As enterprises deploy AI agents across Claude, Cursor, ChatGPT, and custom workflows, the need for centralized governance has created a rapidly evolving AI gateway market. TrueFoundry has positioned itself as a comprehensive ML platform with gateway capabilities, but understanding its pricing structure requires examining both the published tiers and the factors that influence total cost. With per-agent identity becoming critical for enterprise security teams, platform buyers must evaluate not just sticker prices but total cost of ownership across their AI infrastructure.

This article breaks down TrueFoundry's 2026 pricing structure, compares it against alternative approaches, and outlines the key factors that influence total costs for enterprise AI gateway deployments.

Key Takeaways

• TrueFoundry's published pricing starts at $0 for Developer (up to 3 users, 10K requests per user per month), while Pro costs $25 per user per month with 20K requests per user, pooled across the team, plus $20 per additional 100K requests
• TrueFoundry was named a Gartner Representative Vendor for AI Gateways in February 2026, reflecting growing market recognition
• Request-based pricing models can create cost forecasting challenges for agentic workloads, as multi-step agents generate multiple billable gateway requests per workflow
• Shadow AI detection remains a critical gap in gateway-only architectures that see only routed traffic, not local agent activity on developer machines
• TrueFoundry supports Virtual Account Tokens for service identities and now documents Agent Identity as a beta capability for registered agents
• Enterprise buyers should evaluate deployment flexibility, with air-gapped and VPC options adding significant value for data-sensitive workloads

Understanding TrueFoundry's AI Gateway: What It Is and Key Features

TrueFoundry operates as a broader AI/ML platform that includes gateway functionality alongside model serving, fine-tuning, and GPU orchestration. The platform has grown significantly since its 2021 founding, culminating in recognition as a Gartner Representative Vendor for AI Gateways in February 2026 and the acquisition of Seldon AI in June 2026.

Core Gateway Capabilities

• Support for 1,600+ models across major providers including OpenAI, Anthropic, and self-hosted options
• Virtual MCP abstraction for Model Context Protocol server management
• Per-agent budgets, quotas, RBAC, and centralized access controls
• Guardrails with multi-engine support for content filtering
• OpenTelemetry and Prometheus integration for observability
• Request-level audit logging for compliance documentation

Authentication and Authorization

TrueFoundry supports multiple MCP authentication modes, including per-user OAuth, client credentials, identity-provider tokens, token passthrough, and a documented Okta token-exchange/OBO configuration. Agent Identity is currently beta, with broader per-hop token exchange still rolling out. SSO and SCIM provisioning are available on Enterprise.

Agent Identity Considerations

TrueFoundry supports Virtual Account Tokens for service and application identities and now documents Agent Identity as a separate beta capability for registered agents. Agent Identity allows an agent to remain independently identifiable while acting as itself or, in supported flows, on behalf of another principal. Buyers should evaluate the maturity and delegation model of this beta capability alongside established gateway identity controls.

For organizations prioritizing MCP governance with tool-level access control and SCIM-driven RBAC, purpose-built solutions may offer stronger alignment with security requirements.

TrueFoundry's AI Gateway Pricing Models and Factors (2026 Forecast)

TrueFoundry maintains a transparent pricing structure in the enterprise AI gateway category, with three distinct tiers that scale based on request volume and user count.

Published Pricing Tiers

• Developer: Free tier for up to 3 users, limited model accounts, MCP servers, and Virtual MCPs. Includes 10K requests per user per month with 7-day log retention. Suitable for evaluation and small development teams.
• Pro: $25 per user per month with no published user limit. Each user contributes 20K AI Gateway requests and tool calls per month to a pooled team allowance. Additional usage costs $20 per 100K requests. Includes full MCP Gateway, unlimited Virtual MCPs, guardrails, and 30-day log retention.
• Enterprise: Custom pricing with custom usage limits. Includes SSO, SCIM, audit logs, VPC/on-prem deployment, air-gapped deployment options, data residency controls, and enterprise support. Agent Gateway and Agent Harness are currently custom-priced.

Cost Analysis Reveals Hybrid Pricing Model

TrueFoundry separates seat pricing from gateway traffic. Pro costs $25 per user per month and includes 20K requests and tool calls per user, pooled across the team. Usage above the included allowance costs $20 per additional 100K requests. Enterprise pricing and Agent Gateway pricing are custom.

