The agent runtime built for the enterprise reality
Multi-cloud. On-prem. Air-gapped. Regulated. Forge runs the same SKILL.md agent on EKS, AKS, GKE, on-prem Kubernetes, bare metal, or air-gapped infrastructure — with computed trust, egress enforcement, and audit logging built into the runtime, not bolted on as an enterprise tier.
What "enterprise" actually means for an agent runtime
The agent runtime market has produced excellent products for desktop knowledge work, for AWS-native deployment, for Azure-native deployment, for GCP-native deployment, and for hosted sessions in Anthropic's cloud. What it has not produced — and structurally cannot produce, because each major vendor's incentive is to keep workloads in their own cloud — is a runtime that runs anywhere your services already run without picking a cloud, a model, or a managed-service dependency. That gap is where Forge lives.
The infrastructure you actually have, not the one a vendor would prefer. Enterprise deployments span multiple clouds, include on-prem, frequently include air-gapped environments for regulated workloads, and never sit cleanly on a single hyperscaler. A runtime that requires AWS, requires Azure, or requires Anthropic's cloud isn't a runtime — it's a cloud commitment. Forge runs on any container infrastructure and on bare metal, with no managed-service dependency.
The security model your security team can actually review. Computed-not-declared trust. Per-skill egress allowlists enforced at the runtime layer. Denied tools as explicit contracts in skill frontmatter. Audit logs with correlation IDs at every level. The autowire pipeline scans every skill before it executes. These aren't enterprise add-ons; they're how the open-source runtime works. When your security team asks "how do we prevent this agent from doing things we didn't authorize," the answer doesn't require a separate paid product.
The compliance posture your auditors will actually accept. Forge Hub adds signed skills, RBAC, and an on-prem registry for organizations under SOC 2, ISO 27001, HIPAA, FedRAMP, or other frameworks. Audit logs stream to your existing SIEM. Identity binds to the IdP you already operate. State persists wherever you already store state. No vendor-specific compliance dependency is created.
The cost predictability your CFO will actually sign off on. Forge is open source. You pay for the infrastructure you already operate and the model API spend you already budget. No consumption-priced managed-runtime fee, no per-session billing surprises, no cloud-vendor lock-in on agent execution. At scale, this is the difference between an agent program that survives a budget review and one that doesn't.
Where Forge fits in the enterprise agent runtime landscape
Forge does not try to be every agent runtime. The honest carve, made explicitly:
Hyperscaler-managed runtimes
If your stack is 100% AWS, AgentCore is excellent. If you're Microsoft-native with Entra and M365, Foundry is integrated for that. If you're GCP-native with Gemini-grounding, Gemini Enterprise is the deepest fit. Forge is for enterprises that span clouds, run on-prem, or need air-gapped deployment.
Anthropic-hosted runtimes
Claude Managed Agents is the right runtime for event-driven workloads that run in Anthropic's cloud on Anthropic models with MCP-reachable tools. Forge runs the same SKILL.md format in your environment, on your model choice, with direct access to systems that can't be exposed via public MCP.
Agent frameworks
LangChain, CrewAI, and other frameworks give you Python abstractions to build agents. Forge gives you a runtime to deploy them. The two are not competitors; LangChain logic can live inside a SKILL.md script.
Personal agent tooling
Claude Cowork and similar desktop assistants run on a knowledge worker's laptop for personal productivity. Forge runs as a service in your infrastructure, for your organization. Different blast radius, different security model.
Three audiences. One document. One runtime.
Enterprise agent programs fail when they're built by developers in isolation. The operational expertise lives elsewhere — in the SREs who write runbooks, the security engineers who document controls, the compliance officers who codify procedures. Forge is designed so all three audiences can collaborate through the SKILL.md document.
Domain experts who author skills
If you write runbooks today, you can write Forge skills today. SKILL.md is markdown with structured frontmatter — the same prose you use in operational documentation, with declarations for what tools and integrations the skill needs. The Skill Builder UI in forge ui is designed for non-developer authors. No Python, no framework.
Developers who write the tools
When LLM reasoning isn't enough, developers contribute scripts (in any language) that skills invoke as tools. The skill body declares the tools by name; the developer implements them; the domain expert iterates on the procedure without touching code. Clear separation, version-controlled together.
Platform engineers who operate the runtime
A single static Go binary. Standard Kubernetes manifests. Egress enforcement at the runtime layer. Audit logs through OpenTelemetry to your existing SIEM. Local model inference via forge brain for air-gapped deployments. The runtime is infrastructure you already know how to operate.
In production
Forge is in production deployment with enterprise customers in robotics, logistics, fintech, and platform engineering. Customer case studies are in publication — contact us if you'd like a direct reference.
For the full architectural argument and a representative production scenario, read the long-form positioning post.
What Forge Hub adds for enterprise
The core Forge runtime is open source under Apache 2.0 and includes the security model, egress enforcement, computed trust, and audit logging by default — not gated as enterprise features. Forge Hub adds the controls enterprise procurement and security teams require to deploy at scale. Items marked preview or contact us are in active development or available on request.
| Capability | Open-source Forge | Forge Hub Enterprise |
|---|---|---|
| Runtime | Full single-binary runtime, all execution features | Same — no runtime gating |
| Egress enforcement | Per-skill allowlists, runtime-enforced | Same |
| Computed trust | Autowire pipeline, security analyzer | Same |
| Audit logging | Structured logs with correlation IDs | Same, plus SIEM-ready streaming connectors (preview) |
| Skill registry | Local skills, community registry | Private organizational registry, on-prem option (contact us) |
| Skill signing | Cryptographic verification supported | Signed skill distribution, key management (preview) |
| RBAC | Filesystem permissions | Workspace-driven RBAC, fine-grained roles (contact us) |
| Model inference | Pluggable model providers, `forge brain` local | Same |
| Support | Community via Discord and GitHub | SLA-backed enterprise support (contact us) |
| Compliance | SOC 2 / ISO 27001 audit-ready posture | Compliance documentation, attestations, controls evidence (contact us) |
| Deployment | Self-deploy on any infrastructure | Reference architectures, deployment assistance (contact us) |
Ready to deploy in production?
The Forge runtime is open source. Enterprise support, signed skills, and the on-prem registry are available through Forge Hub.