RecourseOS

RecourseOS is a consequence verification platform designed for autonomous systems operating on live infrastructure. Its primary function is to analyze proposed changes—such as those initiated by Terraform plans, shell commands, or MCP tool calls—by evaluating the potential impact, recoverability, and risk before actions are executed. The system aims to prevent destructive or unrecoverable mutations by enforcing a verification step that assesses the blast radius, checks for protective mechanisms like deletion protection and backups, and issues cryptographically signed attestations for every decision.

The platform supports over 175 resource types across AWS, GCP, and Azure, applying deterministic rules for known resources and semantic classification for unknowns. If evidence is insufficient or the risk is high, RecourseOS escalates decisions for human review rather than granting automatic approval. Its evaluation engine normalizes all mutations into a standard MutationIntent, which is processed to generate a machine-readable ConsequenceReport. Each report is signed with Ed25519, ensuring auditability and verifiability by external parties using the public key.

RecourseOS integrates natively with MCP servers, command-line interfaces, and continuous integration pipelines. It is deployable wherever agents execute, including environments such as Terraform Cloud, kubectl, shell, and MCP, with multi-region enforcement for global coverage. The platform offers SDKs and integrations for tools like LangChain, LlamaIndex, Go, GitHub Actions, Pulumi, AWS Lambda, Kubernetes Admission Controller, and more. Notifications can be sent to Slack, Discord, PagerDuty, and Opsgenie, and developer tools include a CLI, VS Code extension, and Raycast extension.

The intended audience includes teams managing autonomous agents and infrastructure automation, particularly those seeking to enforce safety checks and maintain audit trails for changes in production environments. RecourseOS operates with per-mutation evaluation, requiring no persistent trust, and ensures that every action is independently verified before execution. Its focus on structured, signed, and auditable consequence verification addresses the need for robust safeguards in automated infrastructure management.