# Cloud coding agents that use your own API key or subscription

Which cloud coding agents accept your own API key, cloud account, or Claude and ChatGPT subscription, who bills model usage for each, and what to ask a vendor.

- Canonical: https://replicas.dev/resources/coding-agents-bring-your-own-key

A cloud coding agent needs a model account to run on. It can use a Claude or ChatGPT subscription you already pay for, an API key or cloud account billed per token, or credits that the agent vendor sells you. This guide lists which cloud coding agents accept each one.

Vendor terms change often, so every claim about a vendor below links to its own page. We checked them all on November 5, 2026.

## Which cloud coding agents accept your own model account

This table covers agents that run tasks in the cloud, not on your computer. In alphabetical order:

| Agent | Claude or ChatGPT subscription | Your own API key or cloud account | Who bills model usage |
|---|---|---|---|
| Claude Code cloud sessions | Yes, a Claude subscription is required: Pro, Max, Ultra, Team, or Enterprise with premium or Chat + Claude Code seats ([Anthropic](https://code.claude.com/docs/en/claude-code-on-the-web)) | No. Cloud sessions need a claude.ai sign-in and aren't available with an API key, a Console sign-in, Bedrock, or Google Cloud ([Anthropic](https://code.claude.com/docs/en/claude-code-on-the-web)) | Anthropic, against your plan's usage limits, with no separate charge for the cloud VM |
| Cursor Cloud Agents | No | Not documented. Cursor's API key page says custom keys only work with chat models, and doesn't mention Cloud Agents ([Cursor](https://cursor.com/help/models-and-usage/api-keys)) | Cursor, from your plan's usage at the selected model's rate ([Cursor](https://cursor.com/docs/cloud-agent)) |
| Devin | No | Not documented ([Devin](https://devin.ai/pricing)) | Devin, from a plan allowance, with extra usage at API pricing |
| Factory | No | In the Droid CLI and desktop app only. Personal custom models don't appear in Factory's hosted web or mobile platforms ([Factory](https://docs.factory.com/model-independence/byok)) | Factory, or your provider for custom models |
| GitHub Copilot cloud agent | No | Not documented. GitHub's BYOK page lists VS Code, JetBrains, Xcode, Copilot CLI, the Copilot app, and the Copilot SDK, not the cloud agent on GitHub.com ([GitHub](https://docs.github.com/en/copilot/concepts/models/bring-your-own-key)) | GitHub, through your Copilot plan ([GitHub](https://docs.github.com/en/copilot/concepts/agents/coding-agent/about-coding-agent)) |
| Google Jules | No. Paid tiers come with Google AI Pro or Ultra, for personal @gmail.com accounts only ([Google](https://jules.google/docs/usage-limits/)) | Not documented | Google, as daily and concurrent task limits per plan |
| OpenAI Codex cloud | Yes, a ChatGPT sign-in is required ([OpenAI](https://learn.chatgpt.com/docs/auth)) | No. OpenAI says API key sign-in supports local Codex work, and features that rely on cloud services are limited or unavailable ([OpenAI](https://learn.chatgpt.com/docs/auth)) | OpenAI, against your ChatGPT plan |
| OpenHands Cloud | No | Yes, on a free tier limited to one user, with daily and parallel conversation limits ([OpenHands](https://www.openhands.dev/blog/bring-your-own-llm-is-now-available-to-all-openhands-cloud-users)) | Your provider, or OpenHands' own LLM provider at cost |
| Replicas | Yes, Claude and ChatGPT subscriptions, up to 10 of each per person or organization ([Replicas](https://docs.replicas.dev/admin/credentials)) | Yes: Anthropic and OpenAI API keys, AWS Bedrock, Microsoft Foundry, OpenRouter, Ollama, Cursor API keys, and OpenAI- or Anthropic-compatible endpoints ([Replicas](https://docs.replicas.dev/admin/credentials)) | Your provider. Replicas charges for workspaces and doesn't resell model usage ([Replicas](/pricing)) |

"Not documented" means the vendor's own pages don't say the agent accepts your key. Ask the vendor before you rely on it.

