AgentBox is an SDK for running coding agents (Claude Code, Codex, OpenCode) inside sandboxes (Docker, E2B, Modal, Daytona, Vercel). One API. Swap the agent or the sandbox and your code doesn't change.
Think of it as what the AI SDK did for LLMs, but for agent + runtime.
Most wrappers call agents in non interactive mode (claude --print, codex exec). AgentBox instead boots each agent's native server inside the sandbox (Codex app-server over JSON-RPC, OpenCode serve over HTTP/SSE, Claude Code's SDK WebSocket transport) and drives it from the host, behaving like an interactive session.
Live demo: https://agentbox-demo-175164121374.us-west1.run.app
Feedback welcome, especially on the abstractions and which providers to add next!"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"Show HN: AgentBox \u2013 SDK to Run Claude Code, Codex, or OpenCode in Any Sandbox"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"https://github.com/TwillAI/agentbox-sdk"}},"_tags":["story","author_willydouhard","story_47876788","show_hn"],"author":"willydouhard","created_at":"2026-04-23T15:19:20Z","created_at_i":1776957560,"num_comments":0,"objectID":"47876788","points":8,"story_id":47876788,"story_text":"Hi HN,
AgentBox is an SDK for running coding agents (Claude Code, Codex, OpenCode) inside sandboxes (Docker, E2B, Modal, Daytona, Vercel). One API. Swap the agent or the sandbox and your code doesn't change.
Think of it as what the AI SDK did for LLMs, but for agent + runtime.
Most wrappers call agents in non interactive mode (claude --print, codex exec). AgentBox instead boots each agent's native server inside the sandbox (Codex app-server over JSON-RPC, OpenCode serve over HTTP/SSE, Claude Code's SDK WebSocket transport) and drives it from the host, behaving like an interactive session.
Live demo: https://agentbox-demo-175164121374.us-west1.run.app
Feedback welcome, especially on the abstractions and which providers to add next!","title":"Show HN: AgentBox \u2013 SDK to Run Claude Code, Codex, or OpenCode in Any Sandbox","updated_at":"2026-06-03T02:04:39Z","url":"https://github.com/TwillAI/agentbox-sdk"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"madarco"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"Hi HN, I'm Marco,
I've built this MIT-licensed CLI command to parallelize my work with claude / codex / PI, because with worktrees I had ports conflicts, agents conflicting for browser use, or touching external files.
It copies your project, skills, subscription oauth, and can optionally carry external files, supports LFS, and adds Chrome and screensharing via webVNC, so I use it for Godot game for eg.
To use:
agentbox claude # launch on docker, also with pi, codex, opencode\n agentbox hetzner claude # also digitalocean, e2b, vercel, daytona, islo, Sprites, Tenki\n agentbox remote-docker add buildbox marco@10.0.0.9\n agentbox docker:buildbox claude\n\nI use it to drive multiple subagents from a "Firstmate" agent:
I request several changes or gives him a backlog of items and claude will splits them in groups into parallel cloud sandboxes to avoid conflicts, and drive the sessions with the special "agentbox drive" commands, for eg. to wait-for idle, plan-approval, questions etc.
I find it really usefull so far, since it allows me to keep using my existing IDE/tools.
More info on https://github.com/madarco/agentbox"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"Show HN: Agentbox \u2013 Teleport your repo into sandboxes with no worktree juggling"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"https://github.com/madarco/agentbox"}},"_tags":["story","author_madarco","story_49728120","show_hn"],"author":"madarco","children":[49908334],"created_at":"2026-09-16T15:01:48Z","created_at_i":1789570908,"num_comments":0,"objectID":"49728120","points":4,"story_id":49728120,"story_text":"Hi HN, I'm Marco,
I've built this MIT-licensed CLI command to parallelize my work with claude / codex / PI, because with worktrees I had ports conflicts, agents conflicting for browser use, or touching external files.
It copies your project, skills, subscription oauth, and can optionally carry external files, supports LFS, and adds Chrome and screensharing via webVNC, so I use it for Godot game for eg.
