TrueFoundry review: the AI gateway, and why Bifrost is not it

TrueFoundry is an enterprise AI platform whose gateway is one layer of a larger control plane, deployable into VPC, on-prem, air-gapped, or multi-cloud Kubernetes. It publishes its pricing, which most enterprise gateways do not. And it is not Bifrost: that is Maxim AI's open-source gateway, a different product from a different company.

By the Continuum team. We build a workbench that runs Claude Code, Codex, and their peers, so the model rates quoted here are the ones our own cost analytics ship with.

The short version

TrueFoundry AI Gateway is a commercial enterprise gateway with RBAC, guardrails, semantic caching, intelligent routing, and an MCP gateway, claiming sub-3ms internal latency and more than 10 billion requests processed a month. Pricing is published: Developer at $0 for 50,000 requests a month and 3 users, Pro at $499 a month for 1 million requests and 10 users, Pro Plus at $2,999 a month for 25 users, and Enterprise from 10 million requests. A self-hosted gateway plane runs roughly $600 to $1,000 a month. Bifrost, which people often search alongside TrueFoundry, is an unrelated Apache 2.0 Go gateway from Maxim AI with about 11 microseconds of overhead at 5,000 RPS.

What you need to know
  • Bifrost is not a TrueFoundry product. It is Maxim AI's Apache 2.0 Go gateway. Different company.
  • TrueFoundry's AI Gateway is commercial, not open source. Its open projects are TrueForge, KubeElasti, Cognita, CruiseKube.
  • Published pricing: $0 / $499 / $2,999 per month, plus custom Enterprise. Rare transparency in this tier.
  • The meter is requests, not tokens. A one-token call costs the same as a 200k-token call.
  • Claims: sub-3ms internal latency, 10B+ requests a month, 99.99% uptime, 1,600+ models.
  • Deployment is the real differentiator: VPC, on-prem, air-gapped, multi-cloud Kubernetes.

First: Bifrost is a different product

A large share of the searches that land on "TrueFoundry AI Gateway" also carry the word Bifrost, so this belongs at the top rather than in a footnote.

Two different products from two different companies. Verified August 2026.
BifrostTrueFoundry AI Gateway
CompanyMaxim AI (maximhq on GitHub)TrueFoundry
LicenceApache 2.0, open sourceCommercial
LanguageGo, single binaryPlatform, deployed on Kubernetes
Performance claim~11 microseconds overhead at 5,000 RPSSub-3ms internal latency
GitHub stars7.4kn/a (not open source)
PriceFree$0 / $499 / $2,999 per month

If you came here wanting the fast open-source Go gateway, that is Bifrost and it is free. The rest of this page is about TrueFoundry.

What TrueFoundry's gateway does

TrueFoundry sells an enterprise AI platform. The AI Gateway is one product inside it, alongside an MCP Gateway and an Agent Gateway. That framing matters: you are buying a control plane, and the LLM proxy is a component of it.

Feature and claim summary from truefoundry.com/ai-gateway.checked aug 2026
AreaWhat is claimed
Unified APIOne API across OpenAI, Claude, Gemini, Groq, Mistral, and 250+ LLMs
Catalogue"All 1,600+ models" elsewhere on the same site
LatencySub-3ms internal latency under enterprise-scale workloads
Scale10B+ requests processed per month
Availability99.99% uptime with automatic failovers
GovernanceRBAC on models, rate limits, quotas, access control
GuardrailsPII filtering, toxicity detection, custom rules
PerformanceSemantic caching and intelligent routing
ToolingMCP integration for tools; a separate MCP Gateway product
Cost30% average cost optimisation

Pricing, which they actually publish

Publishing enterprise gateway pricing at all is unusual. LiteLLM, Kong Enterprise, and Databricks all route you to a sales conversation. TrueFoundry puts numbers on a page, and that is worth crediting even where the numbers are high.

From truefoundry.com/pricing.checked aug 2026
TierPriceRequests/monthUsers
Developer$050,0003
Pro$499/month1,000,00010
Pro Plus$2,999/month1,000,00025
EnterpriseCustom10,000,000+Custom
Deployment and overage.
ItemCost
SaaS onlyNo infrastructure cost beyond optional data storage
Self-hosted gateway plane~$600 to $1,000 per month
Control plane plus gateway planeEnterprise
Overage bundle2M requests and 5 API keys for an additional $499/month

Note also that Pro and Pro Plus include the same 1 million requests. The $2,500 difference buys 15 extra users, priority support, and dedicated onboarding, not more capacity. If you are a small team that needs the support tier, that is an expensive way to get it.

Deployment, which is the real reason to buy it

Everything above is table stakes for a 2026 gateway. What TrueFoundry has that a hosted aggregator cannot offer is the deployment matrix.

