TrueFoundry

• Palo Alto Networks could fold the Portkey gateway it now owns into Prisma AIRS at a price that removes the reason to buy a standalone gateway alongside it.
• The major cloud providers host the models and could ship equivalent routing, guardrails and request logging in their own gateways, leaving TrueFoundry selling mainly its on-premises and air-gapped installation.
• The gateway exposes one OpenAI-compatible endpoint for every provider rather than a proprietary interface, so a rival can present the same interface to a customer's applications.
• TrueFoundry's reported revenue could stay far below its reported request volume, which would leave the gateway line dependent on another funding round.
• No named customer account in the reviewed record identifies a security team or CISO as the gateway's budget owner, so the security positioning may not reach a security budget.
• Seldon's customers could keep or choose another control plane, which would leave the acquisition producing little gateway adoption.

TrueFoundry sells the AI Gateway to enterprises whose engineering teams already call several model providers and have no single place to govern those calls. The documentation describes the product as the proxy layer between a customer's applications and the LLM providers and MCP servers they reach. Its page for platform owners says they can centralize AI, MCP and agent interactions through a unified gateway layer, which names the platform team as the audience.

TrueFoundry addresses a second buyer as well. Its page for CISOs argues that controls must extend beyond individual applications once AI systems gain access to enterprise data, tools and workflows, and it offers policy enforcement, an audit trail of prompts and tool calls, and telemetry export into existing SIEM workflows. That page describes a security buyer that the customer accounts in the record do not identify.

The pain itself is described rather than measured. The reviewed sources carry no independent quantification of what fragmented model access, untracked spend or missing audit trails cost an enterprise. TrueFoundry supplies the framing, and its release quotes a Gartner projection about how many teams will adopt this class of product. [s2, s17, s3, s11]

The published capability list is concrete and specific. TrueFoundry's docs describe one OpenAI-compatible endpoint for every provider, drop-in support for the OpenAI and Anthropic SDKs, role-based access control with scoped keys, and guardrails covering personal data, prompt injection, content moderation and custom policies. The product page advertises a catalog of more than 1,600 models.

Two companion products push the same control point past model calls. The MCP Gateway registers MCP servers and applies per-server role-based access control and OAuth 2.0 with dynamic discovery, and the Agent Gateway adds one execution layer where policy governs who can deploy, run or modify an agent. The CISO page describes writing those controls as OPA or Cedar policies over prompts, responses, agent actions and MCP tool calls.

Every performance number belongs to TrueFoundry. The claim of sub-3ms internal latency under enterprise-scale workloads is the company's own measurement, and the reviewed record contains no third-party benchmark, no independent evaluation and no inspectable implementation of the gateway. [s2, s1, s4, s14, s3]

Palo Alto Networks settled part of this question by buying a rival. It closed its acquisition of Portkey on May 29, 2026, saying the deal establishes the AI Gateway as a mission-critical control plane for the enterprise and making Portkey the core AI Gateway for Prisma AIRS, its AI security platform. A security platform is folding that control point into its own product.

TrueFoundry answers with deployment shape and reach. It offers the gateway as a hosted service or installed in a customer's own cloud, on-premises or air-gapped, and says no data leaves the customer's domain. An SDxCentral article carries the company's framing of the Seldon acquisition, under which TrueFoundry operates as a gateway and control plane above model servers such as Seldon's Core, building an independent alternative to the AI tooling bundled with AWS, Google Cloud and Azure.

What the record does not contain is an independent comparison. No independent source measures TrueFoundry's controls against a rival's product, so what TrueFoundry offers a buyer as difference is a deployment option and a documented span across models, tools and agents. [s15, s16, s10, s5, s13, s1]

TrueFoundry sells the gateway from the bottom of the market upward. The pricing page publishes a free tier covering 50,000 requests a month, two paid tiers at published monthly prices, and a custom enterprise tier for organisations running AI at scale with strict compliance. That shape supports self-serve evaluation before any sales conversation.

