Model performance might radically improve in time, and that might change the Amdahl's Law calculations here. If you're paying for Turbo or Plaid or whatever, yeah, you maybe have the money to spend on a better harness too. I'd say that ideally these workloads become live migrate-able, that we can CRIU checkpoint/restore them across systems, ideally, anytime, so that we can give performance people performance when it actually counts, like the build concern above, when the agent is fast. LLM's built for speed like LFM2.5-8B-A1B (DiffuseGemini feels unlikely as it's fast, but low concurrency, but perhaps?), double the speed of many models, so that 20 minutes could become significantly less. But right now it feels like we need a lot of cheap not-performance critical harnesses that can sit around running, and that performance for them is not critical.\nhttps://www.liquid.ai/blog/lfm2-5-8b-a1b
If folks aren't aware, I also suggest taking a peak at Google Ax and Google Scion, two agent runtimes designed for scaling out, that are both kind of neat. https://github.com/google/ax https://github.com/googlecloudplatform/scion https://news.ycombinator.com/item?id=47675213"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"A low-carbon computing platform from your retired phones"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["google"],"value":"https://research.google/blog/a-low-carbon-computing-platform-from-your-retired-phones/"}},"_tags":["comment","author_jauntywundrkind","story_48515336"],"author":"jauntywundrkind","comment_text":"If you are trying to run lots of Pi or Hermes or whatever corporate whatever agent junk you have, to make a bunch of always on efficient agentic systems available to people, en masse, with low start-up costs, and high efficiency, there's a host of reasons that doesn't matter.
The big obvious central smoking gun that you'll get to in computer science 200 level classes is Amdahl's Law, which states:
> the overall performance improvement gained by optimizing a single part of a system is limited by the fraction of time that the improved part is actually used
You queue up some work for an agent. The LLM is going to do a bunch of work over time, and spend 20 minutes crunching on a task. Let's generously say it takes your PC 2 minute of it's CPU time for it to do the tool calls, to run the build, to run tests. If we expand this to 10 minutes to run it on a phone, that's indeed starting to be a big enough difference to notice. But in 99.9999% of cases, I don't think the harness consumes that much CPU and I don't think the growth factor is 5x to move to phone, and even if it did, it's still only an increase from 22 to 30 minutes: it's an async job either way, and the time budget is not dominated by the phone or PC running the harness.
Ideally yes, there's some intelligence to see: oh, we are about do to a build. Send the build to the build server, that's a 384 core 1U with terabytes of memory bandwidth and let it do that. But most work is not like running builds and tests. The harness doesn't need that. We need some small local computers cheap that we can have lots of running.
Model performance might radically improve in time, and that might change the Amdahl's Law calculations here. If you're paying for Turbo or Plaid or whatever, yeah, you maybe have the money to spend on a better harness too. I'd say that ideally these workloads become live migrate-able, that we can CRIU checkpoint/restore them across systems, ideally, anytime, so that we can give performance people performance when it actually counts, like the build concern above, when the agent is fast. LLM's built for speed like LFM2.5-8B-A1B (DiffuseGemini feels unlikely as it's fast, but low concurrency, but perhaps?), double the speed of many models, so that 20 minutes could become significantly less. But right now it feels like we need a lot of cheap not-performance critical harnesses that can sit around running, and that performance for them is not critical.\nhttps://www.liquid.ai/blog/lfm2-5-8b-a1b
If folks aren't aware, I also suggest taking a peak at Google Ax and Google Scion, two agent runtimes designed for scaling out, that are both kind of neat. https://github.com/google/ax https://github.com/googlecloudplatform/scion https://news.ycombinator.com/item?id=47675213","created_at":"2026-06-13T20:49:30Z","created_at_i":1781383770,"objectID":"48521303","parent_id":48518215,"story_id":48515336,"story_title":"A low-carbon computing platform from your retired phones","story_url":"https://research.google/blog/a-low-carbon-computing-platform-from-your-retired-phones/","updated_at":"2026-06-13T21:05:24Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jauntywundrkind"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"I'd evaluated both Google's Agent Substrate (that underlies Ax) and their Scion project. I really enjoy how Scion operates with existing tools really well. Ax/Agent Substrate is much more a greenfield independent effort, it's own thing.