To use:
agentbox claude # launch on docker, also with pi, codex, opencode\n agentbox hetzner claude # also digitalocean, e2b, vercel, daytona, islo, Sprites, Tenki\n agentbox remote-docker add buildbox marco@10.0.0.9\n agentbox docker:buildbox claude\n\nI use it to drive multiple subagents from a "Firstmate" agent:
I request several changes or gives him a backlog of items and claude will splits them in groups into parallel cloud sandboxes to avoid conflicts, and drive the sessions with the special "agentbox drive" commands, for eg. to wait-for idle, plan-approval, questions etc.
I find it really usefull so far, since it allows me to keep using my existing IDE/tools.
More info on https://github.com/madarco/agentbox","title":"Show HN: Agentbox \u2013 Teleport your repo into sandboxes with no worktree juggling","updated_at":"2026-09-30T13:02:11Z","url":"https://github.com/madarco/agentbox"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"karamfrederic"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"Agentboxd \u2013 Email inboxes for AI agents that treat mail as hostile"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"https://agentboxd.com"}},"_tags":["story","author_karamfrederic","story_49855372"],"author":"karamfrederic","children":[49855373],"created_at":"2026-09-26T11:07:33Z","created_at_i":1790420853,"num_comments":0,"objectID":"49855372","points":2,"story_id":49855372,"title":"Agentboxd \u2013 Email inboxes for AI agents that treat mail as hostile","updated_at":"2026-09-26T11:43:43Z","url":"https://agentboxd.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"handfuloflight"},"title":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["agentbox"],"value":"Agent Box"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"https://agent-box.sh/"}},"_tags":["story","author_handfuloflight","story_48815504"],"author":"handfuloflight","created_at":"2026-07-07T09:34:56Z","created_at_i":1783416896,"num_comments":0,"objectID":"48815504","points":1,"story_id":48815504,"title":"Agent Box","updated_at":"2026-07-07T09:38:37Z","url":"https://agent-box.sh/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"uptownhr"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"Show HN: Agent Box \u2013 Instant Sandbox VM for Claude Code(Macs)"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"https://github.com/Zabaca/agent-box"}},"_tags":["story","author_uptownhr","story_46891064","show_hn"],"author":"uptownhr","children":[46891118],"created_at":"2026-02-04T20:19:22Z","created_at_i":1770236362,"num_comments":0,"objectID":"46891064","points":1,"story_id":46891064,"title":"Show HN: Agent Box \u2013 Instant Sandbox VM for Claude Code(Macs)","updated_at":"2026-03-05T23:31:47Z","url":"https://github.com/Zabaca/agent-box"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rcarmo"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"Agentbox: Contain your coding agents (literally)"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"https://github.com/rcarmo/agentbox"}},"_tags":["story","author_rcarmo","story_46648444"],"author":"rcarmo","created_at":"2026-01-16T16:42:11Z","created_at_i":1768581731,"num_comments":0,"objectID":"46648444","points":1,"story_id":46648444,"title":"Agentbox: Contain your coding agents (literally)","updated_at":"2026-03-05T23:26:02Z","url":"https://github.com/rcarmo/agentbox"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"danoandco"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"Hey HN, we're Willy and Dan, co-founders of Twill.ai (https://twill.ai/). Twill runs coding CLIs like Claude Code and Codex in isolated cloud sandboxes. You hand it work through Slack, GitHub, Linear, our web app or CLI, and it comes back with a PR, a review, a diagnosis, or a follow-up question. It loops you in when it needs your input, so you stay in control.
Demo: https://www.youtube.com/watch?v=oyfTMXVECbs
Before Twill, building with Claude Code locally, we kept hitting three walls
1. Parallelization: two tasks that both touch your Docker config or the same infra files are painful to run locally at once, and manual port rebinding and separate build contexts don't scale past a couple of tasks.
2. Persistence: close your laptop and the agent stops. We wanted to kick off a batch of tasks before bed and wake up to \nPRs.
3. Trust: giving an autonomous agent full access to your local filesystem and processes is a leap, and a sandbox per task felt safer to run unattended.
All three pointed to the same answer: move the agents to the cloud, give each task its own isolated environment.
