Alternatives guidePublished August 16, 2026· Last verified August 16, 2026

Databricks Unity AI Gateway Alternatives (2026): 7 Enterprise Options

Unity AI Gateway (GA August 2026) is a strong governance layer — for AI that routes through a Databricks workspace. Teams with multi-platform estates, sovereignty constraints or compliance-evidence requirements have seven credible options, compared here with citations.

Alejo HernandezAlejo HernandezCTO, Kosmoy

Databricks relaunched its gateway as Unity AI Gateway — the product that began life as Mosaic AI Gateway — and took it to general availability on August 4, 2026 (GA announcement). For a Databricks-standardized organization it is a genuinely strong control layer: budgets that can hard-block spend, PII and safety guardrails enforced inline, MCP servers governed as Unity Catalog securables, and per-request cost in system tables a finance team can query with SQL — with rate limits, fallbacks and traffic splitting included at no extra charge.

The reason to read an alternatives page is scope, not quality. The gateway exists only as a Databricks-managed service, governs the traffic that opts in by pointing a base URL at a workspace, and documents no EU AI Act, ISO/IEC 42001 or NIST AI RMF evidence tooling as of August 16, 2026. This guide compares seven alternatives — independent gateways, open-source proxies, an API platform, a self-hosted AI platform, an edge gateway and a rival platform-native option — using the same ten-axis rubric we apply to every vendor, with every claim cited to documentation.


Why teams look beyond Databricks Unity AI Gateway

Start with what Unity AI Gateway does well, because inside its home estate the bar is high: governance rides the platform you already run. MCP servers register as Unity Catalog securables with tool-level filtering and centralized audit; budgets alert or hard-block per user and workload; service policies apply PII redaction, safety, jailbreak and hallucination filters before routing; and usage lands in system tables next to your data (docs). Core routing features are free, so a Databricks shop's marginal cost of adoption is close to zero.

The first switch driver is estate shape. The gateway requires a Unity Catalog-enabled workspace, and governance is opt-in per application — an app is governed once it repoints its base URL at the workspace endpoint. Enterprises whose AI also lives on Azure AI Foundry, Bedrock, Vertex, Salesforce or ServiceNow — or that are not Databricks customers at all — need a control point that does not assume the lakehouse is the center.

The second is deployment control. Unity AI Gateway runs only as a Databricks-managed service on AWS, Azure or GCP. No self-hosted, on-premises or air-gapped option is documented as of August 16, 2026 — a hard stop for sovereignty-constrained buyers. Five of the seven alternatives below can run in infrastructure you control.

The third is compliance evidence. The gateway produces audit-quality logs and policy records in Unity Catalog, but documents no EU AI Act, ISO/IEC 42001 or NIST AI RMF evidence generation, framework mapping or use-case risk classification as of August 16, 2026. Teams whose auditors expect framework-mapped artifacts need tooling the gateway does not attempt.

The rest are breadth, openness and maturity. The gateway fronts OpenAI, Anthropic, Cohere, Bedrock, Vertex and any OpenAI-compatible provider — a workspace-anchored model rather than the 100-to-1,600-model catalogs of the independent gateways (external models). There is no open-source core to audit or fork. And GA is weeks old as of this writing, with Smart Routing, service policies and agent services still in Beta — teams that prefer battle-tested products will note that several alternatives have been GA for years.

How we chose the alternatives

  • A real runtime data path. Every entry proxies AI traffic and can enforce policy on it — no workflow-only governance tools.
  • Documented capabilities only, verified against vendor documentation as of August 16, 2026; vendor performance numbers are reported as claims, not facts.
  • Deployment sovereignty stated plainly. Databricks-managed-only is the target's biggest constraint, so each profile says exactly where the alternative runs: SaaS, hybrid, self-hosted or air-gapped.
  • Migration friction. Unity AI Gateway's own adoption mechanism is a base-URL repoint; each alternative is judged on whether arriving (or leaving) is likewise an endpoint change rather than a rewrite.
  • Architecture diversity. Independent gateways, open-source proxies, an API platform, a self-hosted AI platform, an edge gateway and a hyperscaler platform — so the list covers different estate shapes rather than seven flavors of one answer.