Named deployments exist and most are the company's own accounts. TrueFoundry's press room names examroom.ai, Aviva Credito and Adopt AI as AI Gateway users, crediting examroom.ai with 60-plus AI solutions governed for 500,000 users. Its FloQast case study, whose title names no product line, reports 15.4 billion tokens processed and more than 80,000 guardrail evaluations at roughly 53ms of overhead. A VentureBeat article names one of the gateway users, stating that Adopt AI routes more than 15 million requests and 40 billion input tokens through the gateway.

Volume and revenue point in different directions. A January 2026 VentureBeat article states that TrueFailover operates as a resilience layer on top of TrueFoundry's AI Gateway, which already processes more than 10 billion requests a month for Fortune 1000 companies. The same article reports more than 30 paid customers and more than $1.5 million in annual recurring revenue for the prior year, with the customer base quadrupling, and it attributes none of the three to a product line. [s5, s1, s7, s8]

The founding team's documented backgrounds span machine learning, infrastructure engineering and quantitative finance. A VentureBeat article states that Nikunj Bajaj, Abhishek Choudhary and Anuraag Gutgutia founded the company in 2021, that all three were Meta engineers, and that they had met as classmates at IIT Kharagpur. The team page adds that Bajaj led machine learning for one of Facebook's flagship conversational AI products and that Choudhary reached Senior Staff Engineer in infrastructure before leaving.

Anuraag Gutgutia brought quantitative-finance experience to the founding team. TrueFoundry says he was a vice president of portfolio management at WorldQuant, applying quantitative models to large funds.

No prior exit, sustained publication record or industry recognition appears for any of the three in the reviewed sources, and the record identifies no security role among them. [s8, s6]

TrueFoundry publishes a trust centre with documents behind it. The page lists a SOC 2 Type II attestation and offers its report, 2025 and 2026 penetration-test reports and a complete policy packet, with the documents behind a request for access. TrueFoundry's product pages claim SOC 2, HIPAA and GDPR standards, and the pricing page lists GDPR and HIPAA ready deployments and certificates among its plan features.

Deployment is where a regulated buyer gets more than a certificate. TrueFoundry offers installation in a customer's own cloud, on-premises or air-gapped and says no data leaves the customer's domain, and its CISO page describes exporting telemetry through OpenTelemetry into Splunk, Datadog, Grafana and existing SIEM workflows.

That assurance package is conventional rather than distinguishing. The record documents no authorisation that only TrueFoundry holds, and the reports behind the trust centre open only on request. [s12, s16, s5, s3]

Add analyzed competitors to compare them side by side with TrueFoundry.

TrueFoundry aims the AI Gateway at enterprises whose engineering teams already call several model providers. Its page for platform owners addresses that team directly, saying they can centralize AI, MCP and agent interactions through a unified gateway layer while enforcing consistent governance. The customer accounts in the record do not say which team bought or operates the gateway.

TrueFoundry names a security segment as well. Its page for CISOs argues that controls must reach beyond individual applications once AI systems gain access to enterprise data, tools and workflows, and lists security workflows such as SOC investigation assistants and triage agents. No customer account in the reviewed record names a security team as the buyer or the operator.

One gateway user named in the record operates in a regulated sector. TrueFoundry describes Aviva Credito, which it names as an AI Gateway user, as a Mexico-based lender. Its wider case-study library adds Innovaccer, a healthcare intelligence cloud handling protected health information, without naming the gateway.

The gateway's documented controls cover the three things a governance buyer asks for. TrueFoundry's docs describe role-based access control with scoped keys, per-user and per-model rate limits with budgets, and guardrails for personal data, prompt injection, content moderation and custom policies, all behind one OpenAI-compatible endpoint.

The documented breadth runs past model calls. The MCP Gateway applies per-server role-based access control and OAuth 2.0 with dynamic discovery to tool calls, the Agent Gateway governs who can deploy, run or modify an agent, and the CISO page describes writing those rules as OPA or Cedar policies over prompts, responses, agent actions and MCP tool calls. One vendor covers model traffic, tool traffic and agent execution.

The performance evidence is the vendor's own. The sub-3ms internal latency figure is TrueFoundry's own measurement, and the reviewed record holds no third-party benchmark, no independent evaluation and no inspectable implementation.