Two patterns stand out. Anthropic and OpenAI run their own cloud agents only on their own subscriptions, so neither accepts the other's account. Most other vendors bill model usage themselves and don't document a way to bring your own account to their cloud agent.

## Subscriptions, API keys, and vendor credits

A cloud agent gets its model from one of three kinds of account:

| Account | Who bills you | How usage is measured | Limits |
|---|---|---|---|
| Claude or ChatGPT subscription | Anthropic or OpenAI | A flat monthly price per person | 5-hour and weekly usage limits, shared with the person's other Claude or ChatGPT use |
| API key or cloud account (Bedrock, Foundry, Vertex) | The model provider or cloud provider | Per token, at published rates | Rate limits and your own spend limits |
| Vendor credits | The agent vendor | The vendor's own unit, such as tasks, requests, or usage at API rates | The plan's allowance, then overage |

A subscription costs the same no matter how much of the limit you use, so it suits steady daily work. An API key costs nothing when idle and has no 5-hour window, so it suits automations and bursts of parallel tasks. Vendor credits put the model and the agent on one bill, but you pay the vendor's rate and can't use a subscription you already pay for.

Teams often combine them: each engineer uses their own subscription, and a shared API key or cloud account covers automations and anything that runs past a subscription's limit.

## Questions to ask a cloud agent vendor

1. Which model accounts can the cloud agent itself use? Many products accept your key in their IDE or CLI but not in their cloud agent.
2. Can each person use their own subscription, or does everyone share one organization account?
3. When a subscription hits its usage limit mid-task, does the task stop, or move to another account?
4. Where are the keys stored, and who in the organization can read them?
5. Can admins choose which accounts the agent uses, and in what order?
6. Does the vendor add a markup or a fee on model usage?

## Where Replicas fits

Replicas runs Claude Code, Codex, Cursor, OpenCode and Pi in cloud workspaces, on the model accounts you already have. Each person can connect their own Claude or ChatGPT subscription, and an admin can add shared organization accounts and keys. A person's own credentials take priority over the organization's.

When a subscription reaches its usage limit, the chat moves to the next account in the [fallback chain](https://docs.replicas.dev/admin/credentials#fallback-chains) you set, such as a second subscription or an API key, and continues. Replicas charges for workspace time and doesn't mark up model usage. Credentials are passed into the workspaces that use them, and any organization member who can open a workspace can read the keys inside it. A personal key only decides whose workspaces get it, so use a key you're willing to share with your organization. See [Credentials](https://docs.replicas.dev/admin/credentials) for every supported account type.


## FAQ

### Which cloud coding agents let you bring your own API key?
As of November 5, 2026, Replicas and OpenHands Cloud accept your own API key in their cloud agents. Factory accepts custom models in its CLI and desktop app but not its hosted web platform. Claude Code cloud sessions and Codex cloud need a Claude or ChatGPT sign-in, and Cursor, Devin, GitHub Copilot and Jules don't document it for their cloud agents.

### Can I use my Claude subscription with a cloud coding agent?
Yes. Claude Code cloud sessions run on Pro, Max, Ultra, Team and eligible Enterprise plans, and count toward that plan's usage limits. Replicas can also run Claude Code on your Claude subscription, and lets each person connect up to 10.

### Does Codex cloud work with an OpenAI API key?
No. OpenAI says Codex cloud requires signing in with ChatGPT. An API key works for the Codex CLI, IDE extension and desktop app when they run locally.

### Do Cursor Cloud Agents support your own API key?
Cursor doesn't document it. Its API key page says custom keys only work with chat models, and Cloud Agents are billed from your Cursor plan at the selected model's rate.

### Is a subscription or an API key cheaper for a coding agent?
A subscription costs a flat monthly price with 5-hour and weekly usage limits, so it suits steady daily work. An API key is billed per token with no usage window, so it suits automations and bursts of parallel tasks. Many teams use both, with the API key as a fallback when a subscription hits its limit.