The alternatives at a glance

ProductBest forDeploymentOpen sourcePricing model
PortkeyPlatform teams standardizing LLM traffic, cost tracking and guardrails across many product teams.SaaS, hybrid (data plane in your VPC) or air-gapped (enterprise)MIT (gateway core); platform proprietaryFree dev tier; Pro from $49/mo; enterprise by quote
KosmoyRegulated enterprises that need governance enforced in the runtime path, in their own infrastructure.Self-hosted — single-tenant, your own Kubernetes (air-gap capable)ProprietaryEnterprise subscription; no self-service tier.
LiteLLMPlatform teams standardizing LLM access on a self-hosted, OSS-first gateway — with the engineers to run it.Self-hosted (Docker, Kubernetes, Helm, Terraform) — air-gap supportedMIT core; enterprise/ directory under a commercial licenseOSS core free; enterprise license by quote (contact sales), with a free 7-day trial
Kong AI GatewayPlatform and API teams — especially existing Kong customers — extending proven gateway governance to LLM, MCP and agent traffic.Konnect (SaaS control plane, customer-hosted data planes) or fully self-managed; air-gap marketed for EnterpriseApache-2.0 core (incl. ai-proxy); many AI plugins Enterprise/Konnect-onlyOSS gateway free; Konnect has free and self-serve tiers; Enterprise by quote — AI Gateway is not sold separately
TrueFoundryPlatform and ML infrastructure teams in regulated sectors that run Kubernetes and want gateway governance plus model serving on their own GPUs.SaaS, hybrid, or self-hosted on your Kubernetes — documented air-gapped installProprietary platform; OSS side projects (aitori, Apache-2.0)Free developer tier, self-serve paid plans and a custom enterprise tier — figures on the pricing page
Cloudflare AI GatewayDeveloper and platform teams — especially existing Cloudflare/Workers shops — that want a free, zero-ops edge proxy for caching, reliability, multi-provider routing and cost control.SaaS only — runs on Cloudflare's global edge; no self-hosted or VPC optionProprietaryCore gateway features free on all plans; log storage tied to the Workers plan; guardrails billed as Workers AI inference; Unified Billing adds a 5% fee on provider credits.
Azure AI FoundryEnterprises standardized on Microsoft Azure that want a single vendor for the full generative-AI and agent lifecycle and value tight integration with Entra, Defender, Purview and Agent 365 in exchange for Azure lock-in.Azure PaaS only — managed Azure service; no customer-VPC, on-premises or air-gapped install of Foundry itself. Sovereignty limited to Azure regions and Microsoft sovereign-cloud offeringsProprietary Azure PaaS; adjacent SDKs are OSS (e.g. the PyRIT red-teaming framework, Azure AI Evaluation SDK), but the platform is closedConsumption-based: model inference billed per token, no separate charge to run Foundry-native agents, with separate licenses for some governance/security features (Defender for Cloud, Agent 365). See pricing page.

Last verified August 16, 2026 against each vendor's public documentation.

Capability shape vs Databricks Unity AI Gateway

Each panel shows one alternative across the same ten capability axes (0–10); the dashed outline is Databricks Unity AI Gateway for reference. The further a shape reaches on a spoke, the stronger that capability.

Portkey
INVSECOBSGWGRDCTNCMPEVLBLDSOV
Kosmoy
INVSECOBSGWGRDCTNCMPEVLBLDSOV
LiteLLM
INVSECOBSGWGRDCTNCMPEVLBLDSOV
Kong AI Gateway
INVSECOBSGWGRDCTNCMPEVLBLDSOV
TrueFoundry
INVSECOBSGWGRDCTNCMPEVLBLDSOV
Cloudflare AI Gateway
INVSECOBSGWGRDCTNCMPEVLBLDSOV
Azure AI Foundry
INVSECOBSGWGRDCTNCMPEVLBLDSOV
dashed = Databricks Unity AI GatewayINV AI Inventory & Discovery · SEC Security & Shadow AI · OBS Observability & FinOps · GW Gateway & Policy Control · GRD Guardrails & Runtime Safety · CTN Agent Containment · CMP Compliance & Audit · EVL Testing, Evals & Red-teaming · BLD Agent Building · SOV Deployment Sovereignty

See it live

How Kosmoy scores on these axes — see it on your own use case.

Book a demo

30 minutes, straight to the product. Or email sales@kosmoy.com.