So we built what we wanted. The first version was pure delegation: describe a task, get back a PR. Then multiplayer, so the whole team can talk to the same agent, each in their own thread. Then memory, so "use the existing logger in lib/log.ts, never console.log" becomes a standing instruction on every future task. Then automation: crons for recurring work, event triggers for things like broken CI.
This space is crowded. AI labs ship their own coding products (Claude Code, Codex), local IDEs wrap models in your editor, and a wave of startups build custom cloud agents on bespoke harnesses. We take the following path: reuse the lab-native CLIs in cloud sandboxes. Labs will keep pouring RL into their own harnesses, so they only get better over time. That way, no vendor lock-in, and you can pick a different CLI per task or combine them.
When you give Twill a task, it spins up a dedicated sandbox, clones your repo, installs dependencies, and invokes the CLI you chose. Each task gets its own filesystem, ports, and process isolation. Secrets are injected at runtime through environment variables. After a task finishes, Twill snapshots the sandbox filesystem so the next run on the same repo starts warm with dependencies already installed. We chose this architecture because every time the labs ship an improvement to their coding harness, Twill picks up the improvement automatically.
We\u2019re also open-sourcing agentbox-sdk, https://github.com/TwillAI/agentbox-sdk, an SDK for running and interacting with agent CLIs across sandbox providers.
Here\u2019s an example: a three-person team assigned Twill to a Linear backlog ticket about adding a CSV import feature to their Rails app. Twill cloned the repo, set up the dev environment, implemented the feature, ran the test suite, took screenshots and attached them to the PR. The PR needed one round of revision, which they requested through Github. For more complex tasks, Twill asks clarifying questions before writing code and records a browser session video (using Vercel's Webreel) as proof of work.
Free tier: 10 credits per month (1 credit = $1 of AI compute at cost, no markup), no credit card. Paid plans start at $50/month for 50 credits, with BYOK support on higher tiers. Free pro tier for open-source projects.
We\u2019d love to hear how cloud coding agents fit into your workflow today, and if you try Twill, what worked, what broke, and what\u2019s still missing."},"title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: Twill.ai (YC S25) \u2013 Delegate to cloud agents, get back PRs"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://twill.ai"}},"_tags":["story","author_danoandco","story_47720418","launch_hn"],"author":"danoandco","children":[47721827,47721911,47721952,47722067,47722143,47722251,47722489,47722894,47723053,47723090,47723506,47723944,47723947,47724244,47724470,47724685,47724790,47726441,47726505,47726513,47727181,47727825,47728129,47728452,47728671,47729237,47729698,47729750,47729916,47729997,47730075,47730110,47730687,47731560,47734885,47739774],"created_at":"2026-04-10T16:22:13Z","created_at_i":1775838133,"num_comments":95,"objectID":"47720418","points":77,"story_id":47720418,"story_text":"Hey HN, we're Willy and Dan, co-founders of Twill.ai (https://twill.ai/). Twill runs coding CLIs like Claude Code and Codex in isolated cloud sandboxes. You hand it work through Slack, GitHub, Linear, our web app or CLI, and it comes back with a PR, a review, a diagnosis, or a follow-up question. It loops you in when it needs your input, so you stay in control.
Demo: https://www.youtube.com/watch?v=oyfTMXVECbs
Before Twill, building with Claude Code locally, we kept hitting three walls
1. Parallelization: two tasks that both touch your Docker config or the same infra files are painful to run locally at once, and manual port rebinding and separate build contexts don't scale past a couple of tasks.
2. Persistence: close your laptop and the agent stops. We wanted to kick off a batch of tasks before bed and wake up to \nPRs.
3. Trust: giving an autonomous agent full access to your local filesystem and processes is a leap, and a sandbox per task felt safer to run unattended.
All three pointed to the same answer: move the agents to the cloud, give each task its own isolated environment.
So we built what we wanted. The first version was pure delegation: describe a task, get back a PR. Then multiplayer, so the whole team can talk to the same agent, each in their own thread. Then memory, so "use the existing logger in lib/log.ts, never console.log" becomes a standing instruction on every future task. Then automation: crons for recurring work, event triggers for things like broken CI.
