Forge vs Gemini Enterprise Agent Platform
How Forge compares to Google’s Gemini Enterprise Agent Platform (formerly Vertex AI Agent Builder) — open Agent Skills standard, model neutrality, and cross-cloud portability versus the Gemini-first GCP agent platform.
TLDR
Gemini Enterprise Agent Platform is Google Cloud’s managed platform for building and deploying AI agents — the rebranded and expanded successor to Vertex AI Agent Builder, announced at Cloud Next 26 in April 2026. It pairs the open-source Agent Development Kit (ADK) with the Gemini Enterprise managed runtime (Agent Engine), Workspace Studio for no-code building, and the Model Garden of 200+ models. The platform is optimized for Gemini-powered agents that ground in Google’s search infrastructure and integrate with the GCP ecosystem (BigQuery, Cloud Run, Workspace).
Forge is an open-source agent runtime built on Anthropic’s Agent Skills standard, distributed as a single static binary. It runs the same SKILL.md agent on any container infrastructure, with any LLM provider, with no GCP dependency.
Gemini Enterprise optimizes for “Gemini-grounded agents on GCP.” Forge optimizes for “agents that run anywhere, on any model.”
Different Problems, Different Architectures
Gemini Enterprise Agent Platform is built around two strong Google assets: Gemini as the reasoning model, and Google’s grounding infrastructure (web search, Google Workspace, BigQuery) as the knowledge layer. The platform’s value compounds when an agent’s primary job is to reason over Google-indexed information, query BigQuery, or operate over Workspace content. For organizations already invested in Gemini, BigQuery, and GCP-native data products, the integration is the point.
Forge does not try to be a grounding platform. It does not have an opinion about search providers, vector stores, or knowledge bases — those are skills you compose into your agent, drawn from whichever provider matches your data reality. What Forge does have an opinion about is the runtime contract: a portable, hardened, single-binary execution environment with computed trust, egress enforcement, and an open agent format, deployable to any infrastructure.
The architectural difference is whether the runtime is shaped around a specific cloud’s knowledge stack or shaped to be neutral about it.
Feature Comparison
| Capability | Forge | Gemini Enterprise Agent Platform |
|---|---|---|
| Where the managed runtime runs | Any container runtime — GCP, AWS, Azure, on-prem, air-gapped | GCP only (Agent Engine) |
| SDK portability | N/A — runtime is the binary | ADK is open-source and “deploy anywhere” (Cloud Run, GKE, own infra); managed Agent Engine is GCP-only |
| Skill / agent format | SKILL.md — open Agent Skills standard | ADK code (Python, TypeScript, Go, Java, Kotlin) or Workspace Studio no-code config |
| Cloud lock-in | None | GCP project, region, billing, IAM (for managed platform) |
| Identity | Pluggable — any IdP or none | Google Cloud IAM, Workforce Identity Federation |
| Model choice | Anthropic, OpenAI, Gemini, Ollama, local GGUF | Gemini family + Model Garden (200+ models incl. Claude Opus 4.7, Llama, Mistral, Gemma) — Gemini-first in practice |
| Grounding / knowledge | Composable via skills (any provider) | Google Search, Vertex AI Search, BigQuery, Workspace native |
| Network isolation | Container + per-skill egress allowlists | VPC Service Controls, Private Service Connect |
| Audit logging | Structured logs with correlation IDs | Cloud Logging, Cloud Audit Logs |
| A2A protocol | Native (JSON-RPC 2.0) | Native (A2A v1.0 default at GA) |
| MCP protocol | Native | Native (via ADK) |
| Distribution surfaces | CLI, Docker, K8s, Slack, Telegram, A2A, custom | Workspace, Dialogflow, Cloud Run, Agent Engine, custom |
| Pricing model | Open source — pay for infrastructure you already own | Per-agent pricing (post-rebrand) + Gemini tokens + grounding queries + GCP infrastructure |
| On-prem / air-gapped | First-class — single binary, local inference via forge brain | Managed platform not supported; ADK can self-host on own infra but loses Agent Engine, Workspace Studio, and grounding integrations |
| What you write | A SKILL.md markdown file | ADK code + agent configuration, or Workspace Studio no-code |
The Model Question
Gemini Enterprise supports models from the Model Garden — over 200 of them, including Anthropic’s Claude Opus 4.7, Meta’s Llama family, Mistral, Gemma, and open-weight models. In documentation, this reads as model neutrality. In practice, the platform’s strongest features — grounding with Google Search, Workspace integration, Vertex AI Search integration — are designed around Gemini’s tool-calling patterns and Gemini’s native multimodal capabilities. The ecosystem assumption is that Gemini is the primary model and others are alternatives.
This is not necessarily wrong; Gemini is a genuinely strong model and tight coupling lets the platform optimize. But for organizations whose model strategy is multi-provider (Anthropic for code, OpenAI for general reasoning, Gemini for grounded queries, local models for sensitive data), Gemini Enterprise is a less natural fit.