The alternatives, one by one

01

Portkey

AI gateway & LLM-ops control plane

Portkey is an AI gateway and control plane for production AI — one API to 1,600+ models across 45+ providers, with observability, guardrails, prompt management and MCP/agent access control — acquired by Palo Alto Networks in May 2026.

The closest independent equivalent: a managed gateway control plane with more provider breadth and a deployment ladder Unity AI Gateway does not attempt — owned by Palo Alto Networks since May 2026 as the gateway layer of Prisma AIRS.

Where it beats Databricks Unity AI Gateway

  • Provider and model breadth: one OpenAI-compatible API to 1,600+ models across 45+ providers, versus five named provider families plus OpenAI-compatible endpoints fronted from a workspace.
  • Deployment flexibility: an MIT open-source gateway core, SaaS, hybrid VPC data plane and a documented air-gapped enterprise tier — Unity AI Gateway is Databricks-managed only.
  • Request-level observability with 21+ metrics, OpenTelemetry and data-lake export, plus an adoption path (free dev tier, Pro from $49/mo) that requires no platform contract.

Where it falls short

  • Nothing matches the Databricks-native integration: no Unity Catalog securables, no lakehouse system tables, no first-party MLflow 3 evaluation loop.
  • Cost governance is per-key budgets and rate limits; Databricks' alert-or-block budgets plus enriched billing records are the stronger surface for finance-grade attribution.
  • Post-acquisition roadmap uncertainty is real: the public docs changelog stops at April 2026, and future packaging under Prisma AIRS is unknown.
Deployment: SaaS, hybrid (data plane in your VPC) or air-gapped (enterprise)Open source: MIT (gateway core); platform proprietaryPricing: Free dev tier; Pro from $49/mo; enterprise by quote
02

Kosmoy

AI management platform

A self-hosted control plane for enterprise AI: one inventory, one policy gateway, one audit trail and a containment sandbox for every model, agent and MCP server a company runs.

The pick when the requirement is governing the estate, not one platform's traffic. Kosmoy is an AI management platform — inventory, gateway, compliance evidence and agent containment in one self-hosted product — with Databricks as one of the estates it can sit above.

Where it beats Databricks Unity AI Gateway

  • Estate-wide inventory: four registries, including a master agent registry that harvests agents from Azure AI Foundry, Bedrock, Vertex, Salesforce and ServiceNow — Unity AI Gateway inventories what is registered in Unity Catalog.
  • Compliance as a product: EU AI Act, ISO/IEC 42001 (aligned) and NIST AI RMF evidence bundles generated from registry state plus gateway logs — the target documents none of these as of August 16, 2026.
  • Containment and sovereignty: kernel-enforced Action Capsule sandboxing with a kill switch, delivered as single-tenant software in your own Kubernetes, air-gap capable — in production at Italy's central bank and Europe's largest defence and aerospace group.

Where it falls short

  • Lakehouse-native cost analytics: Databricks' per-request system tables and enriched billing records are a stronger surface for finance-grade SQL slicing than Kosmoy's cost and usage monitoring.
  • Inside Databricks the integration gap is structural: Kosmoy cannot govern workspace-native serving and MCP as natively as the platform's own catalog does.
  • No free or self-service tier — procurement runs through enterprise sales — and the evaluation suite shipped mid-2026, newer than the pure-play eval platforms.
Deployment: Self-hosted — single-tenant, your own Kubernetes (air-gap capable)Open source: ProprietaryPricing: Enterprise subscription; no self-service tier.
03

LiteLLM

Open-source LLM proxy & AI gateway

LiteLLM is BerriAI's open-source proxy and Python SDK that puts 100+ LLM providers behind one OpenAI-compatible API — with spend tracking, guardrails, MCP and A2A gateways, and an enterprise license that adds SSO, RBAC and audit logs on the same self-hosted deployment.

The open-source answer: an MIT-licensed proxy (~53.6k stars) that does the routing, budget and key-management job on any infrastructure that runs a container — with none of the platform attached.

Where it beats Databricks Unity AI Gateway

  • Runs anywhere: Docker, Kubernetes, Helm, Terraform — explicitly air-gap supported — versus a gateway that exists only inside a managed workspace.
  • Open source with zero license cost: the MIT core self-hosts everything, and the enterprise tier is a license key on your own deployment, not a hosted service.
  • No platform commitment, and community pace: 100+ providers, weekly stable releases and a Rust migration announced June 2026 targeting sub-1ms overhead.

