SODAX SDKDigital asset flows you can build and deploy with AI.
Promoted
AgentHost
Five AI agents. A Sovereign box. Your cloud 24-7 Persistence
7 followers
Five AI agents. A Sovereign box. Your cloud 24-7 Persistence
7 followers
AgentHost puts Claude, Codex, Gemini, Kimi K3, and Hermes in one persistent control room on your own cloud. Shared task board, per-engine cost ledger, governance gates, audit trails. Deploy with one command. Your keys, your data, your rules.
Free Options
Launch Team / Built With
I built AgentHost because my AI agents died every time I closed my laptop. Context gone. 30 minutes rebuilding every morning.
Now five agents run 24/7 on a cloud box in my own account. They write code while I sleep. Every action has a budget cap and an audit trail. When one agent tried to merge broken code, another caught it and stopped it.
It runs on your infrastructure. Your cloud, your keys, your data. Founding 50 is $499 once with code TOMORROWSHERE. No AgentHost subscription — you pay your own cloud and model costs directly.
The useful part of this launch is the feedback. If you try it, tell me where the handoff, cost view, or recovery flow breaks down. I'll be here all day.
https://agenthost.space
honestly the multi engine setup looks super useful, but it would be really nice if there was a way to set per task budget caps that actually pause execution when you hit the limit, kind of like a soft kill switch for runaway costs. right now im guessing you just get a number at the end
@debuggerpro It's a running tally.
The governance layers are designed to allow you to set strictness levels and autonomy levels:
This is a feature we don't have yet, and I will strongly consider a "per task" budgeting and control mechanism.
17 days ago this was a dare: one command that moves your whole AI agent setup (Claude Code, Codex, Gemini, Hermes, Ollama) to a box in YOUR own cloud account.
774 commits later, here it is.
Why I built it: my agents died every time my laptop lid closed.
Every platform that promised to fix that wanted my API keys, my data, and $500 a month.
So AgentHost has one rule: there is no backend.
The CLI runs on your laptop, credentials get redacted before anything leaves it, and your secrets go into your own cloud provider's encrypted store.
There are no AgentHost servers to trust because there are no AgentHost servers.
The part I'm proudest of: the agents that built this reviewed each other's code the whole way.
A different AI reviews every merge.
A human approves anything irreversible.
This launch was red-teamed by the box itself, and it once blocked its own release with 4 critical findings.
What you get today: the CLI is free and open source.
Founders seats are $499 with code TOMORROWSHERE if you want the guided path.
I'll be here all day. Ask me anything, especially the hard security questions. Those are my favorite.
One thing I'd love to see is a visual diff or replay of each agent's work over time, so when governance flags a task I can scrub through exactly what Claude versus Kimi did and decide faster. Right now the audit log is text-heavy and it's hard to spot where an engine drifted off-policy.
@quinntaylor This is the sharpest read of the gap we have. The per-engine event stream shipped to the codebase this week: one durable transcript, every engine labeled. A scrubbable replay over it is exactly where that's pointed. The text-heavy audit log is the receipt today; the timeline is the receipt you can scrub. No date promised until it exists.
One thing I'd love to see is a way to set budget caps per engine that auto-pause jobs when a daily or weekly threshold gets hit, so a runaway loop on one model doesn't blow up the whole ledger. Would be a small addition but really useful for anyone juggling several providers at once.