I think Scion has so much more mature a disosition: you could write OpenCode plugins that enhance the runner, and use that locally, and use it in Scion. With Ax/Agent Substrate, you are opting in to a pretty huge stack that is just Agent Substrate, that is their runners, their harness, their substrate. I do think their actor model is pretty neat! It's neat having the agent have such primacy! But it feels so much less integrative, is such it's own thing. Scion, to me, is much more interesting an effort, that similarly helps scale out agentic workloads.
https://github.com/googlecloudplatform/scion"},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"AX \u2013 Google\u2019s Open Agentic Orchestrator"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://agentexecutor.io"}},"_tags":["comment","author_jauntywundrkind","story_49780797"],"author":"jauntywundrkind","children":[49781315,49796661],"comment_text":"I'd evaluated both Google's Agent Substrate (that underlies Ax) and their Scion project. I really enjoy how Scion operates with existing tools really well. Ax/Agent Substrate is much more a greenfield independent effort, it's own thing.
I think Scion has so much more mature a disosition: you could write OpenCode plugins that enhance the runner, and use that locally, and use it in Scion. With Ax/Agent Substrate, you are opting in to a pretty huge stack that is just Agent Substrate, that is their runners, their harness, their substrate. I do think their actor model is pretty neat! It's neat having the agent have such primacy! But it feels so much less integrative, is such it's own thing. Scion, to me, is much more interesting an effort, that similarly helps scale out agentic workloads.
https://github.com/googlecloudplatform/scion","created_at":"2026-09-20T23:15:49Z","created_at_i":1789946149,"objectID":"49781116","parent_id":49780797,"story_id":49780797,"story_title":"AX \u2013 Google\u2019s Open Agentic Orchestrator","story_url":"https://agentexecutor.io","updated_at":"2026-09-23T02:35:48Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"briga"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"I don't think there is really any convergence going on. The agentic ecosystem is continuing to multiply on a daily basis and everyone and their grandma has written a new agent framework--people are stepping over each other to get these new projects out the door.
That said, I think Google's ADK ecosystem and this new AX platform is promising--I would expect Google to maintain this and other tooling around this for years to come.
To the Googlers out there: is Google using this at any capacity for internal projects?"},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"AX \u2013 Google\u2019s Open Agentic Orchestrator"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://agentexecutor.io"}},"_tags":["comment","author_briga","story_49780797"],"author":"briga","children":[49782009,49782704,49782936,49783002,49783664,49785086],"comment_text":"I don't think there is really any convergence going on. The agentic ecosystem is continuing to multiply on a daily basis and everyone and their grandma has written a new agent framework--people are stepping over each other to get these new projects out the door.
That said, I think Google's ADK ecosystem and this new AX platform is promising--I would expect Google to maintain this and other tooling around this for years to come.
To the Googlers out there: is Google using this at any capacity for internal projects?","created_at":"2026-09-21T01:21:32Z","created_at_i":1789953692,"objectID":"49781958","parent_id":49781801,"story_id":49780797,"story_title":"AX \u2013 Google\u2019s Open Agentic Orchestrator","story_url":"https://agentexecutor.io","updated_at":"2026-10-01T11:30:44Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rakyll"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"Disclaimer: I'm one of the co-creators of this project.
AX is a layer that is closer to job orchestration, Agent Substrate. It's NOT an agentic framework. We use Antigravity for a few generative features but are abstracting away some of these components so anyone can bring their own. AX is trying to solve some tedious things everyone has to deal with. Layering execution with the underlying stateful worker, wiring up the network correctly, providing sub "task" identity, provisioning the right environment, proving stateful branching, auto discovery.
We want to keep Agent Substrate free of generative features and still need a layer above for some integrations that require agentic concepts and identity.
Google compute services are under GCP and I work on Kubernetes. So the comparison with Gemini CLI is not relevant here."},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"AX \u2013 Google\u2019s Open Agentic Orchestrator"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://agentexecutor.io"}},"_tags":["comment","author_rakyll","story_49780797"],"author":"rakyll","comment_text":"Disclaimer: I'm one of the co-creators of this project.
AX is a layer that is closer to job orchestration, Agent Substrate. It's NOT an agentic framework. We use Antigravity for a few generative features but are abstracting away some of these components so anyone can bring their own. AX is trying to solve some tedious things everyone has to deal with. Layering execution with the underlying stateful worker, wiring up the network correctly, providing sub "task" identity, provisioning the right environment, proving stateful branching, auto discovery.
