Hi, this looks really powerful, in that it seems to have many use cases.
One question I had:
Does every sandbox change end (when ready for production) in a pull request? If marketing sends me a pull request and I hate the code, what's the flow like for me to fix it?
The sandbox’s lifecycle is independent of the PR so that the same user or someone else in your team, like an engineer can come back to the same session and keep building until the result is ready to merge. So you can choose to iterate on the first drafts on the same sandbox. Curious, what flow you envision in your case?
Thanks! Lots of use cases. For the workflow you mentioned, the idea is marketing sends you a PR with the live preview. You can see the UI changes and if you need to change anything, you can open the session, which will let you modify, backtrack, or continue their work inside the same sandbox they used.
all of our customers are using Anthropic APIs for programmatic use. Codex and other providers let you use Oauth. But inside of a sandbox, you can technically use max plan since it is the same as using Claude locally.
I have a suggestion - an assistant which can help to set up all these agents, perhaps based on templates. You already covered various use cases, but it's not clear if it's something concrete.
I think a lot of people who might be interested in this product might be interested in an easy set-up process. Even if it doesn't really save time for an experienced ops person, a lot of people would rather talk to a bot than fill a form.
Good point. We launched our cli recently exactly for this. It comes with skills, so you can use your own local setup (Claude Code, Cursor, Codex), to build up templates, spin up sessions, and set things up. You can scope what agents can and can't do. I wouldn't recommend using an agent to set up guardrails. There should be some human oversight for this.
thanks! we used to build a lot of internal software / scripts for conferences or one-off data crunching. We saw LLMs getting better and how the world would go from building software for localhost to needing it to deploy. I then wrote an essay about my thoughts as software goes to zero back in January. https://www.gustrigos.com/essays/on-abundant-ephemeral-softw...
It's doing context management in the same way, but adds a learning loop based on code reviews. Currently works with Gitlab, Forgejo and Gitea but happy to accept a PR for Github support or co-write it.
- you mention proxying keys. One issue that we run into is that there are a bunch of tools that are really useful but require keys to be on disk (e.g. aws cli -- yes yes you can do IAM permissions but still). How do you guys think about those? (Especially since your setup onboarding is 'just install from npm or mise')
- poking around on the github, saw that you guys were at one point on fly.io. Did you guys end up switching off them? What motivated that if so?
- the CLI integration is cool! Is that actually teleporting remote sessions down to a local machine? Or is it more a window into the remote sandboxes?
would love to share notes! If you want to get in touch separately feel free at amol at noriagentic dot com.
Nori looks really cool, will set up sometime to exchange notes. But with regards to your questions:
- Proxying keys: We allow users to setup keys in the sandbox and use CLIs for some cases but we also support an egress gateway that intercepts and injects keys on the way out, which supports the major api integrations we offer.
- We still allow fly.io for deployments (so after a sandbox has a final app, you can deploy it and move out of the ephemeral sandbox). We never used them for sandboxing, but we will integrate into https://sprites.dev/ soon.
- For the CLI, we allow you to SSH into the remote sandox since a lot of workload add too much stress to local machines
Curious about what sandbox provider you use to power Nori and how you are handling the secrets/keys issue?
we're on fly, going to add modal support soon. I dont think our users care all that much, but we care for dev ergonomics.
keys are tricky. We don't have a great answer. I mean the proxy inject works well enough for mcp, but there is just such a long tail of tools that do something like 'read a key from disk and encrypt it before sending it out' which makes proxy management just a pain
Hi, this looks really powerful, in that it seems to have many use cases.
One question I had:
Does every sandbox change end (when ready for production) in a pull request? If marketing sends me a pull request and I hate the code, what's the flow like for me to fix it?
The sandbox’s lifecycle is independent of the PR so that the same user or someone else in your team, like an engineer can come back to the same session and keep building until the result is ready to merge. So you can choose to iterate on the first drafts on the same sandbox. Curious, what flow you envision in your case?
Thanks! Lots of use cases. For the workflow you mentioned, the idea is marketing sends you a PR with the live preview. You can see the UI changes and if you need to change anything, you can open the session, which will let you modify, backtrack, or continue their work inside the same sandbox they used.
I wonder how this would be looked upon by the ever changing rules of claude code.
If someone from Anthropic sees this, would love to know if I can use my max plan here.
all of our customers are using Anthropic APIs for programmatic use. Codex and other providers let you use Oauth. But inside of a sandbox, you can technically use max plan since it is the same as using Claude locally.
I have a suggestion - an assistant which can help to set up all these agents, perhaps based on templates. You already covered various use cases, but it's not clear if it's something concrete.
I think a lot of people who might be interested in this product might be interested in an easy set-up process. Even if it doesn't really save time for an experienced ops person, a lot of people would rather talk to a bot than fill a form.
Good point. We launched our cli recently exactly for this. It comes with skills, so you can use your own local setup (Claude Code, Cursor, Codex), to build up templates, spin up sessions, and set things up. You can scope what agents can and can't do. I wouldn't recommend using an agent to set up guardrails. There should be some human oversight for this.
This is amazing, what was the inspiration?
thanks! we used to build a lot of internal software / scripts for conferences or one-off data crunching. We saw LLMs getting better and how the world would go from building software for localhost to needing it to deploy. I then wrote an essay about my thoughts as software goes to zero back in January. https://www.gustrigos.com/essays/on-abundant-ephemeral-softw...
Checked license it said copyrighted which makes this unsuable for me.
my question is what advantage does this have over real FOSS agentic sandboxes
We're using a split license: apache-2.0 for the CLI/SDK/sandbox, MIT for templates, AGPLv3 for server components. Not copyrighted.
But beyond that, we are managing a higher level abstraction than sandboxes.
I wrote a simpler version of this for Claude Code as open source if that's useful to you: https://github.com/smithy-ai/smithy-ai
It's doing context management in the same way, but adds a learning loop based on code reviews. Currently works with Gitlab, Forgejo and Gitea but happy to accept a PR for Github support or co-write it.
Really cool! We're working on something similar over at https://norisessions.com/
A few questions
- you mention proxying keys. One issue that we run into is that there are a bunch of tools that are really useful but require keys to be on disk (e.g. aws cli -- yes yes you can do IAM permissions but still). How do you guys think about those? (Especially since your setup onboarding is 'just install from npm or mise')
- poking around on the github, saw that you guys were at one point on fly.io. Did you guys end up switching off them? What motivated that if so?
- the CLI integration is cool! Is that actually teleporting remote sessions down to a local machine? Or is it more a window into the remote sandboxes?
would love to share notes! If you want to get in touch separately feel free at amol at noriagentic dot com.
Nori looks really cool, will set up sometime to exchange notes. But with regards to your questions: - Proxying keys: We allow users to setup keys in the sandbox and use CLIs for some cases but we also support an egress gateway that intercepts and injects keys on the way out, which supports the major api integrations we offer. - We still allow fly.io for deployments (so after a sandbox has a final app, you can deploy it and move out of the ephemeral sandbox). We never used them for sandboxing, but we will integrate into https://sprites.dev/ soon. - For the CLI, we allow you to SSH into the remote sandox since a lot of workload add too much stress to local machines
Curious about what sandbox provider you use to power Nori and how you are handling the secrets/keys issue?
we're on fly, going to add modal support soon. I dont think our users care all that much, but we care for dev ergonomics.
keys are tricky. We don't have a great answer. I mean the proxy inject works well enough for mcp, but there is just such a long tail of tools that do something like 'read a key from disk and encrypt it before sending it out' which makes proxy management just a pain