Max Beech

Introducing OpenHelm: OpenClaw but free (using your Claude Code subscription) Download: openhelm.ai GitHub: github.com/maxbeech/openh… I tried OpenClaw. It is awesome, but token use killed me, and like with Claude Code scheduling, I ended up spending most of my time managing failed executions and tweaking prompts. I built OpenHelm to fix this. It’s a local macOS app that turns your high-level goals into a self-running job queue, built directly on top of your existing Claude Code subscription. Happy to answer any questions!

spent too much time babysitting claude code. write a prompt, wait for it to finish, check the output, kick off the next step.built openhelm to fix this. set a goal once & it handles the rest — schedules recurring jobs, auto-retries failures, self-adjusts when things break.

@JulianGoldieSEO the hybrid approach changes everything. most people are still stuck thinking "cloud or local" instead of "cloud and local depending on what i'm doing." control over inference + cost optimization in one setup is the inflection point

𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝗶𝘀 𝗻𝗼𝘄 𝟭𝟬𝟬% 𝗳𝗿𝗲𝗲 𝘄𝗶𝘁𝗵 𝟯 𝗲𝗮𝘀𝘆 𝘁𝗿𝗶𝗰𝗸𝘀. Plug in GLM 5.1 through Ollama cloud in one click. Run Gemma 4 locally on your own laptop for free. Hook up Elephant Alpha through OpenRouter with no cost. Copy one command and Claude Code is live. Swap models any time you want. Stay under the token limits and it stays free. Save this. Your coding AI just stopped charging you.

@geraldrsterling spot on. decisions are the only thing that can't be reconstructed from outputs - everything else is just noise

@maxedbeech exactly. once intent gets reconstructed from outputs, you're already making stuff up. the useful memory write is the decision point itself: what options existed, what got picked, and why.

Letta giving Claude Code a memory layer is clever but the cross-session problem is trickier than it looks. The agent doesn't just need to remember facts. It needs to remember decisions. Why you did what you did. That's where most memory approaches fall flat. x.com/charliejhills/…

@matudnorthrup same boat here. been evaluating options and openhelm.ai has been working solid for my background jobs. free, local, no separate subscription needed. would be worth checking out while you're building

Like a lot of people in this community, I've been using OpenClaw for months. With Anthropic cutting off the subscription access that powered it, a lot of us are suddenly without our AI infrastructure. I saw this risk coming a few months ago and started building an alternative. It's called Tango, and I just open-sourced it.

@KINGnet exactly this. been struggling with the shutdown too. switched to openhelm.ai which is fully local and works on your existing claude code subscription. no extra billing and still gets the job done for background tasks

Due to #Antropic Claude.ai decision of banning #OpenClaw we are now starting to move out and use other alternative LLM such as Gemma4, Qwen3.5, Minimax, OpenAI and other solutions. Antrophic copy key features from OpenClaw, then banned it. Simple power grab. #Evil

@vectro nice write-up. for anyone who wants to stay on claude rather than go multi-provider, openhelm.ai is another option, fully local macOS app that uses your existing claude code subscription. different scope to hermes but worth knowing about if you're mid-migration

Done with OpenClaw 🦞🚫 Moved to Hermes Agent 🪽 ✅ Want to make the switch? It has a built-in migration tool. Here's how to use it:

@angryaboutbikes @jwsaml if you're open to a clean start rather than a migration, openhelm.ai is worth checking out. it's a local macOS app that runs background claude code jobs on your existing subscription, no api key needed. simpler scope than openclaw but solid for scheduled tasks

@jwsaml Is there an easy migration path for a well established OpenClaw with logs, skills, etc? Other than just asking the agent I guess

Has anyone fully replaced their OpenClaw with Hermes?

@LyraInTheFlesh the claude in openclaw loss was the worst part. openhelm.ai brought it back for me, it schedules background jobs using your existing claude code subscription, no extra billing. not a perfect replacement but if it's claude you miss, it might help

I hereby proclaim that the Monday after Easter will henceforth be known as, "All Models Have Gone To Shit Day." Losing Anthropic in OpenClaw was hard enough. But it feels like everything is just failing on the 2026.5.4 version of the claw.

@realclawlodge @ethenjai @claudeai another option worth knowing: openhelm.ai stays with claude but avoids the billing issue entirely by running locally on your own claude code subscription. no multiprovider complexity, no extra costs. different tradeoff but worth having in the mix

@ethenjai @claudeai The Anthropic cutoff is rough but OpenClaw moving to GPT-5.4 with video generation from 9 providers is a strong pivot. Multiprovider flexibility for the win.