We want to keep Agent Substrate free of generative features and still need a layer above for some integrations that require agentic concepts and identity.
Google compute services are under GCP and I work on Kubernetes. So the comparison with Gemini CLI is not relevant here.","created_at":"2026-09-21T20:57:13Z","created_at_i":1790024233,"objectID":"49793311","parent_id":49781725,"story_id":49780797,"story_title":"AX \u2013 Google\u2019s Open Agentic Orchestrator","story_url":"https://agentexecutor.io","updated_at":"2026-09-28T03:42:03Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sheepscreek"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"> Drawing on agentic runtime research from Google DeepMind alongside deep experience in large-scale isolation, resumption, and scheduling, AX is being built as an open, declarative control plane purpose-built...
The project seems like an open-source initiative born out of the experience of some Googlers but not being used at Google. So, the title appears a bit misleading - people will be misled."},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"AX \u2013 Google\u2019s Open Agentic Orchestrator"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://agentexecutor.io"}},"_tags":["comment","author_sheepscreek","story_49780797"],"author":"sheepscreek","comment_text":"> Drawing on agentic runtime research from Google DeepMind alongside deep experience in large-scale isolation, resumption, and scheduling, AX is being built as an open, declarative control plane purpose-built...
The project seems like an open-source initiative born out of the experience of some Googlers but not being used at Google. So, the title appears a bit misleading - people will be misled.","created_at":"2026-09-21T02:27:46Z","created_at_i":1789957666,"objectID":"49782384","parent_id":49780797,"story_id":49780797,"story_title":"AX \u2013 Google\u2019s Open Agentic Orchestrator","story_url":"https://agentexecutor.io","updated_at":"2026-09-21T14:34:31Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"NBenkovich"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"Hey HN!
Happy to share our work with you. The paper presents our system architecture, provides links to the sources, and compares our approach with solutions such as AWS Bedrock AgentCore, Cloudflare Agents, Google AX, and others.
I'm here to answer on your questions."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["agentic"],"value":"Agyn \u2013 OSS Alternative to AWS AgentCore: FS, Network, Credential Isolation"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://arxiv.org/abs/2605.27575"}},"_tags":["comment","author_NBenkovich","story_48411290"],"author":"NBenkovich","comment_text":"Hey HN!
Happy to share our work with you. The paper presents our system architecture, provides links to the sources, and compares our approach with solutions such as AWS Bedrock AgentCore, Cloudflare Agents, Google AX, and others.
I'm here to answer on your questions.","created_at":"2026-06-05T12:06:50Z","created_at_i":1780661210,"objectID":"48411291","parent_id":48411290,"story_id":48411290,"story_title":"Agyn \u2013 OSS Alternative to AWS AgentCore: FS, Network, Credential Isolation","story_url":"https://arxiv.org/abs/2605.27575","updated_at":"2026-06-05T12:09:36Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cohix"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"Ax, Google's new highly distributed agent executor"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["google","ax"],"value":"https://github.com/google/ax"}},"_tags":["story","author_cohix","story_48307107"],"author":"cohix","children":[48307108],"created_at":"2026-05-28T10:40:12Z","created_at_i":1779964812,"num_comments":1,"objectID":"48307107","points":2,"story_id":48307107,"title":"Ax, Google's new highly distributed agent executor","updated_at":"2026-09-22T14:03:18Z","url":"https://github.com/google/ax"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rnella01"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["google","agentic"],"value":"Google's open source distributed agent runtime"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["google","ax"],"value":"https://github.com/google/ax"}},"_tags":["story","author_rnella01","story_48217440"],"author":"rnella01","created_at":"2026-05-21T03:18:15Z","created_at_i":1779333495,"num_comments":0,"objectID":"48217440","points":3,"story_id":48217440,"title":"Google's open source distributed agent runtime","updated_at":"2026-05-21T07:51:49Z","url":"https://github.com/google/ax"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dbmikus"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"I think it will be, but I don't think you need a standalone machine! If you run things inside a VM, you can get safety and control over access and networks
A standalone machine is nice if you need more compute resources or if you want an always-on machine you can connect to from your laptop, phone, etc.
It doesn't look like Google's AX is quite the plug-and-play fit for running agents on a computer you own, since it requires setting up a K8S cluster, etc.
