Portkey
Overview
Portkey is a production stack for GenAI builders, providing an AI gateway to 1,600+ LLMs from 40+ providers with built-in guardrails, observability, governance, and prompt management. Positions itself as the complete "control plane" for AI in production — not just routing, but security, compliance, and operational management. Serves 24,000+ organizations processing 1 trillion tokens daily, managing $180M+ in AI spending. Open-sourced Production Gateway in March 2026, making enterprise features (circuit breakers, semantic caching, budget limits, MCP Registry with OAuth 2.1) freely available. Raised $15M Series A (Elevation Capital, Lightspeed) in early 2026, expanding into enterprise and agentic AI governance. Partnered with Akto in March 2026 for native AI agent security guardrails, bringing total guardrail count to 60+.
Formidability
Score: 8/10
Portkey is the most feature-complete direct competitor in the AI gateway space. Its combination of routing (1,600+ LLMs from 40+ providers), 60+ guardrails (including Akto AgentGuard for runtime agent security), governance (workspaces, RBAC, audit trails), and observability creates a comprehensive platform that goes well beyond simple API routing. ISO 27001, SOC 2, HIPAA, and GDPR compliance makes it enterprise-ready. The March 2026 open-sourcing of the Production Gateway — including circuit breakers, semantic caching, budget limits, and MCP Registry with OAuth 2.1 — is a major competitive move that makes enterprise-grade features freely available. Scale at 24,000+ organizations, $180M+ managed AI spend, and 1 trillion tokens/day demonstrates significant traction. The $15M Series A signals serious enterprise ambition. Primary risk to OpenRouter is in the enterprise segment and increasingly in the open-source developer segment.
Markets
- Primary: Enterprise teams building production AI applications needing governance and compliance
- Secondary: Platform teams managing multi-model AI infrastructure across organizations
- Tertiary: Startups and developers wanting a managed gateway with guardrails
- Geographic: Global (edge workers worldwide, regional data residency for enterprise)
Products
- AI Gateway — routing to 1,600+ LLMs from 40+ providers, caching, load balancing, failover, rate limiting (including weekly windows and endpoint-scoped limits), gRPC support, 99.999% uptime claim
- Guardrails — 60+ built-in guardrails: jailbreak detection, PII redaction, policy enforcement, Zscaler AI Guard integration, Akto AgentGuard (runtime evaluation of tool invocations, MCP requests, and agent actions)
- Observability — detailed logs, latency metrics, token/cost analytics by app, team, or model
- Governance — workspaces, roles, data residency, audit trails, SSO/SCIM, agent workflow orchestration, multi-team resource hierarchy, required metadata enforcement (blocks requests missing mandated fields)
- Key Vault & Secret References — secure storage of LLM provider API keys with virtual keys, plus external vault integration (AWS Secrets Manager, Azure Key Vault, HashiCorp Vault)
- MCP Gateway & Registry — centralized control plane for MCP servers with versioning, tracking, and OAuth 2.1 (PKCE) authentication
- Prompt Management — version control, A/B testing for prompts
- Portkey Skills — integration with AI coding assistants (Cursor, GitHub Copilot, Claude Code), plus deeper integrations with AI coding wrappers like Conductor (Claude Code GUI with worktree isolation and parallel agents)
- Agent Gateway (beta) — production layer for governing autonomous AI agents: agent registry with governed endpoints, unified auth (API keys, JWT, Basic Auth, MCP auth spec), full execution traces with 40+ metrics, 50+ guardrails (PII/PHI filtering, content moderation), automatic fallbacks and load balancing, scope-based access control per agent/team/user — zero code changes to agents required
- Responses API — cross-provider support for OpenAI Responses API and Anthropic Messages API
- Open-source Production Gateway — self-hostable with enterprise features: circuit breakers, semantic caching, budget limits, model catalog, MCP Registry, usage policies
Pricing
| Tier | Cost | Key Limits |
|---|---|---|
| Developer (Free) | $0 | 10K logs/mo, 3-day log retention, 30-day metrics, deterministic guardrails |
| Production | $49/mo | 100K logs/mo, 30-day log retention, 90-day metrics, LLM & partner guardrails, simple & semantic caching |
| Enterprise | Custom | 10M+ logs, governance, data residency, SSO, VPC, private cloud deployment, data export to data lakes, data isolation, Custom Guardrail Hooks, Advanced Evaluation Templates, Granular Budget & Rate Limits |
| Open Source | Free (self-hosted) | Universal API, circuit breakers, semantic caching, budget limits, MCP Registry, model catalog |
Charges based on recorded logs, not tokens. Production tier: $9 per additional 100K requests (up to 3M). Service account API keys available from Production tier. LLM provider costs are separate.
URLs to Monitor
| URL | Label | Notes |
|---|---|---|
https://portkey.ai/docs/product/ai-gateway |
AI Gateway Docs | Gateway product documentation |
https://portkey.ai/docs/product/guardrails |
Guardrails | Guardrails feature docs |
https://portkey.ai/docs/integrations/llms |
LLM Integrations | Supported LLM providers |
https://portkey.ai/pricing |
Pricing | Pricing page |
https://portkey.ai/docs/changelog |
Changelog | Product updates |
https://portkey.ai/blog |
Blog | Company blog |
Strategy
- Production-first positioning: Not just a gateway — a complete AI production stack (guardrails, governance, observability)
- Enterprise focus: ISO/SOC2/HIPAA/GDPR compliance, regional data residency, SSO/SCIM — targeting regulated industries. $15M Series A accelerates enterprise push
- Guardrails differentiation: 50+ built-in guardrails plus Zscaler AI Guard integration for enterprise DLP
- Open-source gateway expansion: Open-sourced Production Gateway with enterprise features (circuit breakers, semantic caching, MCP Registry, OAuth 2.1) — aggressive developer capture strategy, upselling only on-prem enterprise features (gRPC, SSO, SCIM, AWS KMS)
- Agentic AI play: Agent Gateway (beta) for governing autonomous agents with registry, auth, and observability; MCP Gateway for centralized MCP server management; Portkey Skills for coding assistants — expanding from LLM routing to full agent lifecycle governance
- API compatibility expansion: Responses API across all providers, Messages API universal adapter — reducing switching costs
- Multi-SDK approach: Python, Node.js, and REST API SDKs with OpenAI compatibility
- Coding assistant infrastructure play: Expanding from Portkey Skills (lightweight integration) to full gateway integration with AI coding wrappers like Conductor — positioning as the cost/observability/reliability layer beneath AI coding tools
- Developer content marketing: Active blog producing Claude Code guides, API comparisons — positioning as thought leader