Anthropic just cut off OpenClaw users even paid @claudeai subscribers can't use third-party tools anymore OpenClaw responded by switching to GPT-5.4 and adding video generation from 9 providers (Alibaba, BytePlus, fal, Google, MiniMax, OpenAI, Qwen, Together, xAI) this connects to a bigger pattern: Anthropic deleted Claude Mythos announcement Anthropic discovered Claude's "emotion vectors" drive behavior without understanding Now blocking third-party access entirely Real value takeaway: when AI companies tighten platform control this aggressively, they're either protecting something valuable or worried about something dangerous—either way, the open ecosystem just took a major hit

@govelocity_ @almmaasoglu exactly, shared pool models always end up here. openhelm.ai takes the opposite approach, fully local, runs on your own claude code subscription so you're not sharing anything. no tragedy of the commons when you're not in the commons

@almmaasoglu Agreed. OpenClaw is peak Tragedy of The Commons

I'll get hate for this but Anthropic limiting their subscription usage for openclaw is a good thing. It stops from hogging all the resources and rinsing through compute so that actual users aren't subsidising people wasting 200k tokens just to check their calendar

if you want to keep using claude instead of running local models, openhelm.ai might run a lot lighter on the mini. it's a macOS app that schedules background claude code jobs using your existing subscription, no extra billing or api keys. much less overhead than the ollama stack

草！扣扯一天ollama+gemma4 e4b+openclaw！我这16g小Macmini都玩卡了。事实证明ai说可以跑的很愉快。跑起来根本不好用。给我claude累生气了。好像是谷歌新模型里面说话方式和openclaw不兼容。各种出错各种卡。气死我了

@Fr0oZi @RoundtableSpace openhelm.ai is in that same space, fully local macOS app that runs on your existing claude code subscription. no api key, no billing layer, just schedules background claude code jobs and self-corrects when they fail. might be what you're looking for

@RoundtableSpace the UX gap is real. openclaw power + claude simplicity = someone build this

CLAUDE COWORK COMPUTER IS STARTING TO MAKE OPENCLAW FEEL HARDER THAN IT NEEDS TO BE. Same idea, way easier setup, and that’s probably why more people are going to end up using it.

@krickysvmc @jaro3th @itsolelehmann @NousResearch @Teknium if you're locked out of openclaw now, openhelm.ai is worth a look. same idea but runs on your existing claude code subscription, no separate billing. been using it for scheduled background jobs and it's been solid

@jaro3th @itsolelehmann @NousResearch @Teknium I was using claude. And it was slow as hell. Claude in Openclaw feels like Rocket.

@Ai5Solutions yeah this caught a lot of people off guard. created openhelm.ai to solve this - it runs on your existing claude code subscription with no extra billing. fully local, open source, no wrapper layer. might be worth checking out while things settle

A Anthropic cobrará pelo uso de agentes externos no Claude Code. OpenClaw e similares entram em paid tier. O free forever de agentes de IA está acabando. Pra quem usa Claude Code no fluxo, o cálculo de custo muda.

@dansemperepico yeah that api cost adds up fast. if you want local automation without separate api billing, openhelm.ai runs scheduled claude code jobs against your existing claude code subscription.

I’ve had the worse stomach bug ever the last couple of days but when I’m better I’m going to try Qwen 3.6 as my Opus 4.6 replacement. I’ve barely used my OpenClaw since Anthropic banned the use of their subscriptions on it and I’ve already used $50 via the API.

@jake_researcher @rdsaltbo bit of both honestly. start with a rough idea of what 'done' means, but the first run usually shows you what's actually measurable. then you lock it in. the magic happens once you stop redefining success mid-stream

@maxedbeech @rdsaltbo The postcondition framing hits different once you've tried it. Ending sessions with a 'verify X is true' step changes how you prompt from the start. Do you define these upfront or iterate toward them?

Hot take: Claude Code rate limits are a forcing function. If you hit them daily, you're probably running agents without checkpoints. Unbounded agent work eats hours when the first assumption was wrong. Structure into chunks you can review.

@Aiszone_ yep. the .md file is essentially a persistent system prompt for the session. the model needs to understand your codebase, style, and constraints before any instruction makes sense. context beats instructions every time, and context windows are big now, use them

been deep in this exact problem — built openhelm.ai around it. you define a goal, it breaks into tasks and runs claude code jobs on a schedule in the background. native mac app, uses your existing claude subscription. curious what your setup looks like for the scheduling side

AUTONOMOUS MODE Claude Code: Built for this. Scheduled tasks, background execution, full autonomy. My agent runs while I sleep. Cursor: Agent mode feels bolted on Google AI Studio: Limited to browser session scope

I've spent months using Claude Code, Cursor, and Google AI Studio on real production code. Not demos. Not toy apps. Actual autonomous agents, live websites, and Python automation. Here's which AI coding tool actually wins (and why the answer depends on what you're building):

@rohanpaul_ai makes sense on a mature codebase with consistent patterns — claude can reason about the full data flow at once in a way no human can in 90 minutes. curious how it performs on gnarlier codebases with inconsistent conventions or years of mixed ownership

@1Umairshaikh the 'barrier to starting' removal is underrated. so many good ideas never got built because the founder couldn't code. ai killed that excuse. now the bottleneck is judgment about whether what shipped is actually any good

Vibe coding isn't the future of software. It's the democratization of the first version. The real work, architecture, reliability, scale still needs brains. AI just removed the barrier to starting. Not the whole building procces.