Where it falls short

  • Guardrails orchestrate third-party engines; Unity AI Gateway ships PII redaction, safety, jailbreak and hallucination filters natively, plus Contextual Service Policies (Beta) over agent actions.
  • Enterprise governance depth: admin-action audit logs and RBAC, but no catalog-of-record model like Unity Catalog securables and no compliance evidence tooling.
  • You own the operations — Postgres, Redis, upgrades, on-call — where the target arrives fully managed with the workspace.
Deployment: Self-hosted (Docker, Kubernetes, Helm, Terraform) — air-gap supportedOpen source: MIT core; enterprise/ directory under a commercial licensePricing: OSS core free; enterprise license by quote (contact sales), with a free 7-day trial
04

Kong AI Gateway

AI gateway on the Kong API platform

Kong AI Gateway is the AI extension of Kong's API gateway: a plugin-based data path that proxies, secures, rate-limits, caches and observes LLM, MCP and agent-to-agent traffic — self-managed or via the Konnect SaaS control plane.

The API-platform answer: Kong extends a battle-tested Apache-2.0 gateway (~43.8k stars) to LLM, MCP and agent-to-agent traffic, with data planes that stay in your network even under the Konnect SaaS control plane.

Where it beats Databricks Unity AI Gateway

  • API-gateway lineage: one runtime and one operating model for REST, LLM, MCP and A2A traffic — Agent Gateway went GA in April 2026 — where Unity AI Gateway governs AI traffic only.
  • Self-managed data planes: Apache-2.0 core, fully self-managed Enterprise, and customer-hosted data planes under Konnect, so prompts need not transit a vendor-managed service.
  • MCP security depth independent of any data platform: per-tool ACLs, OAuth2 scope-based tool filtering and RFC 8693 token exchange.

Where it falls short

  • FinOps is token-aware rate limiting plus Konnect analytics — no alert-or-block spend budgets and no billing-grade cost tables like the target's system tables.
  • The strongest AI plugins (semantic prompt guard, PII sanitizer, LLM analytics) are Enterprise/Konnect-only; the free OSS tier covers mainly ai-proxy basics.
  • No evaluation tooling of any kind, and no compliance evidence tooling — Kong's EU AI Act content is positioning backed by audit logs.
Deployment: Konnect (SaaS control plane, customer-hosted data planes) or fully self-managed; air-gap marketed for EnterpriseOpen source: Apache-2.0 core (incl. ai-proxy); many AI plugins Enterprise/Konnect-onlyPricing: OSS gateway free; Konnect has free and self-serve tiers; Enterprise by quote — AI Gateway is not sold separately
05

TrueFoundry

Enterprise AI gateway & Kubernetes-native ML platform

TrueFoundry is a Kubernetes-native enterprise AI platform that combines LLM, MCP and Agent gateways with model serving, fine-tuning and GPU orchestration — deployable as SaaS, hybrid, self-hosted or fully air-gapped.

The self-hosted AI platform: LLM, MCP and Agent gateways plus model serving, fine-tuning and GPU orchestration on your own Kubernetes — a Representative Vendor in the February 2026 Gartner Market Guide for AI Gateways.

Where it beats Databricks Unity AI Gateway

  • Platform breadth on infrastructure you control: the gateway comes with vLLM/Triton serving, fine-tuning and fractional GPUs — Unity AI Gateway governs traffic and leaves model hosting to other Databricks products.
  • A documented air-gapped install: images and Helm charts mirrored to a customer-controlled OCI registry with no outbound dependencies.
  • Agent-level FinOps engineering: per-agent identities, cost-velocity circuit breakers and instant tool revocation.

Where it falls short

  • Governance is gateway-scoped: no Unity Catalog-style catalog of record, and no compliance evidence tooling — its posture is SOC 2 Type 2, HIPAA and GDPR.
  • Guardrails and observability largely orchestrate external engines rather than matching the target's native service-policy depth.
  • A smaller vendor digesting its own June 2026 acquisition (Seldon AI), with a Kubernetes-first operating model that is heavier than switching on a workspace feature.
Deployment: SaaS, hybrid, or self-hosted on your Kubernetes — documented air-gapped installOpen source: Proprietary platform; OSS side projects (aitori, Apache-2.0)Pricing: Free developer tier, self-serve paid plans and a custom enterprise tier — figures on the pricing page
06

Cloudflare AI Gateway

Edge AI gateway

Cloudflare AI Gateway is an edge proxy between applications and major AI providers, adding caching, rate limiting, spend limits, logging and analytics, dynamic routing with fallbacks and retries, stored provider keys and Llama-Guard-based guardrails on Cloudflare's global network.