I think what's needed is something like a zero-setup combo of Tailscale and Firecracker
I'm trying to work towards that with my startup (https://github.com/gofixpoint/amika) but the bring-your-own-computer part doesn't work quite yet."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["google","ax"],"value":"AX \u2013 Google\u2019s Open Agentic Orchestrator"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://agentexecutor.io"}},"_tags":["comment","author_dbmikus","story_49780797"],"author":"dbmikus","comment_text":"I think it will be, but I don't think you need a standalone machine! If you run things inside a VM, you can get safety and control over access and networks
A standalone machine is nice if you need more compute resources or if you want an always-on machine you can connect to from your laptop, phone, etc.
It doesn't look like Google's AX is quite the plug-and-play fit for running agents on a computer you own, since it requires setting up a K8S cluster, etc.
I think what's needed is something like a zero-setup combo of Tailscale and Firecracker
I'm trying to work towards that with my startup (https://github.com/gofixpoint/amika) but the bring-your-own-computer part doesn't work quite yet.","created_at":"2026-09-21T02:10:34Z","created_at_i":1789956634,"objectID":"49782274","parent_id":49781979,"story_id":49780797,"story_title":"AX \u2013 Google\u2019s Open Agentic Orchestrator","story_url":"https://agentexecutor.io","updated_at":"2026-09-21T03:21:42Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"calebkaiser"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"That's not quite right. A recurring trend in ML is figuring out how to get an elastic interface for the the particular quirks of ML workloads (source: I worked on an open source one years ago for inference). Recently, there's been a lot of interest in doing this for agent workloads. Google recently released something called ax that is similar in spirit. The core of it is this: https://github.com/agent-substrate/substrate
At a high level, agent sandboxes have peculiar needs. Agents are really bursty, but also long lived. You need low latency suspend/resume calls. Checkpointing and recovery have some particular considerations. And naive approaches are often really wasteful, but over optimizing without harming durability, isolation, or consistency in performance can get tricky.
This is the new hot infra topic for agent swarms, for the time being. Whether that's impressive or not is up to the reader I guess, but it's a pretty involved project nonetheless."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"DeepSeek Elastic Compute (DSec)"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://arxiv.org/abs/2609.22978"}},"_tags":["comment","author_calebkaiser","story_49859112"],"author":"calebkaiser","children":[49863943,49865014],"comment_text":"That's not quite right. A recurring trend in ML is figuring out how to get an elastic interface for the the particular quirks of ML workloads (source: I worked on an open source one years ago for inference). Recently, there's been a lot of interest in doing this for agent workloads. Google recently released something called ax that is similar in spirit. The core of it is this: https://github.com/agent-substrate/substrate
At a high level, agent sandboxes have peculiar needs. Agents are really bursty, but also long lived. You need low latency suspend/resume calls. Checkpointing and recovery have some particular considerations. And naive approaches are often really wasteful, but over optimizing without harming durability, isolation, or consistency in performance can get tricky.
This is the new hot infra topic for agent swarms, for the time being. Whether that's impressive or not is up to the reader I guess, but it's a pretty involved project nonetheless.","created_at":"2026-09-27T04:16:11Z","created_at_i":1790482571,"objectID":"49863306","parent_id":49861443,"story_id":49859112,"story_title":"DeepSeek Elastic Compute (DSec)","story_url":"https://arxiv.org/abs/2609.22978","updated_at":"2026-09-27T19:53:50Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Intendant"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"Location: Austin, TX
Remote: Yes (5+ yrs fully remote)
Willing to relocate: Yes, for the right role
Open to: Full-time or contract
Technologies: Claude Agent SDK, MCP (author + consumer), RAG, evals,\n Langfuse, LLM-ops; AWS (EKS, Lambda, Bedrock) + Azure (AKS, ADO);\n Databricks, Terraform, Kubernetes; Python, TypeScript, Rust
R\u00e9sum\u00e9: https://drive.google.com/file/d/13j_Ax_0soXWrtoMfFORqV4A34kn...
LinkedIn: linkedin.com/in/branperez
GitHub: github.com/BrandonPerez-Dev
Email: perez.brand.on.off at gmail dot com
I build AI agents and the platforms they run on. This year, on WEX's\n incubator team, I built a platform on the Claude Agent SDK: a runtime,\n an agent builder, a builder MCP, and skill-graph composition for\n sub-agents. It took new-agent build time from weeks to days, and I used\n it to build several agents. The customer-facing ones I demoed to WEX's\n C-suite.
