Solo.io Pricing: Open-Source Gateways and Enterprise Scope

Solo.io pricing is quote-led for commercial scope, and the product boundary matters more than a guessed license number. Solo's pricing page says there is no single price list because environments and use cases differ. The current portfolio includes agentgateway for LLM, MCP, and agent traffic, while kgateway handles general API traffic. Both expose open-source paths, with enterprise software and support layered around specific deployments.

TL;DR

• Solo.io asks buyers to request a tailored estimate rather than publishing one universal enterprise rate card.
• agentgateway targets LLM, MCP, and agent traffic as a Linux Foundation open-source project.
• kgateway is a CNCF Sandbox API gateway under Apache 2.0, with enterprise extensions and support available separately.
• Separate software and support from infrastructure and operations. Keep telemetry, migration, and any model-provider charges separate too.

Solo.io pricing starts with the product boundary

The Solo.io pricing page states that pricing depends on the environment and use case. It asks buyers to submit details for a tailored estimate. The page names Solo Enterprise for agentgateway and Istio, plus kagent and agentregistry in its description, while the product menu places kgateway under cloud connectivity.

A useful request begins by naming the product. "Solo gateway" can refer to materially different control planes and traffic types. agentgateway targets AI-native protocols and model traffic. kgateway targets Kubernetes-native API routing across conventional application protocols. Istio belongs to service-mesh scope. Combining them in one unqualified line prevents procurement from tying cost to an architectural owner.

I would send back any Solo quote request that says only "gateway platform." The noun is too broad to price and too broad to operate.

agentgateway covers LLM and MCP traffic

Solo's agentgateway product page describes a Rust-based gateway for LLM, MCP, and agent traffic. The project is open source and hosted by the Linux Foundation. The page names model routing and observability, plus inline guardrails and centralized provider keys. Rate limits and token-based cost controls are also included.

Build the quote from expected model requests and token volume, providers, and MCP servers. Add agent identities and tool authorization, then specify telemetry exports and the environments where the gateway runs. State which enterprise functions and support terms the organization expects Solo to provide around the open-source data plane.

The Gloo AI Gateway transition article explains why current naming matters. Older Gloo AI Gateway material now points toward agentgateway, while Gloo Gateway became kgateway. New orders should use current names and identify the exact version and product family.

kgateway remains a general API gateway

The kgateway product page describes a self-managed, Envoy-based API gateway and a CNCF Sandbox project. It states that the open-source project uses the Apache 2.0 license and that enterprise extensions are available. The same page presents Solo Enterprise for kgateway as the path for support and hardened images, plus additional enterprise functions.

Kgateway's commercial scope should follow clusters and routes, plus traffic policies, environments, and support requirements. It can route REST and gRPC, plus GraphQL and WebSocket traffic. Other application traffic also fits this general API gateway category, which extends beyond an AI-native LLM or MCP gateway.

Keep the products separate even when they share an infrastructure team. A diagram with a blue kgateway box in front of application APIs and a green agentgateway box in front of model and MCP routes will prevent a surprising amount of confusion during vendor review.

Open source removes one fee and creates other costs

Open-source availability can remove a proprietary software license from one deployment path. It leaves infrastructure and ownership intact. Kubernetes capacity and network egress still need budgets. So do telemetry storage and secrets. Upgrades and vulnerability response need budgets, along with backups and on-call coverage. A self-managed gateway also needs tested rollback and recovery procedures.

Enterprise software and support can shift some of that risk through supported releases and hardened images. They can also provide response commitments and vendor engineering access. Ask Solo to identify which benefits attach to each commercial product. Require the quote to separate subscription or support fees from professional services and infrastructure operated by the customer.

The AI gateway architecture guide maps the components that usually sit around a gateway. Use it as a checklist, then assign a named team to every database and control plane. Assign the certificates and policy store too, along with every telemetry destination in the selected Solo topology.

Build a quote that can survive renewal

Send Solo an environment sheet before requesting a number. For agentgateway, include LLM providers and MCP clients and servers. Add model routes and agent identities, then state expected traffic and guardrail requirements. Include observability destinations. For kgateway, include Kubernetes clusters and application routes. Add protocols and policy needs, then specify upgrade windows.

Ask the response to state:

• The exact Solo Enterprise products and open-source projects covered.
• Licensed units and included capacity, plus treatment above any limit.
• Support hours and response commitments, plus supported versions and the security-fix policy.
• Hosted and customer-operated components, plus required infrastructure.
• Onboarding and migration charges, plus training and professional-services charges.
• Contract term and renewal assumptions, plus product-name transition protections.

Model inference belongs on a separate line because the chosen model provider sets that price. Telemetry storage may also sit with another vendor. A Solo quote should never absorb costs it does not control.

DeepInspect

DeepInspect governs authenticated HTTP traffic deliberately routed between users or agents and LLM endpoints. It evaluates application-supplied identity and role, plus data classification and destination under versioned policy, then produces a per-decision record before forwarding allowed traffic.

agentgateway can continue to handle model and MCP routing plus telemetry. It can also handle keys and traffic controls. Kgateway can continue to route general application APIs. DeepInspect's scope stays at the authenticated model request boundary, so each layer can be priced against its actual job. Book a technical deep dive at deepinspect.ai.