The zero-ops option: an edge proxy whose core features — caching, rate limiting, routing, retries, spend limits — are free on all plans, from a public company, with no platform prerequisite.

Where it beats Databricks Unity AI Gateway

  • Adoption cost: the core gateway is free at every plan level and needs no workspace, no catalog and no contract — the lowest barrier in the category.
  • Edge scale and simplicity: nothing to host, and Cloudflare has publicly documented scaling the gateway to billions of logs.
  • Real spend enforcement without a platform: enforced dollar spend limits scopable by model, provider or custom metadata, shipped June 2026.

Where it falls short

  • Governance depth: guardrails rest on a single moderation model (Llama Guard 3 8B), RBAC is coarse, and MCP traffic is not governed — no counterpart to Unity Catalog securables or contextual service policies.
  • Sovereignty is lower still: SaaS-only, all traffic transits Cloudflare's edge, and Cloudflare's own Data Localization Suite docs list AI Gateway as incompatible with Regional Services.
  • No org-wide inventory, no compliance evidence tooling and no agent containment as of August 16, 2026.
Deployment: SaaS only — runs on Cloudflare's global edge; no self-hosted or VPC optionOpen source: ProprietaryPricing: Core gateway features free on all plans; log storage tied to the Workers plan; guardrails billed as Workers AI inference; Unified Billing adds a 5% fee on provider credits.
07

Azure AI Foundry

Unified enterprise AI development platform (hyperscaler AI studio) on Azure — model catalog, agent service, evaluations, content safety, observability and a governance control plane

Microsoft's Azure-native platform (rebranded 'Microsoft Foundry' at Build 2026) for building, deploying, evaluating, governing and operating generative-AI models and agents, spanning a large model catalog, the Foundry Agent Service, Azure AI Content Safety guardrails and the Foundry Control Plane.

The platform-native alternative for Azure estates — the same architectural bet as Unity AI Gateway, made on Microsoft's cloud: governance, guardrails, evaluations and an agent runtime arrive with the platform (rebranded Microsoft Foundry at Build 2026).

Where it beats Databricks Unity AI Gateway

  • Agent building and runtime: the Foundry Agent Service offers managed endpoints, sandboxed hosted sessions and a dedicated Entra identity per agent — Unity AI Gateway registers and governs agents but builds nothing.
  • Native evaluation and red teaming: risk and safety evaluators plus the PyRIT-based AI Red Teaming Agent, where Databricks delegates evaluation to MLflow 3.
  • Certification posture: Foundry Models achieved ISO/IEC 42001:2023 certification (July 2025), letting customers inherit AI-management-system controls.

Where it falls short

  • Cross-provider routing: Foundry's enforcement is scoped to the Azure catalog and agent runtime, while Unity AI Gateway brokers OpenAI, Anthropic, Cohere, Bedrock, Vertex and any OpenAI-compatible provider.
  • Governance fragments across Microsoft licenses — Entra Agent ID, Defender for Cloud, Agent 365 — versus the target's one-catalog story.
  • It answers platform gravity with more platform gravity: Azure PaaS only, no self-hosted or air-gapped option, and irrelevant to a non-Azure estate.
Deployment: Azure PaaS only — managed Azure service; no customer-VPC, on-premises or air-gapped install of Foundry itself. Sovereignty limited to Azure regions and Microsoft sovereign-cloud offeringsOpen source: Proprietary Azure PaaS; adjacent SDKs are OSS (e.g. the PyRIT red-teaming framework, Azure AI Evaluation SDK), but the platform is closedPricing: Consumption-based: model inference billed per token, no separate charge to run Foundry-native agents, with separate licenses for some governance/security features (Defender for Cloud, Agent 365). See pricing page.