Before that I built SSM Health's first internal agent platform. The\n "AI Agent Factory" took new-agent spin-up from weeks to hours. I stood\n up their first LLM observability stack, Langfuse on AWS EKS, and built\n the HIPAA-compliant cross-cloud ML infrastructure underneath it. I came\n up through cloud and platform engineering, so I build the infra myself.
I also wrote Ostia, an open-source Rust MCP server that sandboxes AI\n agent shell calls at the OS level with Landlock, seccomp, and Linux\n namespaces: github.com/BrandonPerez-Dev/ostia. Once it was solid, I\n started handing agents small CLIs as tools through that sandbox instead\n of standing up a separate MCP server per integration."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Who wants to be hired? (June 2026)"}},"_tags":["comment","author_Intendant","story_48357724"],"author":"Intendant","comment_text":"Location: Austin, TX
Remote: Yes (5+ yrs fully remote)
Willing to relocate: Yes, for the right role
Open to: Full-time or contract
Technologies: Claude Agent SDK, MCP (author + consumer), RAG, evals,\n Langfuse, LLM-ops; AWS (EKS, Lambda, Bedrock) + Azure (AKS, ADO);\n Databricks, Terraform, Kubernetes; Python, TypeScript, Rust
R\u00e9sum\u00e9: https://drive.google.com/file/d/13j_Ax_0soXWrtoMfFORqV4A34kn...
LinkedIn: linkedin.com/in/branperez
GitHub: github.com/BrandonPerez-Dev
Email: perez.brand.on.off at gmail dot com
I build AI agents and the platforms they run on. This year, on WEX's\n incubator team, I built a platform on the Claude Agent SDK: a runtime,\n an agent builder, a builder MCP, and skill-graph composition for\n sub-agents. It took new-agent build time from weeks to days, and I used\n it to build several agents. The customer-facing ones I demoed to WEX's\n C-suite.
Before that I built SSM Health's first internal agent platform. The\n "AI Agent Factory" took new-agent spin-up from weeks to hours. I stood\n up their first LLM observability stack, Langfuse on AWS EKS, and built\n the HIPAA-compliant cross-cloud ML infrastructure underneath it. I came\n up through cloud and platform engineering, so I build the infra myself.
I also wrote Ostia, an open-source Rust MCP server that sandboxes AI\n agent shell calls at the OS level with Landlock, seccomp, and Linux\n namespaces: github.com/BrandonPerez-Dev/ostia. Once it was solid, I\n started handing agents small CLIs as tools through that sandbox instead\n of standing up a separate MCP server per integration.","created_at":"2026-06-02T00:47:48Z","created_at_i":1780361268,"objectID":"48364468","parent_id":48357724,"story_id":48357724,"story_title":"Ask HN: Who wants to be hired? (June 2026)","updated_at":"2026-06-02T00:51:05Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"coolwulf"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"Hey HN,
I've been building CoolWulf AI (https://coolwulfai.com), a self-hosted personal AI assistant. After seeing OpenClaw blow up, I wanted to share what I've been working on \u2014 a different approach to the same problem.
*Why I built this:*
I tried OpenClaw and found the Node.js/TypeScript stack heavy for what's essentially a local agent. pnpm, Node 22+, React \u2014 lots of moving parts. I wanted something that's a single binary, zero runtime dependencies, and feels native on macOS. So I built it in Go.
*How it's different from OpenClaw:*
- *Single binary, no runtime needed.* Download, set your API key, run. No Node.js, no pnpm, no build step. One ~100 MB binary.\n- *Built in Go.* Fast startup, low memory footprint, compiles to a native executable. No garbage collector pauses from a JS runtime sitting in the background.\n- *macOS-native integrations.* Deep AppleScript-based control of Apple Notes, Reminders, Calendar, Terminal.app, and WeChat desktop. These aren't browser hacks \u2014 they use the native accessibility and scripting APIs.\n- *WeChat support.* This was a big one for me. WeChat desktop on macOS is a native Cocoa/Qt app with a readable accessibility tree. CoolWulf can read messages, send messages, search contacts, and navigate chats \u2014 all via AX APIs and CGEvent. I haven't seen another AI agent do this.\n- *Simpler setup.* Web-based first-time wizard. Configure your LLM provider, connect Gmail/Calendar via OAuth, enable messaging connectors \u2014 all from the browser. No terminal wizards, no YAML files.
