One task, many agents: switching runtimes without losing history

The agent landscape changes monthly. Locking your workflow to one runtime means re-platforming every time a better model or CLI ships. We designed Agentis around adapters so that the task, its history and its governance never depend on which agent executed it.

The adapter contract

An adapter receives the full task context — description, project, repo, attachments, accumulated knowledge — and streams back a uniform event log: messages, tool calls, questions, artifacts, tokens. The platform does not care whether the runtime is Claude Code, OpenCode or a custom container in your cluster.

This is also what makes environment policy enforceable: approved models, allowed environments and permission boundaries are configured per project once, and every adapter inherits them. Switching runtimes is a dropdown, not a migration.