Decision guide

Your AI estate is standardized on DatabricksKeep Unity AI Gateway — governance arrives with the platform, and core routing is free
You need an independent control point across clouds and platformsKosmoy (self-hosted suite with compliance and containment) or Portkey (managed gateway breadth)
You want an open-source proxy and nothing moreLiteLLM — MIT core, self-hosted anywhere including air-gapped
You govern APIs and AI traffic with one teamKong AI Gateway — REST, LLM, MCP and A2A on one runtime, data planes in your network
You need evaluation depth alongside the gatewayPair the gateway with MLflow 3 or a dedicated eval platform — no gateway on this list is an eval suite
Compliance workflows and agent containment are the requirementKosmoy — EU AI Act / ISO 42001 / NIST AI RMF evidence plus kernel-enforced containment

Questions buyers ask

Is Mosaic AI Gateway the same product as Unity AI Gateway?

Yes. Unity AI Gateway is the current name of the product Databricks launched as Mosaic AI Gateway; the rename accompanied its move onto Unity Catalog, and general availability followed on August 4, 2026. If your notes, vendor decks or configuration still say Mosaic AI Gateway, they refer to the same gateway — Databricks' documentation tree retains legacy pages under the old name.

Can Databricks Unity AI Gateway run self-hosted or air-gapped?

No. As of August 16, 2026 the gateway runs only as a Databricks-managed service inside AWS, Azure or GCP workspaces; no self-hosted, on-premises or air-gapped deployment is documented. If that is disqualifying, the self-hostable entries on this list are Kosmoy (single-tenant Kubernetes, air-gap capable), LiteLLM (MIT, air-gap supported), TrueFoundry (documented air-gapped install), Kong (self-managed data planes) and Portkey's air-gapped enterprise tier.

What is the closest like-for-like alternative to Unity AI Gateway?

No independent product replicates the Unity Catalog integration, so like-for-like means the gateway job: multi-provider routing, budgets, guardrails, MCP governance and cost visibility. Portkey is the closest managed equivalent and LiteLLM the closest open-source one. Both are OpenAI-compatible — the same base-URL mechanism Unity AI Gateway uses to onboard external apps — so trialing either alongside it is an endpoint change, not a rewrite.

Which alternative helps with EU AI Act compliance?

Kosmoy is the only entry that documents EU AI Act tooling: risk classification in an AI systems registry and framework-mapped evidence bundles (EU AI Act, ISO/IEC 42001 aligned, NIST AI RMF) generated from registry state plus gateway logs. Unity AI Gateway documents no framework-specific tooling as of August 16, 2026, and neither do the other gateways here. Azure AI Foundry's ISO/IEC 42001 certification is a real asset, but it is an attestation you inherit, not evidence tooling you operate.

Is Unity AI Gateway free?

Partially. Rate limits, fallbacks and traffic splitting ship free with a Unity Catalog-enabled workspace and there is no separate gateway license — but payload logging and usage tracking bill Databricks DBUs, external-provider tokens are paid to the provider, and the prerequisite is being a Databricks customer ([pricing](https://www.databricks.com/product/pricing/ai-gateway)). Among the alternatives, Cloudflare's core gateway is free on all plans, and LiteLLM's and Kong's open-source cores are free to self-host — your cost becomes the engineers who run them.


Sources

Every factual claim about another vendor on this page traces to that vendor's own published material or a named third-party source below.