*What it does:*
- 20+ LLM providers (OpenAI, Claude, Gemini, DeepSeek, Groq, Ollama, local models via vLLM/LM Studio)\n- Messaging: WhatsApp, Teams, Telegram, Slack, WeChat\n- Email: Gmail and Yahoo Mail with full OAuth\n- Calendar: Google Calendar + Apple Calendar\n- Browser automation: Chrome CDP + Playwright via MCP\n- Task management: org-mode style with scheduled tasks, cron jobs, automatic execution\n- Persistent memory: SQLite + vector embeddings for semantic search across conversations\n- Background heartbeat: runs 24/7, monitors your systems, executes due tasks, sends alerts\n- Web dashboard for chat, tasks, scheduled jobs, and settings
*Architecture choices:*
Go was the right call. The binary compiles in seconds, cross-compiles trivially, and the concurrency model (goroutines for heartbeat, browser sessions, connector polling) maps perfectly to an always-on agent. SQLite with vector extensions (sqlite-vec) gives us semantic memory without running a separate vector DB.
Try it: https://coolwulfai.com
Happy to answer questions about the Go implementation, WeChat automation, or the macOS accessibility approach."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: CoolWulf AI \u2013 A personal AI assistant built in Go, optimized for macOS"},"url":{"matchLevel":"none","matchedWords":[],"value":"http://coolwulfAI.com"}},"_tags":["story","author_coolwulf","story_47044036","show_hn"],"author":"coolwulf","children":[47044074],"created_at":"2026-02-17T05:32:16Z","created_at_i":1771306336,"num_comments":1,"objectID":"47044036","points":1,"story_id":47044036,"story_text":"Hey HN,
I've been building CoolWulf AI (https://coolwulfai.com), a self-hosted personal AI assistant. After seeing OpenClaw blow up, I wanted to share what I've been working on \u2014 a different approach to the same problem.
*Why I built this:*
I tried OpenClaw and found the Node.js/TypeScript stack heavy for what's essentially a local agent. pnpm, Node 22+, React \u2014 lots of moving parts. I wanted something that's a single binary, zero runtime dependencies, and feels native on macOS. So I built it in Go.
*How it's different from OpenClaw:*
- *Single binary, no runtime needed.* Download, set your API key, run. No Node.js, no pnpm, no build step. One ~100 MB binary.\n- *Built in Go.* Fast startup, low memory footprint, compiles to a native executable. No garbage collector pauses from a JS runtime sitting in the background.\n- *macOS-native integrations.* Deep AppleScript-based control of Apple Notes, Reminders, Calendar, Terminal.app, and WeChat desktop. These aren't browser hacks \u2014 they use the native accessibility and scripting APIs.\n- *WeChat support.* This was a big one for me. WeChat desktop on macOS is a native Cocoa/Qt app with a readable accessibility tree. CoolWulf can read messages, send messages, search contacts, and navigate chats \u2014 all via AX APIs and CGEvent. I haven't seen another AI agent do this.\n- *Simpler setup.* Web-based first-time wizard. Configure your LLM provider, connect Gmail/Calendar via OAuth, enable messaging connectors \u2014 all from the browser. No terminal wizards, no YAML files.
*What it does:*
- 20+ LLM providers (OpenAI, Claude, Gemini, DeepSeek, Groq, Ollama, local models via vLLM/LM Studio)\n- Messaging: WhatsApp, Teams, Telegram, Slack, WeChat\n- Email: Gmail and Yahoo Mail with full OAuth\n- Calendar: Google Calendar + Apple Calendar\n- Browser automation: Chrome CDP + Playwright via MCP\n- Task management: org-mode style with scheduled tasks, cron jobs, automatic execution\n- Persistent memory: SQLite + vector embeddings for semantic search across conversations\n- Background heartbeat: runs 24/7, monitors your systems, executes due tasks, sends alerts\n- Web dashboard for chat, tasks, scheduled jobs, and settings
*Architecture choices:*
Go was the right call. The binary compiles in seconds, cross-compiles trivially, and the concurrency model (goroutines for heartbeat, browser sessions, connector polling) maps perfectly to an always-on agent. SQLite with vector extensions (sqlite-vec) gives us semantic memory without running a separate vector DB.