Potential Cost Drivers for TrueFoundry Users

Request-based pricing creates forecasting challenges. Every API call counts equally against limits regardless of whether it's a simple classification or a complex multi-tool agent workflow. TrueFoundry counts each LLM request and MCP tool call toward gateway usage. Multi-step agents can generate multiple billable gateway requests within a single workflow, making request volume an important input to cost forecasting.

Per-user costs scale linearly. Organizations adding users to Pro pay $25 per additional seat, with pooled request allowances growing proportionally. Teams exceeding the pooled allowance should budget for overage charges.

Infrastructure costs for self-hosted deployments. Organizations choosing VPC or on-premise deployment operate under Enterprise custom pricing, with infrastructure responsibilities varying by deployment model.

Mintive's pricing structure uses custom quotes based on team size and requirements, with per-user licensing for active AI agent users and platform fees that scale with usage and team size.

Evaluating TrueFoundry's AI Gateway for Machine Learning Platforms

TrueFoundry's heritage as an ML platform shapes its gateway approach. The June 2026 Seldon AI acquisition added predictive ML serving capabilities, positioning the platform for organizations that need model deployment alongside gateway functionality.

Strengths for ML Teams

• Integrated model serving with vLLM, TGI, and Triton support
• Fine-tuning workflows within the same platform
• GPU orchestration for training and inference workloads
• Unified usage and cost observability across model operations and gateway traffic, with model-provider charges billed separately

Considerations for Governance-Focused Buyers

The platform scope creates adoption surface area. Organizations that only need an LLM proxy or MCP gateway are adopting a platform with significant functionality beyond their core requirements. This platform breadth can introduce complexity considerations.

Key Considerations for ML Teams

Evaluate whether you need model serving or just governance. If your primary requirement is securing MCP connections and managing agent access to enterprise tools, a dedicated MCP Gateway may provide better fit than a full ML platform.

Consider the deployment model. TrueFoundry offers managed SaaS as well as VPC/on-prem deployment on Enterprise. Customer-operated deployments use Kubernetes-based infrastructure and require more internal platform expertise than the managed SaaS option.

Assess vendor consolidation benefits. Organizations already using TrueFoundry for model serving may benefit from adding gateway functionality within the same platform.

Comparing AI Gateway Costs: TrueFoundry vs. Other Approaches

The AI gateway market spans multiple pricing models, from open-source self-hosted options to managed SaaS platforms. Understanding the full cost picture requires examining software licensing, infrastructure, implementation, and ongoing operational expenses.

Total Cost of Ownership Considerations

Software licensing models vary widely. TrueFoundry Pro uses per-user pricing with pooled request allowances and overage charges. Enterprise and Agent Gateway pricing are custom. Mintive uses custom quotes based on team size, usage, and requirements. Open-source projects may have no software licensing fee but require customer-operated infrastructure.

Infrastructure responsibility depends on deployment model. Managed SaaS generally shifts more infrastructure responsibility to the vendor, while VPC or self-hosted deployments require additional customer-operated infrastructure and platform work. The CISA Zero Trust guidance recommends evaluating data residency and network isolation requirements when selecting deployment models.

Implementation and operational costs compound over time. Self-hosted deployments transfer more platform maintenance, security patching, upgrades, and infrastructure operations to internal teams. The required staffing depends on the deployment architecture and existing platform capabilities.

Published pricing does not support fixed Year 1 TCO comparisons across deployment models. Buyers should model software, usage, infrastructure, implementation, and internal operating costs using their own workload assumptions and vendor quotes.

Factors Influencing Total Cost of Ownership

Usage growth in agentic workloads. TrueFoundry counts each LLM request and MCP tool call toward gateway usage. Multi-step agents can generate multiple billable gateway requests within a single workflow, making request volume an important input to cost forecasting.

Compliance documentation costs. Platforms without tamper-evident audit trails may require additional tooling or manual processes to satisfy regulatory requirements, adding indirect costs.

Mintive's LLM Gateway provides per-team budgets and model restrictions with cost tracking by user, agent, model, and tool, enabling chargeback-grade visibility.

Shadow AI Detection: A Key Differentiator in AI Gateway Solutions

Gateway architectures that only see routed traffic face a fundamental visibility gap. When developers install Claude Code or Cursor on their machines, AI activity occurs locally without passing through centralized governance infrastructure.

Why Shadow AI Poses a Risk

Ungoverned tool access. Local AI clients can read sensitive files, execute commands, and access MCP servers without audit logging. A developer using an unapproved MCP server for database access creates compliance exposure that gateway-only monitoring cannot detect.