TrueFoundry offers a bottom-up evaluation path alongside an enterprise sales motion. It publishes a free tier covering 50,000 requests a month and two paid tiers at published monthly prices, and reserves the enterprise tier for a conversation with sales.

Named references come almost entirely from the vendor. TrueFoundry's press room names examroom.ai, Aviva Credito and Adopt AI as AI Gateway users, and a VentureBeat article ties just one of them, Adopt AI, to gateway traffic, stating that it routes more than 15 million requests and 40 billion input tokens through it.

The traction picture depends on which number a buyer trusts. A VentureBeat article ties the request volume to the product, stating that TrueFailover operates as a resilience layer on top of TrueFoundry's AI Gateway, which already processes more than 10 billion requests a month for Fortune 1000 companies. The same article reports more than 30 paid customers, more than $1.5 million in annual recurring revenue for the prior year and a quadrupling customer base as company-wide figures, and it attributes none of the three to a product line.

TrueFoundry publishes its prices. The pricing page lists a free developer tier, paid tiers, and a custom enterprise tier for organisations running AI at scale with strict compliance, with the free tier covering 50,000 requests each month and all listed models.

The plan features name the assurances a security buyer would look for. GDPR and HIPAA ready deployments and certificates appear among them, and the enterprise plan TrueFoundry describes for organisations running AI at scale with strict compliance is the one priced by conversation.

Delivery is the customer's choice. TrueFoundry describes the gateway as deployable as SaaS, VPC-hosted, on-premise or air-gapped, so the same software either runs as a hosted service or is installed in a network the customer owns with no data leaving its domain.

That range is the delivery difference. TrueFoundry offers the gateway air-gapped, so the same product reaches a network with no internet connection.

Support is contractual rather than operational. TrueFoundry advertises 24/7 support with SLA-backed response times, and its plan table lists enterprise-grade service levels above the standard one, which is an assurance about answering rather than about running the product for the customer.

TrueFoundry's trust centre lists a SOC 2 Type II attestation and offers its report, 2025 and 2026 penetration-test reports and a complete policy packet, with the documents behind a request for access.

Data residency is the stronger part of the story. TrueFoundry says no data leaves the customer's domain and offers installation in the customer's own cloud, on-premises or air-gapped, and its CISO page describes exporting telemetry through OpenTelemetry into Splunk, Datadog, Grafana and existing SIEM workflows.

The listed assurances are conventional and the underlying reports require access. The record documents no authorisation that only this vendor holds, and the SOC 2, penetration-test and policy documents behind the trust centre open only on request.

TrueFoundry is assembling a stack rather than a point product. An SDxCentral article carries the company's framing of its Seldon acquisition, under which TrueFoundry operates as a gateway and control plane above model servers such as Seldon's Core, building an independent alternative to the AI tooling bundled with AWS, Google Cloud and Azure.

The ecosystem play runs through MCP. The MCP Gateway registers internal and third-party MCP servers in one catalogue with approval flows over which roles reach each server, and the AI Gateway page names prebuilt connections to enterprise tools such as Slack, GitHub, Confluence and Datadog.

Observability plugs into what a customer already runs. TrueFoundry's telemetry is OpenTelemetry-compliant and its CISO page describes export into Splunk, Datadog, Grafana and existing SIEM workflows, so the gateway adds a control point without asking the buyer to replace its monitoring.

The founders' documented backgrounds span machine learning, infrastructure engineering and quantitative finance, with no security role identified in the reviewed sources. A VentureBeat article states that Nikunj Bajaj, Abhishek Choudhary and Anuraag Gutgutia founded the company in 2021, that all three were Meta engineers, and that they met as classmates at IIT Kharagpur.

The specialisms differ across the three. The team page says Bajaj led machine learning for one of Facebook's flagship conversational AI products, Choudhary reached Senior Staff Engineer in infrastructure before leaving, and Gutgutia was a vice president of portfolio management at WorldQuant.

No prior exit, sustained publication record or industry recognition appears for any of them in the reviewed sources. The product is now addressed partly to CISOs, and the record identifies no security role among the founders.