  • SaaS, with the option of keeping data storage in your own infrastructure.
  • Self-hosted gateway plane with TrueFoundry's control plane, at roughly $600 to $1,000 a month.
  • Both planes self-hosted, on Enterprise.
  • VPC, on-prem, air-gapped, or multi-cloud, managed via Helm on Kubernetes.

Air-gapped is the word doing the work. If your compliance posture means no request metadata may leave your network, the hosted options in this category are all disqualified and the choice narrows to TrueFoundry, Kong, self-hosted LiteLLM, or Bifrost. Between those, the split is whether you want a commercial platform with support (TrueFoundry, Kong) or free software you operate yourself (LiteLLM, Bifrost).

The honest assessment

Strengths

  • Published pricing. You can build a business case without a sales call, which is genuinely rare here.
  • Deployment flexibility. Air-gapped and on-prem Kubernetes, managed with Helm, with a real control plane.
  • Breadth beyond the gateway. MCP Gateway and Agent Gateway matter if your architecture is agentic.
  • Enterprise posture. RBAC, quotas, guardrails, and support tiers designed for organisations rather than teams.

Weaknesses

  • Request-based pricing disconnects cost from value: cheap calls and expensive calls bill identically.
  • Inconsistent marketing numbers. 250+ versus 1,600+ models on the same site, and an unsourced 30% saving claim.
  • The gateway is not open source. Their open projects (TrueForge, KubeElasti, Cognita, CruiseKube) are other things.
  • Pro Plus is poor value on capacity: same 1 million requests as Pro for six times the price.
  • You are buying a platform. If you only need an LLM proxy, this is a lot of surface area to adopt.

Who should buy it

An enterprise that needs an AI control plane inside its own network, has more than one team consuming models, and wants a vendor to call. If you are a startup that wants a fast proxy, Bifrost or LiteLLM does the gateway job for free and you can revisit the platform question when governance becomes the binding constraint.

Questions people ask

What is TrueFoundry?

TrueFoundry is an enterprise AI platform whose AI Gateway provides a unified API across OpenAI, Claude, Gemini, Groq, Mistral, and other providers, with RBAC, rate limits and quotas, guardrails including PII filtering and toxicity detection, semantic caching, intelligent routing, and MCP integration. It claims sub-3ms internal latency, 10 billion-plus requests processed per month, and 99.99% uptime, and deploys to VPC, on-prem, air-gapped, or multi-cloud Kubernetes.

How much does TrueFoundry cost?

Developer is $0 for 50,000 requests a month and 3 users. Pro is $499 a month for 1 million requests and 10 users. Pro Plus is $2,999 a month for the same 1 million requests but 25 users, priority support, and dedicated onboarding. Enterprise is custom from 10 million requests. A self-hosted gateway plane costs roughly $600 to $1,000 a month, and an overage bundle of 2 million requests plus 5 API keys is an additional $499 a month.

Is Bifrost the same as TrueFoundry?

No. Bifrost is an open-source LLM gateway from Maxim AI, Apache 2.0, written in Go, with roughly 11 microseconds of overhead at 5,000 RPS and 7.4k GitHub stars. TrueFoundry is a separate company selling a commercial AI platform. They appear together in search results largely because TrueFoundry publishes competitive comparison content about Bifrost.

Is TrueFoundry open source?

The AI Gateway is not. TrueFoundry maintains open-source projects including TrueForge (an agent harness), KubeElasti, Cognita, and CruiseKube, but the gateway and the platform around it are commercial. If open source is a requirement, LiteLLM (MIT), Bifrost (Apache 2.0), and Portkey's gateway (MIT) are the options in this category.

Is TrueFoundry pricing per token or per request?

Per request. A 30-token call and a 200,000-token agent turn each count as one request against your monthly allowance. That favours workloads made of a small number of large calls and penalises workloads made of many small ones, which is the opposite of how token-metered gateways behave. Work out your own request-to-token ratio before comparing prices.

TrueFoundry or Kong for an enterprise AI gateway?

Kong if you already run Kong Gateway, because the AI plugins are then an incremental config change on infrastructure your team already operates, and because Kong governs MCP and agent-to-agent traffic with dedicated plugins. TrueFoundry if you do not have an API gateway to build on and want a purpose-built AI control plane with published pricing and air-gapped deployment. Kong Konnect Plus prices per unique LLM model; TrueFoundry prices per request.

Sources

Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.

  1. TrueFoundry AI Gateway features, latency and scale claims, deployment options
  2. TrueFoundry pricing tier prices, request allowances, self-hosted plane cost
  3. maximhq/bifrost on GitHub Maxim ownership, Apache 2.0, 11 microseconds at 5,000 RPS
  4. TrueFoundry vs Bifrost (vendor comparison) cited as an example of vendor-authored comparison content
Try it

A control plane
for services.
Not for laptops.

Continuum covers coding-agent spend your platform gateway cannot see. Free workbench, org spend caps.

free app · your subscriptions · local-first