Pyrana vs Gemini Enterprise Agent Platform

Architecture

Google's platform and Pyrana sit at different layers. Gemini Enterprise Agent Platform builds, scales, governs, and optimizes agents on Google Cloud. Pyrana is an application platform: identity, authorization, storage, orchestration, approvals, audit, and context built once, with each business application defined as configuration on top. Google's unit is the agent. Pyrana's unit is the application, in which agents are one mechanism.

Google describes the platform as "an evolution of Vertex AI." On April 22, 2026, Vertex AI became Agent Platform, Agent Engine became Agent Runtime, and Agent Engine Sessions and Memory Bank became Agent Platform Sessions and Agent Platform Memory Bank. The platform groups its parts into Build (ADK, Agent Studio, Model Garden), Scale (Agent Runtime, Sessions, Memory Bank), Govern (Agent Registry, Agent Identity, Agent Gateway), and Optimize (evaluation, simulation, observability, prompt optimization).

Pyrana's App configuration names an application's pages, operations, resources, agents, workflows, tool bindings, access profiles, and approval policies. Every call, from a page, the copilot, the API, MCP, a schedule, or an event, passes one eight-step gate before it runs.

Gemini Enterprise, the Gemini agent, and the developer platform

Google separates the employee app from the developer platform. Gemini Enterprise is the app employees use, sold in Business, Standard, Plus, Pay-as-you-go, and Frontline editions. It provides oversight of agents from Google, third parties, and internal teams. Admins register custom agents hosted on Agent Runtime, agents built with A2A, and Dialogflow agents, and employees build their own with the no-code Workflow Builder.

On October 8, 2026, Google Cloud announced the Gemini agent at Gemini at Work, calling it "a universal agent for work." Google says it plans the work, uses skills and tools, connects to business systems, chooses the best model for each job, and includes cost controls and enterprise governance. Coverage reports that it is in private preview, available in the Gemini Enterprise app and Workspace, and routes work across Gemini and Claude models.

The developer platform is where custom agents are built and run. The Gemini Enterprise app is where many of them reach employees. Pyrana plays a different role from both: it is the application where a business process, its records, and its approvals live.

Context and knowledge

Google offers memory and several retrieval services. Pyrana curates knowledge as governed claims with evidence. Sessions hold state within an interaction, and Memory Bank carries context across sessions. RAG Engine, Vector Search, and Agent Search retrieve over indexed content. Pyrana's context engine, cortIQ, stores knowledge as Context Units: a claim, its evidence passage, a type, a classification, and a content-hash identity.

The difference is change control and proof. In cortIQ, classification gates every change: absolute units never auto-change, regulated units wait for a person, tribal units are conflict-checked, and new units supersede old ones rather than overwrite them. Each run accounts for every unit as retrieved, injected, or cited, and citations are content-addressed, so the business context behind an output is provable.

Agent capabilities

Google offers a full agent toolkit centered on ADK. Pyrana offers structure for agents working inside an application. Agent Runtime gives ADK full integration, provides managed templates for LangChain, LangGraph, AG2, and LlamaIndex, and accepts CrewAI and custom frameworks through a custom template. Model Garden offers Gemini plus Claude, Llama, DeepSeek, Grok, and other models. Pyrana's Harness supports native tool calling, planning, subagents, checkpoints, and workflow fan-out.

In a Pyrana workflow, one run dispatches specialized agents per entity and group in parallel. Each pulls only what its job needs through tools scoped to it, reads what siblings and prior runs published, and publishes back as versioned resources. A model is assigned per agent and is not switched mid-run, and tools hold the query so agents pass only parameters.

Governed actions and approvals

Google governs whether an agent may reach a tool. Pyrana routes consequential calls to people and records exactly what they approved. Agent Gateway blocks connections by default unless an IAM policy grants them, Model Armor screens prompts and responses, and Semantic Governance policies (preview) check tool calls against plain-language business rules and return allow or deny. ADK Tool Confirmation, marked experimental, pauses a tool call for a person.

In Pyrana, a tool approval policy over typed arguments decides which invocations need a person. A tripped policy freezes the operation, exact arguments, scope, and evidence under a hash and queues it to a named approval lane. Reviewer eligibility is checked at decision time, the frozen invocation is revalidated before dispatch, and dispatch is idempotent. A queued request is visibly different from a posted one.

