A Coding Agent That Keeps Your Files Local and Your Model Swappable
Most ways of running an AI coding agent push in one of three directions: your proprietary code goes to a third-party API, you are locked into a single model provider, or you pay per-seat subscriptions regardless of how much you use them. AWS's machine-learning blog frames its answer as a way around those constraints for teams with data-residency requirements, cost-sensitive workloads, or a need to move between models. The agent is OpenCode, an open source, terminal-native coding assistant built in Go. Paired with open weight models on Amazon Bedrock, AWS says, it gives you an assistant that runs locally while inference happens securely within your own AWS account. The split is the interesting part: the piece that reads and edits files and runs shell commands sits on your machine, and only the model call leaves it. That is a data-governance argument more than a capability argument, and it comes from the company selling the inference.
AI-generated discussion · Different perspectives
The pitch is about where the data goes, not what the agent can do. Residency and cost are the whole product here.
Both travel together, though. The agent's file editing and shell access stay local, while the model layer can move — the separation is what makes swapping models possible at all.