Forge takes model neutrality as a starting design constraint. Every embedded skill in the Forge runtime works with Anthropic, OpenAI, Gemini, or local GGUF inference (forge brain) — the choice is per-deployment, not per-platform. A skill that uses Anthropic for code review and Gemini for grounded search is a normal Forge agent.
The Knowledge and Grounding Difference
Where Gemini Enterprise is most differentiated is its grounding layer. Google’s web index, Workspace content, Vertex AI Search, and BigQuery are all directly accessible as grounding sources, with citation back to the source documents. For agents whose primary task is to reason over enterprise-internal documents stored in Drive or to query operational data in BigQuery, the integration is genuinely deep — and no other hyperscaler can match it.
Forge does not compete with this. Forge’s design treats knowledge sources as skills — a Tavily search skill, an internal-document-search skill, a BigQuery skill (if you write it). The runtime doesn’t have an opinion about which one you use. If your knowledge lives in Google’s stack and your agents are grounding-heavy, Gemini Enterprise’s depth is real value. If your knowledge lives elsewhere (Confluence, Notion, internal wikis, multi-database environments), Forge’s “compose your own knowledge skills” approach is more flexible.
The ADK Detail Worth Knowing
The Agent Development Kit (ADK) is open source and explicitly designed to “deploy anywhere” — Python, TypeScript, Go, Java, or Kotlin code that you can containerize and run on your own infrastructure, on Cloud Run, on GKE, or on the managed Agent Engine. That means the SDK doesn’t lock you to GCP.
What does lock you to GCP is the rest of the platform: the Agent Engine managed runtime, Workspace Studio, the Model Garden integration, grounding to Google Search and BigQuery, and the per-agent pricing model that came with the Gemini Enterprise rebrand. An ADK agent self-hosted outside GCP is a perfectly valid pattern — but you’ve now opted out of most of what makes the managed platform a managed platform.
This is similar to how Microsoft Agent Framework (the SDK used by Hosted agents in Foundry) can technically run outside Azure but loses Foundry’s integration depth in the process.
Security and Compliance Model
Gemini Enterprise inherits GCP’s enterprise security primitives: VPC Service Controls for data exfiltration prevention, Private Service Connect for private networking, Cloud KMS for encryption, Cloud Audit Logs for compliance, and Workforce Identity Federation for external identity integration. For GCP-native organizations with mature security tooling, these primitives are well-integrated and the operational story is clean.
Forge’s security model travels with the runtime instead of being supplied by the cloud. Computed trust evaluates each SKILL.md before execution. Per-skill egress allowlists are enforced at the runtime layer. Denied tools are explicit contracts in frontmatter. Audit logs are structured and correlation-tagged. The model works identically on GCP, AWS, Azure, or on-prem — useful when your security and compliance team needs to apply the same controls across environments.
The two models address the same concerns through different abstractions: cloud-native security primitives versus runtime-native security contracts.
When to Use Gemini Enterprise Agent Platform
Gemini Enterprise is the right choice when:
- Your organization is GCP-native and intends to remain so.
- Your agent’s primary reasoning model is Gemini, with grounding in Google Search, Workspace, or Vertex AI Search.
- Your knowledge base lives in BigQuery, Drive, or other Google-native stores.
- You want tight integration with Google Cloud IAM, VPC Service Controls, and Cloud Audit Logs.
- Your distribution surface is Workspace, Dialogflow, or Cloud Run.
- You prefer a managed runtime with no operational burden beyond agent code.
- You’re comfortable with per-agent pricing and the broader Gemini Enterprise consumption model.
When to Use Forge
Forge is the right choice when:
- Your infrastructure includes non-GCP clouds, on-prem, or air-gapped environments.
- Your model strategy is multi-provider — Anthropic, OpenAI, Gemini, and local models in the same agent fleet.
- Your knowledge sources are heterogeneous (Confluence, Notion, internal wikis, multiple databases).
- Your identity stack is not Google IAM, and Workforce Identity Federation is not a default option.
- You want an open standard for the agent format itself (SKILL.md / Agent Skills) for portability across runtimes.
- You need on-prem or air-gapped deployment with local model inference.
- You prefer a single-binary, open-source runtime to a managed cloud service.
The Bottom Line
Gemini Enterprise Agent Platform is Google’s best execution of “agents that get the most out of Gemini and GCP.” For organizations whose strategy aligns with that — GCP-native infrastructure, Gemini as primary model, Google’s grounding stack as the knowledge layer — it is a strong, integrated choice, and the April 2026 rebrand makes it a more coherent enterprise product than the predecessor Vertex AI Agent Builder.
Forge is the open-source, model-neutral, cloud-neutral runtime for organizations whose reality doesn’t align with any single hyperscaler’s stack. It does not try to win the grounding-and-knowledge fight; it tries to be the best portable runtime for the agent format itself.
The most honest summary: Gemini Enterprise is GCP’s answer to “Gemini-powered agents grounded in Google data.” Forge is the answer to “production agents that don’t pick a model or a cloud.”
Read the full positioning: Forge is the runtime for agents that run next to services →
Also see: Forge vs AgentCore · Forge vs Foundry · Forge vs Claude Managed Agents