This space is crowded. AI labs ship their own coding products (Claude Code, Codex), local IDEs wrap models in your editor, and a wave of startups build custom cloud agents on bespoke harnesses. We take the following path: reuse the lab-native CLIs in cloud sandboxes. Labs will keep pouring RL into their own harnesses, so they only get better over time. That way, no vendor lock-in, and you can pick a different CLI per task or combine them.
When you give Twill a task, it spins up a dedicated sandbox, clones your repo, installs dependencies, and invokes the CLI you chose. Each task gets its own filesystem, ports, and process isolation. Secrets are injected at runtime through environment variables. After a task finishes, Twill snapshots the sandbox filesystem so the next run on the same repo starts warm with dependencies already installed. We chose this architecture because every time the labs ship an improvement to their coding harness, Twill picks up the improvement automatically.
We\u2019re also open-sourcing agentbox-sdk, https://github.com/TwillAI/agentbox-sdk, an SDK for running and interacting with agent CLIs across sandbox providers.
Here\u2019s an example: a three-person team assigned Twill to a Linear backlog ticket about adding a CSV import feature to their Rails app. Twill cloned the repo, set up the dev environment, implemented the feature, ran the test suite, took screenshots and attached them to the PR. The PR needed one round of revision, which they requested through Github. For more complex tasks, Twill asks clarifying questions before writing code and records a browser session video (using Vercel's Webreel) as proof of work.
Free tier: 10 credits per month (1 credit = $1 of AI compute at cost, no markup), no credit card. Paid plans start at $50/month for 50 credits, with BYOK support on higher tiers. Free pro tier for open-source projects.
We\u2019d love to hear how cloud coding agents fit into your workflow today, and if you try Twill, what worked, what broke, and what\u2019s still missing.","title":"Launch HN: Twill.ai (YC S25) \u2013 Delegate to cloud agents, get back PRs","updated_at":"2026-07-29T20:35:58Z","url":"https://twill.ai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"bleevht"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["agentbox"],"value":"So, I have been experimenting with Hermes inside my home lab and I decided to build out a dedicated AI agent network that allows easily controlling the traffic that is allowed out of that network and to other parts of my home lab.
I created a dedicated VLAN (999) for my agent. I am currently using a Palo Alto PA-440 that handles my routing and firewall policies. Unifi is carrying my Layer 2 traffic. Then, I created the new network as a new Proxmox SDN VNET.
The interesting part is the traffic controls. Currently I have it setup where my management workstation can connect TO the AI agent box, but not the other way around. Eventually I want to give Hermes limited access to services like Netbox, Proxmox, Kubernetes, etc.
Curious how many of you are isolating self-hosted AI agents and controlling how traffic is sent out into the rest of your home lab."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"[dead]"}},"_tags":["comment","author_bleevht","story_50005452"],"author":"bleevht","comment_text":"So, I have been experimenting with Hermes inside my home lab and I decided to build out a dedicated AI agent network that allows easily controlling the traffic that is allowed out of that network and to other parts of my home lab.
I created a dedicated VLAN (999) for my agent. I am currently using a Palo Alto PA-440 that handles my routing and firewall policies. Unifi is carrying my Layer 2 traffic. Then, I created the new network as a new Proxmox SDN VNET.
The interesting part is the traffic controls. Currently I have it setup where my management workstation can connect TO the AI agent box, but not the other way around. Eventually I want to give Hermes limited access to services like Netbox, Proxmox, Kubernetes, etc.
Curious how many of you are isolating self-hosted AI agents and controlling how traffic is sent out into the rest of your home lab.","created_at":"2026-10-08T13:14:58Z","created_at_i":1791465298,"objectID":"50005453","parent_id":50005452,"story_id":50005452,"story_title":"[dead]","updated_at":"2026-10-08T13:18:54Z"}],"hitsPerPage":8,"nbHits":2496,"nbPages":125,"page":0,"params":"query=agentbox&hitsPerPage=8&advancedSyntax=true&analyticsTags=backend","processingTimeMS":29,"processingTimingsMS":{"_request":{"queue":1,"roundTrip":14},"fetch":{"query":21,"scanning":5,"total":28},"total":29},"query":"agentbox","serverTimeMS":31}