Skip to main content

The Infrastream AI Agent

The native Vertex AI integration is a massive competitive differentiator for Infrastream v2. The Infrastream Agent acts as an expert-level, automated Platform Engineer embedded directly within your infrastructure.

This section details how the agent operates, its strict security boundaries, and how it democratizes cloud operations across different roles in your organization.

1. The Vision (For Business Leaders & CTOs)​

Democratizing Cloud Operations safely.

The Infrastream Agent allows a broad, non-technical audience (like Product Managers or Data Scientists) to express their infrastructure intent in natural language. For example, a user can simply type: "I need a highly-available Postgres database for the new AstraPay project."

Crucially, the AI does not modify infrastructure directly. It operates under a strict "Human-in-the-Loop" workflow:

  1. The Agent interprets the natural language prompt.
  2. It queries the live Spanner state to understand the current topology.
  3. It generates the exact YAML Manifests required, ensuring that Pvotal's elite "blueprints" and corporate security guardrails are respected.
  4. It submits these Manifests as a Pull Request (PR) to the Git repository.

A human engineer must review and approve the PR before the Engine provisions anything. This guarantees that AI-generated infrastructure is structurally sound, cost-effective, and perfectly compliant with your SOC2/PCI standards.

2. Using the Agent (For Developers & DevOps)​

Interacting with the Agent in the Portal.

Developers can interact with the Infrastream Agent directly inside the Cloud Portal UI. The Agent is not just a code generator; it is an active participant in your deployment lifecycle.

  • Writing Manifests: Use the chat interface to ask the Agent to generate complex boilerplate. Example: "Write a DeploymentConfig for the 'billing-service' Application, scaling from 2 to 10 instances, with the DB_PASSWORD injected from Secret Manager."
  • Automated Troubleshooting: Because the Agent has "read-only" access to the cloud environment, you can ask it to diagnose live issues. If a deployment stalls, ask the Agent: "Why did the AstraPay database fail to provision?" The Agent will cross-reference the Engine's RunEvents with live GCP Cloud Logging to pinpoint the exact IAM or quota failure.
  • Security & Performance Audits: You can prompt the Agent to analyze your current state. Example: "Are there any Virtual Machines in the staging environment that are not attached to a secure subnet?"

3. Architecture (For Engineering Evaluators)​

How it works under the hood: Spanner & Vertex AI.

The Infrastream Agent is not a generic LLM wrapper. It is deeply integrated into the platform's core architecture using Retrieval-Augmented Generation (RAG) built directly on top of Google Cloud Spanner.

  • First-Class Embeddings: Every time a resource is provisioned or updated, the Engine computes a high-dimensional vector embedding of its configuration (StateEmbedding) and description (DescriptionHash) using Vertex AI, and stores it directly in the Spanner State Ledger.
  • Topological Awareness: When prompted, the Agent performs semantic similarity searches directly within Spanner. Because the embeddings are stored alongside the graph edges (StateDependencies), the Agent understands the "blast radius" of its suggestions. It knows exactly which Applications depend on which Databases, and which IAM policies bind them together.
  • Semantic Bridging: The Agent utilizes SearchContext and SearchLinks metadata (auto-generated by the Go Runners) to seamlessly bridge the gap between abstract logical manifests (like ApplicationSet) and the physical cloud APIs (like google.run.v2.Service), allowing it to reason about infrastructure exactly like a senior human architect.