> ## Documentation Index > Fetch the complete documentation index at: https://docs.nuphos.ai/llms.txt > Use this file to discover all available pages before exploring further. # Agent runtime > Connect your own terminal coding agent to a workspace-isolated Nuphos runtime. Nuphos is the DevOps platform around your agent. It provides infrastructure context, scoped credentials, tools, memory, approvals, transcripts, and team collaboration; the agent and its model access remain yours. ## Supported agents Nuphos's runtime boundary is the terminal coding agents supported by [OpenAB](https://github.com/openabdev/openab). Those agents retain their native vendor harness; Nuphos does not require the agent itself to implement a Nuphos MCP interface. | Agent | Availability | How it runs | | ------------------------------------------- | ----------------------------------------------------- | ---------------------------------------------------------------------------------------- | | **Claude Code** | Available now in workspace settings | Its native Claude Code harness, bridged by [OpenAB](https://github.com/openabdev/openab) | | **Codex, Gemini CLI, Kiro CLI, and others** | Added to Nuphos as their runtime adapters are enabled | Each vendor's native harness, bridged by [OpenAB](https://github.com/openabdev/openab) | The [remote MCP server](/connectors/mcp) is a separate interface that lets an MCP client hand work to a Nuphos workspace. MCP compatibility is not what determines whether Nuphos can host a terminal coding agent. ## Connect Claude Code You need a Claude Code account and the Claude CLI on a machine where you are signed in. Run: ```bash theme={null} claude setup-token ``` Copy the generated token. The token uses your existing Claude account and subscription; it is not a Nuphos API key. In the desktop app, open **Settings → Workspace → Agent**. Under **Claude Code accounts**, paste the token and choose **Bind token**. Only a workspace **Administrator** can bind or remove a token. Other members can see that an account is connected, but only a masked token and available account metadata are shown. Nuphos provisions the workspace runtime automatically. This normally takes about a minute. Start using Agent when **Runtime status** changes to **Online**. ## What Nuphos adds to the runtime For each conversation, Nuphos gives Claude Code a scoped set of capabilities: * the cloud credentials selected for that conversation; * Kubernetes configuration assembled from the selected accounts and clusters; * built-in Nuphos skills and skills authored by your team; * APIs for infrastructure, observability, databases, knowledge, triggers, and scheduled tasks; * shared team memory and resource context; and * plan creation, human decisions, approvals, and links back into Nuphos. Tool calls, reasoning, plans, and answers stream back into the shared Nuphos transcript. Conversations stay attached to the same runtime session and can resume after a backend or runtime restart. ## Will this get my Claude Code account blocked? Nuphos is designed to avoid that restriction: it does not turn a Claude subscription into a raw API or replace Claude Code with a third-party harness. [OpenAB](https://github.com/openabdev/openab) is a bridge between a terminal coding agent and a conversation surface such as Slack, Discord, or Telegram. Nuphos occupies the same position as one of those surfaces. The prompt and response pass through the bridge, while the request is still run by Claude Code itself using Anthropic's CLI, authentication, tool loop, and harness. This differs from the OpenClaw pattern restricted by Anthropic. That restriction prevents a third-party harness from using Claude subscription authentication to bypass Claude Code's harness and consume the subscription like an API. Nuphos keeps the vendor's harness intact and adds Skills, MCP servers, team memory, credentials, and DevOps context around it. Providers ultimately control their terms and enforcement, but Nuphos does not depend on the subscription-auth bypass that this restriction was designed to stop. ## Credentials and isolation Each workspace gets its own managed runtime. A conversation is authorized with a short-lived token bound to that user, team, and session; it cannot use that token against another Nuphos route or conversation. Cloud access remains subject to both the account's member allow-list and the credentials selected for the session. Nuphos mints or retrieves credentials on demand instead of copying every connected credential into the runtime. See [Access control](/security/access-control) and [IAM trust model](/security/iam-trust). The Claude Code token is encrypted at rest. Removing it retires the managed runtime and prevents new agent turns until an administrator binds an account again. ## Usage and billing Claude Code inference is provided through the Claude account you bind. It does not consume Nuphos agent credits and does not require a paid Nuphos plan. Your usage and any model charges are governed by your agreement with Anthropic. The same principle applies to other terminal coding agents supported by [OpenAB](https://github.com/openabdev/openab): model execution stays inside each vendor's native harness. See [Plans & billing](/billing) for Nuphos workspace limits. ## Troubleshooting * **Setup required** — ask a workspace Administrator to generate and bind a `claude setup-token` token. * **Starting** — provisioning is still in progress; wait about a minute and keep the settings page open for the status to refresh. * **Unreachable** — remove and rebind the token if it was revoked, then contact [hello@nuphos.com](mailto:hello@nuphos.com) if the runtime does not recover.