Polygraph serves as a meta-harness designed to maximize agentic autonomy for developers. It was built by the Nx team to address the limitations agents face when working within the confines of a single repository. While agents are highly capable in isolated environments, they often struggle with cross-repository tasks and fail to maintain context across sessions. Polygraph resolves these issues by enabling agents to work effectively across an entire organization, treating multiple repositories as a unified dependency graph without requiring a complex migration to a single monorepo.

Functionally, Polygraph acts as an infrastructure layer that sits alongside existing AI agents such as Claude Code, Codex, and OpenCode. It connects all private and open-source repositories to which a user has access, allowing for comprehensive visibility and orchestration. It coordinates tasks by spawning a swarm of child agents—one for each repository—while a parent agent maintains the overall task context, planning, and orchestration of work across boundaries. The platform records every session, allowing developers to resume work, reference historical traces, or hand off complex tasks to team members seamlessly.

Some of the key features are:

• Cross-Repo Unity: Automatically indexes repositories to build a dependency graph, enabling agents to read and write across multiple projects as if they were a single synthetic monorepo.
• Session Memory & Resumability: Captures full session history, including PRs, logs, and traces, so work can be resumed at any time by any agent without re-explaining context.
• PR & CI Coordination: Manages the creation, linking, and status tracking of pull requests across multiple repositories, ensuring CI checks are verified across the entire change set.
• OSS Integration: Allows public open-source repositories to be included in sessions for debugging, reproduction, and integration validation without treating them as owned code.
• Multiplexed Operation: Supports terminal multiplexers to view live logs of multiple child agents side-by-side during active development.
• Downstream Testing: Facilitates validation of library changes against consumers using pack-and-copy workflows to test fixes before they reach a registry.

Operationally, Polygraph is used through a CLI tool that developers install to interact with their environment. Users authenticate via the CLI, which handles the necessary OAuth flows, and then start sessions where they can explicitly select relevant repositories or let the system dynamically discover them based on the task description. The agents leverage Polygraph's MCP tools to perform actions like pushing branches, creating PRs, and running CI checks. The platform is designed to respect existing repository permissions while providing the necessary visibility for agents to perform complex, multi-repo updates autonomously.

Some common use cases include:

• Ship feature across repos: Implementing a new feature that requires simultaneous updates to both backend and frontend microservices.
• Ship library change to consumers: Validating a design system or library update across all dependent applications before merging the changes.
• Patch a CVE across every affected repo: Automatically finding all repositories affected by a vulnerability, generating patch PRs, and tracking them through to resolution.
• Ask questions across repos: Performing cross-repo investigations, impact analysis, or pattern transfers when reading documentation and code is the primary deliverable.
• One session for the whole sprint: Managing an entire sprint's worth of progress in a single long-lived session, ensuring no context is lost during handoffs or task switching.