Try it: https://coolwulfai.com
Happy to answer questions about the Go implementation, WeChat automation, or the macOS accessibility approach.","title":"Show HN: CoolWulf AI \u2013 A personal AI assistant built in Go, optimized for macOS","updated_at":"2026-03-05T23:34:13Z","url":"http://coolwulfAI.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jcw90210"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"Googles Agent Identity is already built around SPIFFE but others are not.
I believe the substrate egress-gateway needs to handover the internal SPIFFE one to an external system (e.g. Entra Agent ID). Not sure if that should be part of substrate or kagent/ax/.."},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"AX \u2013 Google\u2019s Open Agentic Orchestrator"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["agentic"],"value":"https://agentexecutor.io"}},"_tags":["comment","author_jcw90210","story_49780797"],"author":"jcw90210","comment_text":"Googles Agent Identity is already built around SPIFFE but others are not.
I believe the substrate egress-gateway needs to handover the internal SPIFFE one to an external system (e.g. Entra Agent ID). Not sure if that should be part of substrate or kagent/ax/..","created_at":"2026-09-21T15:43:24Z","created_at_i":1790005404,"objectID":"49788750","parent_id":49787179,"story_id":49780797,"story_title":"AX \u2013 Google\u2019s Open Agentic Orchestrator","story_url":"https://agentexecutor.io","updated_at":"2026-09-24T14:50:25Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mpeg"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"The project that ax builds on top of does have that disclaimer, and sits on a non-google organisation https://github.com/agent-substrate/substrate"},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"AX \u2013 Google\u2019s Open Agentic Orchestrator"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["agentic"],"value":"https://agentexecutor.io"}},"_tags":["comment","author_mpeg","story_49780797"],"author":"mpeg","children":[49789670],"comment_text":"The project that ax builds on top of does have that disclaimer, and sits on a non-google organisation https://github.com/agent-substrate/substrate","created_at":"2026-09-21T04:43:12Z","created_at_i":1789965792,"objectID":"49783138","parent_id":49781588,"story_id":49780797,"story_title":"AX \u2013 Google\u2019s Open Agentic Orchestrator","story_url":"https://agentexecutor.io","updated_at":"2026-09-21T16:44:00Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"coolwulf"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["google","ax","agentic"],"value":"Hey HN,
I've been building CoolWulf AI (https://coolwulfai.com), a self-hosted personal AI assistant. After seeing OpenClaw blow up, I wanted to share what I've been working on \u2014 a different approach to the same problem.
*Why I built this:*
I tried OpenClaw and found the Node.js/TypeScript stack heavy for what's essentially a local agent. pnpm, Node 22+, React \u2014 lots of moving parts. I wanted something that's a single binary, zero runtime dependencies, and feels native on macOS. So I built it in Go.
*How it's different from OpenClaw:*
- *Single binary, no runtime needed.* Download, set your API key, run. No Node.js, no pnpm, no build step. One ~100 MB binary.\n- *Built in Go.* Fast startup, low memory footprint, compiles to a native executable. No garbage collector pauses from a JS runtime sitting in the background.\n- *macOS-native integrations.* Deep AppleScript-based control of Apple Notes, Reminders, Calendar, Terminal.app, and WeChat desktop. These aren't browser hacks \u2014 they use the native accessibility and scripting APIs.\n- *WeChat support.* This was a big one for me. WeChat desktop on macOS is a native Cocoa/Qt app with a readable accessibility tree. CoolWulf can read messages, send messages, search contacts, and navigate chats \u2014 all via AX APIs and CGEvent. I haven't seen another AI agent do this.\n- *Simpler setup.* Web-based first-time wizard. Configure your LLM provider, connect Gmail/Calendar via OAuth, enable messaging connectors \u2014 all from the browser. No terminal wizards, no YAML files.
*What it does:*
- 20+ LLM providers (OpenAI, Claude, Gemini, DeepSeek, Groq, Ollama, local models via vLLM/LM Studio)\n- Messaging: WhatsApp, Teams, Telegram, Slack, WeChat\n- Email: Gmail and Yahoo Mail with full OAuth\n- Calendar: Google Calendar + Apple Calendar\n- Browser automation: Chrome CDP + Playwright via MCP\n- Task management: org-mode style with scheduled tasks, cron jobs, automatic execution\n- Persistent memory: SQLite + vector embeddings for semantic search across conversations\n- Background heartbeat: runs 24/7, monitors your systems, executes due tasks, sends alerts\n- Web dashboard for chat, tasks, scheduled jobs, and settings
*Architecture choices:*
Go was the right call. The binary compiles in seconds, cross-compiles trivially, and the concurrency model (goroutines for heartbeat, browser sessions, connector polling) maps perfectly to an always-on agent. SQLite with vector extensions (sqlite-vec) gives us semantic memory without running a separate vector DB.