Building the application

On Google's platform, the application around the agent is a separate build. On Pyrana, it is configuration plus a frontend. Agent Runtime exposes deployed agents, and Gemini Enterprise gives them a place in the employee app. The pages, records, review queues, versioned documents, and business rules of an application are built by the team, typically on Cloud Run, Firestore or Cloud SQL, and a frontend framework.

A Pyrana application declares its pages, and the platform hands the agent the exact page context a person is looking at. Outputs are immutable, attributed resource versions; publishing against a stale version is refused. People can ask the copilot for personal pages, compiled without network access or authority and read-only. With clean data available, an application typically takes about a week to stand up.

Governance and security

Google secures agents through cloud identity and network policy; Pyrana authorizes every call as a business operation. Agent Identity issues each agent a SPIFFE-based identity that cannot be impersonated, does not allow long-lived keys, and is logged alongside the user when the agent acts on someone's behalf. Agent Gateway enforces IAM, supports VPC Service Controls, and reads Agent Registry metadata. Pyrana authorizes in layers: tenant, App membership, access profile, scope entitlement, and approval lane.

Pyrana's gate writes the execution row and audit record in one transaction before dispatch and fails closed, so unauthorized things read as absent. Reading a record and approving a change are separate permissions. The user's principal passes through to Databricks and Snowflake row-level security. Pyrana is SOC 2 aligned and supports customer programs under HIPAA, SOX, ISO 27001, and PCI DSS.

Operations

Google operates agents across 23 regions; Pyrana operates applications with a release lifecycle. Agent Runtime runs in 23 Google Cloud regions with us and eu multi-regional endpoints, and Google states these locations provide data-at-rest residency for agent code, sessions, and memories. Memory Bank and Agent Gateway each run in 20 regions. Observability includes the Unified Trace Viewer, and Agent Gateway added Cloud Trace integration in preview in September 2026.

Pyrana's lifecycle covers the whole application: push compiles the App configuration into an immutable version, activate moves the active pointer under audit and requires an App owner, rollback activates a prior version, and deploy ships a frontend image. Workflows run on Temporal workers, so a run waiting for approval survives restarts and deploys. Pyrana runs on Azure, as SaaS on a dedicated private tenant or in the client's own infrastructure for large enterprises.

Cost structure

Google bills platform consumption plus Gemini Enterprise seats; Pyrana is a platform subscription plus implementation. Since June 2026, Google bills Agent Runtime, Code Execution, Sessions, and Memory Bank through three SKUs: Agent Platform Compute at $0.085 per vCPU-hour, Memory at $0.009 per GiB-hour, and Storage at $0.30 per GiB-month. Sessions and Memory Bank moved to these units on September 1, 2026. Model tokens are billed separately, and Gemini Enterprise is licensed per user by edition.

Pyrana's cost is an enterprise platform subscription and implementation, and model usage runs on the client's own keys. Small, medium, and large routing for platform jobs keeps cost in proportion to the task. A first engagement, including SSO, access profiles, the approval policy, and loading the policy library into cortIQ, runs about six weeks.

What it takes to ship the same application

The reference application: a person or agent drafts a purchase request; amounts above a threshold need a finance approver; the request waits durably for the decision; everything is recorded.

Using them together

• Gemini models. Pyrana's Google Gemini adapter lets any Pyrana agent run on Gemini, assigned per agent with the client's own keys.
• Pyrana in Google agents. Each Pyrana App exposes an MCP server. An Agent Runtime agent can reach it through Agent Gateway in egress mode, which applies Google's IAM and content policies, while Pyrana authorizes the call as the user and records it.
• Google agents and tools in Pyrana. Pyrana supports MCP servers and gateways and REST APIs as tool sources, so a Pyrana workflow can call tools and agent endpoints exposed on Google Cloud, with Pyrana's gate and approval policies applied first.
• Workspace content. Pyrana connects to Google Drive as a tool source, so agents can read the documents teams already keep there.

When not to choose Pyrana

• Every component must run inside Google Cloud under your own VPC Service Controls perimeter. Pyrana runs on Azure.
• The goal is an employee assistant across Workspace, with no separate business application. Gemini Enterprise and the Gemini agent are built for that.
• Your developers are committed to ADK and want Google to host, scale, and evaluate their agents with nothing above them.
• The work is model tuning, training, or evaluation research on Google's ML stack rather than a business process with approvals.
• One team is building one agent with no shared users, approvals, or audit needs across applications.