# AI use cases for Agentbox (devilcoolyue/agentbox) Repo: [devilcoolyue/agentbox](https://github.com/devilcoolyue/agentbox) | Use cases: 4 | Page: https://dedupedai.com/use-cases/devilcoolyue-agentbox/ Agentbox is a self-hosted browser workspace that runs Claude Code and Codex CLI in Docker on a Linux server you control, with chat, terminal, files, Git, and usage tracking. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Run Claude Code in a browser workspace on your server: https://dedupedai.com/use-cases/devilcoolyue-agentbox/run-claude-code-in-browser-on-own-server/ - Review agent changes per file and open a PR: https://dedupedai.com/use-cases/devilcoolyue-agentbox/review-agent-changes-and-open-pr/ - Share agent accounts with a team and track usage: https://dedupedai.com/use-cases/devilcoolyue-agentbox/share-agent-accounts-and-track-team-usage/ - Let a workspace reach private services through abox-link: https://dedupedai.com/use-cases/devilcoolyue-agentbox/reach-private-services-from-workspace/ ## How to run Claude Code in a browser workspace on your server with Agentbox Repo: [devilcoolyue/agentbox](https://github.com/devilcoolyue/agentbox) | Difficulty: Intermediate | Time: 60 minutes | Basis: From the README Install Agentbox on a Linux server and code with Claude Code from any browser. **Who it is for:** A developer with a spare Linux server who wants to pick up a coding project from a laptop or phone without setting up tools on each device. **The problem:** Claude Code needs a configured environment, and it has to be set up again on every machine you work from. **The outcome:** Claude Code runs inside a Docker workspace on your server, and you chat with it and see its work from a browser. ### Steps 1. Run `curl -fsSL https://raw.githubusercontent.com/devilcoolyue/agentbox/main/install.sh | sudo bash` on the server (the installer downloads a package and starts Agentbox.service). 2. Open http://YOUR_SERVER_IP:8180 and sign in as `boxadmin` with the generated password printed by the installer. 3. In System settings, open Account pool, add a Claude account, and fetch its models. 4. Create a workspace from an empty project, an uploaded archive, or a Git repository, and select the Claude account. 5. Describe the task in Chat, then check the result in Files and Changes. ### Prompt to paste into your agent ```text Use the agentbox repo at github.com/devilcoolyue/agentbox to set up a Claude Code workspace on the server . Confirm that agentbox.service is running and that the web console answers on port 8180, then report the workspace URL. Do not share the generated admin password in logs. ``` ### Caveat Each workspace defaults to 2 GiB of memory and 2 CPUs, so size the server for the number of workspaces you run at once. Risk, remote script: The installer is piped from the internet straight into sudo bash, so read it first if your policy requires it. Risk, credentials: The Claude subscription, API key, or relay credential you add is stored in Agentbox on your server. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/devilcoolyue-agentbox/run-claude-code-in-browser-on-own-server/ ## How to review agent changes per file and open a PR with Agentbox Repo: [devilcoolyue/agentbox](https://github.com/devilcoolyue/agentbox) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Check each diff the agent made, then commit, push, and open a PR or MR from the browser. **Who it is for:** A developer who lets Codex CLI or Claude Code edit a Git repository and wants to review every change before it leaves the server. **The problem:** Agent edits land in the working tree, and it is easy to commit or push changes you have not read. **The outcome:** You review per-file diffs in the workspace, commit the ones you accept, and open a PR or MR from the same place. ### Steps 1. Create a workspace from your Git repository and select a Codex or Claude account. 2. Describe the change in Chat and let the agent edit files. 3. Open Changes and review the diff for each file. 4. Configure a Git connection with an HTTPS token, SSH, or GitHub or GitLab OAuth. 5. Commit the accepted changes, preview the push, then push and open the PR or MR. ### Prompt to paste into your agent ```text Using the agentbox repo at github.com/devilcoolyue/agentbox, run a workspace on against the repository . Ask the agent to make the change , then list every changed file and its diff summary before any commit. Verify that nothing is pushed until the diffs are approved. ``` ### Caveat The README describes the diff and PR flow but does not say how the agent is prevented from editing files before you review them, so review the Changes panel before you commit. Risk, credentials: The HTTPS token, SSH key, or OAuth grant you connect lets the workspace push to your remote repository. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/devilcoolyue-agentbox/review-agent-changes-and-open-pr/ ## How to share agent accounts with a team and track usage with Agentbox Repo: [devilcoolyue/agentbox](https://github.com/devilcoolyue/agentbox) | Difficulty: Advanced | Time: 60 minutes | Basis: From the README Pool Claude and Codex accounts, scope them per user, and export usage costs as CSV. **Who it is for:** A team lead who gives several trusted developers access to Claude or Codex and needs to know who used which model and what it cost. **The problem:** Shared agent accounts are hard to track, and usage for each request is rarely visible to the person paying. **The outcome:** Each developer uses only the accounts you scoped to them, and you read per-request usage with price snapshots and export it to CSV. ### Steps 1. In System settings, open Account pool and add the subscription, API key, or relay accounts your team will share. 2. Set the models and reasoning effort each account offers, and limit each account to the users who may use it. 3. Give team members sign-in access and have them create their own workspaces with the accounts they can use. 4. Open Usage to filter per-request records and check tokens, cache, latency, and price snapshots. 5. Export the filtered usage records as CSV for your cost review. ### Prompt to paste into your agent ```text Use the agentbox repo at github.com/devilcoolyue/agentbox on the server to scope the account pool to the users . After a test request, open the Usage view and confirm that a record appears with tokens and a price snapshot, then export it as CSV. ``` ### Caveat The README says model usage follows your provider's terms and pricing, so the usage records show what Agentbox recorded, not your provider's invoice. Risk, account pooling: Sharing one subscription across several people may conflict with your provider's terms, so check them before you pool accounts. Risk, credentials: Pooled API keys and relay accounts are stored on the server and are visible to whoever administers Agentbox. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/devilcoolyue-agentbox/share-agent-accounts-and-track-team-usage/ ## How to let a workspace reach private services through abox-link with Agentbox Repo: [devilcoolyue/agentbox](https://github.com/devilcoolyue/agentbox) | Difficulty: Advanced | Time: 60 minutes | Basis: From the README Pair abox-link so an agent can use allowlisted repositories, databases, or services on your network. **Who it is for:** An engineer whose agent needs a private repository, database, or internal service that is not reachable from the public internet. **The problem:** A hosted agent cannot reach services on your private network, so it cannot test against them. **The outcome:** The agent's workspace reaches only the allowlisted services on your network through abox-link. ### Steps 1. Open the Private-network tunnel page in the console and check abox-link status. 2. Download the abox-link client for your machine from the same page. 3. Pair the client with Agentbox using the pairing flow shown in the console. (inferred, check the README) 4. Add each service you want to expose to the allowlist, and keep the list as short as the task needs. 5. Ask the agent to connect to one allowlisted service and confirm the connection works. ### Prompt to paste into your agent ```text Using the agentbox repo at github.com/devilcoolyue/agentbox, pair abox-link on and add only to the allowlist. Ask the workspace agent to connect to that service and report the response status. Confirm that a non-listed address is refused. ``` ### Caveat The README does not describe the pairing steps in detail, so follow the console and the linked documentation for that part. Risk, credentials: Any service you allowlist becomes reachable by agent workspaces, so use read-only accounts where you can. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/devilcoolyue-agentbox/reach-private-services-from-workspace/