Try it: https://coolwulfai.com
Happy to answer questions about the Go implementation, WeChat automation, or the macOS accessibility approach."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: CoolWulf AI \u2013 A personal AI assistant built in Go, optimized for macOS"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"http://coolwulfAI.com"}},"_tags":["comment","author_coolwulf","story_47044036"],"author":"coolwulf","comment_text":"Hey HN,
I've been building CoolWulf AI (https://coolwulfai.com), a self-hosted personal AI assistant. After seeing OpenClaw blow up, I wanted to share what I've been working on \u2014 a different approach to the same problem.
*Why I built this:*
I tried OpenClaw and found the Node.js/TypeScript stack heavy for what's essentially a local agent. pnpm, Node 22+, React \u2014 lots of moving parts. I wanted something that's a single binary, zero runtime dependencies, and feels native on macOS. So I built it in Go.
*How it's different from OpenClaw:*
- *Single binary, no runtime needed.* Download, set your API key, run. No Node.js, no pnpm, no build step. One ~100 MB binary.\n- *Built in Go.* Fast startup, low memory footprint, compiles to a native executable. No garbage collector pauses from a JS runtime sitting in the background.\n- *macOS-native integrations.* Deep AppleScript-based control of Apple Notes, Reminders, Calendar, Terminal.app, and WeChat desktop. These aren't browser hacks \u2014 they use the native accessibility and scripting APIs.\n- *WeChat support.* This was a big one for me. WeChat desktop on macOS is a native Cocoa/Qt app with a readable accessibility tree. CoolWulf can read messages, send messages, search contacts, and navigate chats \u2014 all via AX APIs and CGEvent. I haven't seen another AI agent do this.\n- *Simpler setup.* Web-based first-time wizard. Configure your LLM provider, connect Gmail/Calendar via OAuth, enable messaging connectors \u2014 all from the browser. No terminal wizards, no YAML files.
*What it does:*
- 20+ LLM providers (OpenAI, Claude, Gemini, DeepSeek, Groq, Ollama, local models via vLLM/LM Studio)\n- Messaging: WhatsApp, Teams, Telegram, Slack, WeChat\n- Email: Gmail and Yahoo Mail with full OAuth\n- Calendar: Google Calendar + Apple Calendar\n- Browser automation: Chrome CDP + Playwright via MCP\n- Task management: org-mode style with scheduled tasks, cron jobs, automatic execution\n- Persistent memory: SQLite + vector embeddings for semantic search across conversations\n- Background heartbeat: runs 24/7, monitors your systems, executes due tasks, sends alerts\n- Web dashboard for chat, tasks, scheduled jobs, and settings
*Architecture choices:*
Go was the right call. The binary compiles in seconds, cross-compiles trivially, and the concurrency model (goroutines for heartbeat, browser sessions, connector polling) maps perfectly to an always-on agent. SQLite with vector extensions (sqlite-vec) gives us semantic memory without running a separate vector DB.
Try it: https://coolwulfai.com
Happy to answer questions about the Go implementation, WeChat automation, or the macOS accessibility approach.","created_at":"2026-02-17T05:37:32Z","created_at_i":1771306652,"objectID":"47044074","parent_id":47044036,"story_id":47044036,"story_title":"Show HN: CoolWulf AI \u2013 A personal AI assistant built in Go, optimized for macOS","story_url":"http://coolwulfAI.com","updated_at":"2026-03-05T23:34:13Z"}],"hitsPerPage":20,"nbHits":19,"nbPages":1,"page":0,"params":"query=google+ax+agentic&advancedSyntax=true&analyticsTags=backend","processingTimeMS":13,"processingTimingsMS":{"_request":{"roundTrip":15},"afterFetch":{"format":{"highlighting":1,"total":1},"merge":{"mergeLoop":{"prepareNextHit":1,"total":1},"total":1},"total":2},"fetch":{"query":9,"scanning":1,"total":11},"total":13},"query":"google ax agentic","serverTimeMS":15}