Credential exposure. AI coding assistants may read environment files, SSH keys, and API credentials during normal operation. Without visibility into file reads, security teams cannot identify when sensitive data enters AI context windows.

Compliance gaps. Regulatory frameworks such as the NIST AI Risk Management Framework emphasize comprehensive audit trails. Organizations cannot attest to controls they cannot observe, creating audit risk for local AI usage.

Independent analysis confirms that most platforms only capture traffic routed through their infrastructure, missing substantial AI activity occurring on developer endpoints.

Mintive's Approach to Detecting Unsanctioned AI Activity

Mintive's Agent Monitor extends visibility beyond gateway traffic through lightweight hooks for supported AI clients including Claude Code, Cursor, and Codex.

What Agent Monitor captures:

• File reads including environment files, SSH keys, and configuration files
• Commands executed via bash, git, and package managers
• MCP tool calls from supported agent environments, including off-gateway MCP activity where supported
• Prompts submitted to local AI clients
• Token usage and cost by model, user, and session

Enforcement capabilities:

• Rules can flag, block, or require user confirmation for risky operations
• MDM-pushed configurations enable detect-and-block mode on developer machines
• SIEM export via OTLP or Splunk HEC maintains compliance documentation

For teams evaluating shadow AI detection, this cross-platform visibility represents a significant architectural difference from gateway-only approaches.

Advanced Governance: Per-Agent Identity and Programmable Security Policies

Enterprise AI deployments increasingly involve autonomous agents operating alongside human users. CI/CD pipelines, scheduled summarizers, on-call bots, and background coding agents all require access to enterprise systems. The governance question becomes: how do you know which agent did what?

The Shared Credential Problem

Most gateway platforms authenticate agents using shared service-account tokens or repurposed user API keys. This approach creates several issues:

• Audit logs collapse agent activity into a single identity
• Credential rotation affects all agents sharing the token
• Over-privileged access results from using broad service accounts
• Revoking one compromised agent requires rotating shared credentials

Securing AI Agents: The Mintive Bundle Advantage

Mintive's Agent Gateway addresses these challenges through first-class agent identities with the Agent Bundle primitive.

Per-agent identity management:

• Each agent receives its own OAuth 2.0 client credentials (client ID and secret)
• Agents authenticate as distinct principals in the authorization model
• Credentials can be rotated or revoked independently without affecting other agents
• The "act as agent" OAuth flow delegates user permissions to agents without exposing raw tokens

Scoped tool access:

Agent Bundles package together a curated set of MCP tools appropriate for the agent's role, custom security policies governing what the agent can access, independent audit trails attributing every action to the specific agent, and credential lifecycle management with automatic refresh handling.

Authentication flexibility:

• Bearer keys for simple deployments with named, expiring credentials
• M2M tokens via OAuth client-credentials exchange for short-lived access
• Workload identity federation allowing Kubernetes service accounts or cloud IAM roles to authenticate directly

This architecture enables enterprises to scale from 10 to 100+ agents while maintaining the attribution that compliance and security teams require.

Programmable Security Policies

Mintive's Mint Guard and Gateway Middleware provide layered runtime protection:

• Mint Guard offers managed detection for prompt injection, credentials/secrets, PII, and harmful content
• Rules enable declarative matching on tool names, arguments, and content patterns
• Gateway Middleware runs customer-authored JavaScript for DLP integrations, external classifiers, and custom policy enforcement

Pre-built templates support integration with AWS Bedrock Guardrails, GCP Model Armor, Microsoft Purview, Nightfall, and Skyflow, keeping policy enforcement within existing security tooling.

Deployment Flexibility: SaaS, VPC, and Self-Hosted AI Gateway Options

Enterprise buyers face different deployment requirements based on data sensitivity, regulatory obligations, and infrastructure preferences. The AI gateway market offers varying levels of flexibility.

TrueFoundry Deployment Options

• Managed SaaS for standard deployments
• VPC deployment with self-hosted control plane in customer Kubernetes
• Air-gapped deployment via forward proxy for isolated environments

TrueFoundry's documented air-gapped support addresses highly regulated industries, though implementation requires Kubernetes expertise.

Considerations for Enterprise Deployments

Data residency requirements. Organizations subject to GDPR, data localization laws, or industry-specific regulations must understand where data flows and persists. SaaS deployments should offer regional options for US and EU at minimum.

Network isolation needs. Financial services, healthcare, and defense contractors often require air-gapped or VPC deployments that never expose traffic to public internet paths.

