Why Delego

Both Anthropic and OpenAI now offer cloud-hosted coding agents — but running your work through a cloud agent comes with trade-offs that Delego sidesteps by keeping execution on your machine.

Cloud agents restrict your choices. Codex Cloud, for example, does not let you select which model to run — the platform decides for you. Claude Code on the web similarly ties execution to whatever configuration Anthropic exposes at that point in time. With Delego, the model is just a Linear label you set per job. You can run opus on one issue and sonnet on the next, without changing any configuration.

Cloud agents run in an isolated sandbox, not your environment. They can't reach private repositories without explicit permission grants, can't use your local SSH keys, and can't access services running on your internal network. A Delego runner inherits your entire shell environment — if it works in your terminal, it works through Delego.

Cloud agents introduce account risk. Heavy automated use through a cloud platform can trigger rate limits or policy reviews for unusual activity patterns. Delego runs as your own authenticated CLI session on your own hardware. The provider sees normal terminal usage — because that's exactly what it is.

Cloud agents are another tool to log into. Delego lives inside Linear. You delegate an issue, watch it progress in the same issue view, and review the PR without opening a separate tab. The full history of what the agent did stays attached to the issue permanently, searchable alongside everything else in your Linear workspace.

No vendor lock-in. Cloud coding agents tie you to a single provider. Claude Code in the cloud runs Claude; Codex Cloud runs OpenAI models. With Delego, you choose the agent and model per job using a Linear label — you can run Claude Code on one issue and Codex on the next, picking whichever model fits the task best at that moment.

Parallel execution. You can pair multiple runners to a single Delego account — on the same machine or across multiple machines — and they pull from the same job queue simultaneously. The more runners you pair, the more jobs run in parallel.