  1. AI governance with Unity AI Gateway (Databricks docs) — accessed August 16, 2026
  2. Unity AI Gateway is Generally Available (Databricks blog, Aug 4, 2026) — accessed August 16, 2026
  3. Manage budgets for Unity AI Gateway — accessed August 16, 2026
  4. Monitor Unity AI Gateway cost — accessed August 16, 2026
  5. Unity AI Gateway pricing — accessed August 16, 2026
  6. Kosmoy AI Gateway — accessed August 16, 2026
  7. Kosmoy AI Compliance — accessed August 16, 2026
  8. Unity AI Gateway product page — accessed August 16, 2026
  9. External models in Model Serving (provider list) — accessed August 16, 2026
  10. What's new: service policies, guardrails, observability and cost controls — accessed August 16, 2026
  11. Mosaic AI Gateway updates (former product name; Databricks blog) — accessed August 16, 2026
  12. Portkey open-source gateway repository — accessed July 15, 2026
  13. Portkey docs — what is Portkey — accessed July 15, 2026
  14. Portkey docs — plan & feature comparison (SaaS / hybrid / air-gapped) — accessed July 15, 2026
  15. Portkey docs — observability — accessed July 15, 2026
  16. Portkey docs — guardrails — accessed July 15, 2026
  17. Portkey docs — MCP gateway — accessed July 15, 2026
  18. Portkey pricing — accessed July 15, 2026
  19. Palo Alto Networks press release — Portkey acquisition completed (May 29, 2026) — accessed July 15, 2026
  20. Kosmoy Platform — accessed July 15, 2026
  21. Kosmoy Action Capsule — accessed July 15, 2026
  22. Kosmoy AI Evaluation & Red Teaming — accessed July 31, 2026
  23. LiteLLM GitHub repository (stars, license, activity) — accessed July 15, 2026
  24. LiteLLM README (100+ providers, MCP/A2A, performance claims) — accessed July 15, 2026
  25. LiteLLM enterprise docs (features, SLAs, air-gap, pricing by quote) — accessed July 15, 2026
  26. LiteLLM release notes index (2026 releases) — accessed July 15, 2026
  27. Rust migration announcement (issue #31263, June 25, 2026) — accessed July 15, 2026
  28. Guardrail policy templates (incl. offline/air-gapped mode) — accessed July 15, 2026
  29. MCP deployment docs (registry, exposure controls, air-gap guidance) — accessed July 15, 2026
  30. litellm-agent-runtime (per-session VM coding-agent runtime) — accessed July 15, 2026
  31. Kong AI Gateway product page — accessed July 15, 2026
  32. Kong AI Gateway 3.14 release blog (Agent Gateway GA, A2A) — accessed July 15, 2026
  33. A2A support press release (PR Newswire, April 2026) — accessed July 15, 2026
  34. MCP Tool ACLs announcement (AI Gateway 3.13, January 2026) — accessed July 15, 2026
  35. Kong MCP Registry press release (February 2026) — accessed July 15, 2026
  36. Konnect LLM usage reporting docs — accessed July 15, 2026
  37. Kong EU AI Act positioning blog — accessed July 15, 2026
  38. Kong/kong GitHub repository — accessed July 15, 2026
  39. Kong pricing — accessed July 15, 2026
  40. TrueFoundry air-gapped deployment docs — accessed July 15, 2026
  41. TrueFoundry AI Gateway product page — accessed July 15, 2026
  42. TrueFoundry guardrails overview — accessed July 15, 2026
  43. Gartner Market Guide for AI Gateways recognition (Businesswire, Feb 20, 2026) — accessed July 15, 2026
  44. Agent Gateway launch press release (Businesswire, June 2, 2026) — accessed July 15, 2026
  45. TrueFailover launch (VentureBeat, January 2026) — accessed July 15, 2026
  46. Seldon AI acquisition (SiliconANGLE, June 25, 2026) — accessed July 15, 2026
  47. aitori repository (v0.1.0, June 25, 2026, Apache-2.0) — accessed July 15, 2026
  48. Enterprise MCP access control blog — accessed July 15, 2026
  49. Cloudflare AI Gateway docs — accessed July 15, 2026
  50. AI Gateway pricing — accessed July 15, 2026
  51. Guardrails feature docs — accessed July 15, 2026
  52. Changelog — spend limits (June 5, 2026) — accessed July 15, 2026
  53. Changelog — unified REST API (May 21, 2026) — accessed July 15, 2026
  54. Data Localization Suite docs (AI Gateway compatibility) — accessed July 15, 2026
  55. Blog — scaling AI Gateway to billions of logs — accessed July 15, 2026
  56. Blog — AI Security for Apps GA — accessed July 15, 2026
  57. Microsoft Foundry product page — accessed July 15, 2026
  58. What is Azure AI Foundry (Microsoft Learn) — accessed July 15, 2026
  59. What's new in Microsoft Foundry — Build 2026 — accessed July 15, 2026
  60. Foundry Agent Service — accessed July 15, 2026
  61. Guardrails and controls overview in Microsoft Foundry — accessed July 15, 2026
  62. AI Red Teaming Agent (Microsoft Learn) — accessed July 15, 2026
  63. Extending Defender's AI threat protection to Microsoft Foundry agents — accessed July 15, 2026
  64. Transition Foundry/Copilot agent security to Agent 365 — accessed July 15, 2026
  65. Foundry Models + Security Copilot achieve ISO/IEC 42001:2023 — accessed July 15, 2026

Need governance, not just a swap?

Kosmoy puts an inventory, a policy gateway and a containment sandbox around every AI your teams run — in your own Kubernetes.

Or email sales@kosmoy.com.