PK
Portkey
Open-source AI gateway with guardrails and observability
AI model gatewayResearch complete
Company overview
Portkey provides an AI gateway with guardrails, observability, caching, routing and budget controls. The gateway is published as an open-source project alongside a hosted platform.
Ideal customer: Product engineering and AI platform teams that want to self-host or start on a hosted tier
- Guardrails on inputs and outputs
- Caching and routing across many providers
- Budget and rate controls per key or workspace
GateYourAI verdict
Portkey sits closest to the middle of the security-and-cost thesis for teams that want optionality: an open-source gateway they can run themselves, with guardrails, observability and budget controls in the same layer. Individual capability marks await documentation verification.
Best for: Teams that want an open-source gateway with guardrails and spend controls. Read how we assess platforms.
Security capabilities
A capability is marked available only when an official page or documentation says so. “Not publicly documented” means we could not find it — not that the vendor lacks it.
| Capability | Status | Verification | Source |
|---|---|---|---|
| Prompt-injection detectionIdentifies attempts to override system instructions in prompts or tool output. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Prompt-injection blockingBlocks the request before it reaches the model, rather than only logging it. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Jailbreak detectionDetects known jailbreak and guardrail-evasion patterns. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| PII detectionIdentifies personal information in prompts or responses.Available as a guardrail check. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| PII redactionMasks or removes personal information before it leaves the gateway.Redaction is available through guardrail actions; coverage varies by check. | LimitedLimited. Partially documented. | Partially documented | Portkey AI Gateway documentation |
| Secret & credential scanningDetects API keys, tokens and credentials in prompts or output. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Input inspectionInspects outbound requests before they reach the model.Input guardrails run before the request reaches a model. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Output inspectionInspects inbound model responses before they reach the application.Output guardrails run on model responses. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Content moderationClassifies harmful or policy-violating content. Not equivalent to injection defense. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Policy enforcementConfigurable rules applied per route, team, key or environment.Guardrails are configured per route, key or workspace. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Agent & tool controlsConstrains tool calls, autonomous actions and agent permissions. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| MCP securityControls or inspects Model Context Protocol server traffic.MCP gateway support. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Audit logsDurable, reviewable record of requests and policy decisions.Request logs and traces. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Access controlsRoles, teams, keys and scoped permissions.Workspaces, virtual keys and role-based access. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
Cost-optimization capabilities
Token reduction, avoided calls, cheaper-model routing, budgets and reporting are tracked separately because they are not interchangeable.
| Capability | Status | Verification | Source |
|---|---|---|---|
| Prompt compressionReduces the number of tokens actually sent to the model. | Not publicly documentedNot publicly documented. Not publicly documented. | Not publicly documented | — |
| Exact cachingReturns a stored response for a byte-identical request. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Semantic cachingReturns a stored response for a semantically similar request using embeddings. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Intelligent model routingChooses a model per request based on task, cost or performance.Conditional routing across providers and models. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Automatic fallbackRetries against an alternate model or provider on failure. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Load balancingDistributes traffic across keys, regions or providers. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Token budgetsHard or soft limits expressed in tokens.Budget limits can be expressed in tokens per key. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Dollar budgetsSpending limits expressed in currency.Cost budgets per virtual key or workspace. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Rate limitsRequest throttling. Throttling is not the same as token reduction. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Cost attributionAttributes spend to teams, keys, users or features.Spend is attributed per key, workspace and metadata tag. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Cost analyticsDashboards and reporting on token and dollar usage. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Savings reportingReports the spend avoided by compression, caching or routing. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
Supported models and deployment options
Hosted platform plus self-hosted open-source gateway. Verification pending.
| Capability | Status | Verification | Source |
|---|---|---|---|
| OpenAISupports OpenAI models. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| AnthropicSupports Anthropic models. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Google GeminiSupports Google Gemini models. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| AWS BedrockSupports AWS Bedrock as a provider. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Azure OpenAISupports Azure OpenAI deployments. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| Open-source modelsSupports open-weight models or self-hosted inference endpoints. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| SaaS deploymentVendor-hosted managed service. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| VPC / private cloudDeployable inside a customer cloud account.Private-cloud deployment for enterprise plans. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| On-premiseDeployable in a customer data centre. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Self-hostedCustomer runs and operates the gateway themselves.The gateway can be run from the open-source repository. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey-AI/gateway on GitHub |
| Open-source optionSource code published under an open-source licence.The gateway is published as a public repository. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey-AI/gateway on GitHub |
| APIProgrammatic API for configuration or traffic. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
| SDKOfficial client libraries.Official Python and Node SDKs. | AvailableAvailable. Verified from official documentation. | Verified from official documentation | Portkey AI Gateway documentation |
Pricing status
Verification pending. Open-source gateway available; hosted plan detail not yet verified.
| Capability | Status | Verification | Source |
|---|---|---|---|
| Public pricingList pricing published on the vendor website. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Free tierA permanently free usage tier. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Free trialA time-limited or credit-limited trial. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Enterprise planA named enterprise tier. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
| Contact sales requiredPricing requires contacting the vendor. | Verification pendingVerification pending. Verification pending. | Verification pending | — |
Strengths
- Open-source gateway published on GitHub
- Guardrails, observability and cost controls in one product
Limitations
- Feature parity between the open-source gateway and the hosted platform needs verification
- Some capabilities are enterprise-tier only and need contract-level confirmation
Alternatives to Portkey
NeuralTrust TrustGateSecurity-minded teams that want an open-source gateway they can inspect and runTrueFoundry AI GatewayEnterprises that need gateway capabilities inside their own cloud boundaryKong AI GatewayPlatform teams already running Kong that want AI policy in the same data plane
You can also compare Portkey against up to three other platforms, or read the AI Gate pillar page.
Sources
- Portkey product site Portkey · Official product page · Accessed 2026-07-30
- Portkey AI Gateway documentation Portkey · Official documentation · Accessed 2026-07-30Primary source of truth for capability marks.
- Portkey-AI/gateway on GitHub Portkey · GitHub repository · Accessed 2026-07-30
Last reviewed: 2026-07-30
Newsletter
The AI Gateway Brief
A weekly briefing on AI security threats, gateway technology, token economics, model routing, and the infrastructure controlling enterprise AI.