Operational complexity trade-offs. Self-hosted deployments provide maximum control but transfer operational burden to internal teams. Organizations should assess their SRE capacity before committing to self-managed infrastructure.

Mintive provides managed SaaS with US and EU regional availability, plus VPC and self-hosted deployments on request for enterprises with specific sovereignty requirements. Private network tunnels enable secure connectivity to on-premise databases and internal APIs without inbound firewall rules.

Why Enterprise Governance Enables Safe AI Agent Adoption

The pace of AI agent deployment has outstripped governance infrastructure at most organizations. Developers install Claude Code and Cursor on their machines, connect to databases via MCP servers, and build autonomous workflows before IT and security teams establish controls.

The Meta-Problem Driving Governance Demand

• No telemetry: Organizations don't know which tools agents call or what data they read/write
• Scattered credentials: API keys end up on developer laptops; one leak exposes the entire kingdom
• No access control: Agents reach any connected system with no oversight
• Missing audit trails: Makes it harder to produce evidence for SOC 2, HIPAA-related controls, and internal risk reporting
• Config sprawl: Every developer configures MCP servers locally, creating N points of failure

The Cost of Ungoverned AI

• Compliance challenges. Auditors increasingly ask about AI usage. Organizations without comprehensive audit trails face findings and remediation costs.
• Security incidents. Agents operating with over-privileged credentials or connecting to unvetted MCP servers create attack surface that security teams cannot monitor.
• Shadow AI proliferation. Without governed alternatives, developers route around slow approval processes by using personal accounts and unsanctioned tools.
• Operational inefficiency. Per-machine configuration means duplicated setup effort, inconsistent tool availability, and troubleshooting complexity.

Mintive: Purpose-Built Governance for Enterprise AI Agents

Mintive provides governance infrastructure that makes AI agents deployable, governed, measurable, and swappable. Its platform combines identity management, access controls, monitoring, and runtime security.

Key capabilities include:

• Centralized governance: SSO, SCIM-driven RBAC, Virtual MCPs, tool-level policies, and audit trails.
• Per-agent identity: Agent Bundles provide independent OAuth 2.0 credentials, scoped permissions, and audit attribution without shared service accounts.
• Cross-platform monitoring: Agent Monitor captures supported local AI activity, including file access, commands, and off-gateway MCP calls.
• SIEM integration: OpenTelemetry support enables audit data export to existing security and observability systems.

Customer results demonstrate the impact:

• Coursera: Rolled out governed AI to 2,000 employees within weeks using Virtual MCPs to simplify setup.
• Modern Treasury: Reports saving 30 minutes to 4 hours per support case through AI-powered bank integration lookups.

Organizations can evaluate Mintive through a 7-day free trial with full feature access before committing to an enterprise agreement.

Frequently Asked Questions

How does TrueFoundry's current Pro pricing work?

TrueFoundry Pro costs $25 per user per month with no published user limit. Each user contributes 20K AI Gateway requests and tool calls per month to a pooled team allowance, and additional usage costs $20 per 100K requests. Enterprise pricing is custom, while Agent Gateway and Agent Harness are also currently custom-priced.

What should I evaluate when comparing AI gateway vendors?

Enterprise buyers should evaluate deployment flexibility, authentication models, agent identity support, audit trail completeness, shadow AI detection capabilities, and total cost of ownership. Request vendor-specific quotes using your workload assumptions rather than relying on published list prices for custom-priced tiers.

Can I migrate from TrueFoundry to another AI gateway without significant rework?

Migration complexity depends on your integration depth. Organizations using TrueFoundry primarily for gateway functionality face medium migration effort, requiring translation of configurations to alternative policy formats and reconfiguration of client endpoints. The timeline depends on integration depth and deployment requirements.

How do I handle agents that need different access than their operators?

Agents should operate with their own credentials and scoped permissions rather than inheriting their creator's access. Mintive's Agent Bundles provide each agent with independent credentials, curated tool access, and audit attribution. This enables least-privilege scenarios and service-account access without individual human exposure.

What compliance documentation should I expect from an AI gateway vendor?

Enterprise buyers should evaluate SOC 2 Type II audit status, penetration testing, data residency, sub-processor transparency, and applicable healthcare controls. When a vendor will handle protected health information as a business associate, BAA availability should also be verified. Mintive is SOC 2 Type II audited, is compliant with HIPAA standards, and signs BAAs with customers handling protected health information.