Portkey vs LiteLLM: Which AI Gateway in 2026?
By API7.ai Team
Last updated: June 2026
Portkey and LiteLLM take different shapes: Portkey pairs an open-source gateway with a hosted observability and governance platform, while LiteLLM is a Python-based proxy and SDK with budgets and virtual keys in the open-source core. This guide compares them on architecture, routing, caching, guardrails, observability, governance, self-hosting, and pricing so you can choose the right fit.
Portkey pairs an MIT TypeScript gateway with a paid hosted control plane for observability and governance; LiteLLM is a Python SDK and proxy with broad provider coverage and OSS budgets. Both are MIT at the core, and neither does semantic routing or ensemble — so the choice comes down to provider breadth and a hosted governance UI versus a Python SDK and open-source budgets.
- Teams wanting hosted observability + governance: Portkey
- Python teams wanting an SDK + OSS budgets: LiteLLM
Portkey vs LiteLLM at a glance
Portkey leads on provider breadth and a hosted governance UI; LiteLLM keeps more governance in open source. Neither offers semantic routing or ensemble.
| Dimension | Portkey | LiteLLM |
|---|---|---|
| Best for | Hosted observability + governance | Python SDK + OSS budgets |
| Core & runtime | TypeScript gateway + hosted control plane | Python (SDK + proxy) |
| Open-source license | MIT gateway; platform commercial | MIT core; enterprise/ commercial |
| Provider coverage | 1,600+ models / 45+ providers | 100+ providers |
| Semantic routing | — Rule-based only | — Not documented |
| Ensemble / fusion | — Not documented | — Not documented |
| Caching | ✓ Simple + semantic | ✓ Exact + semantic |
| MCP gateway | ✓ Documented | ✓ In open source |
| Self-host / VPC | OSS self-host; VPC = Enterprise | Docker/K8s; no managed VPC |
| SSO / SCIM | Enterprise-only | Enterprise-only |
What is Portkey?
Portkey is an AI gateway pairing an open-source TypeScript gateway with a paid hosted control plane for observability, prompt management, and governance.
Portkey is an AI gateway that pairs an open-source TypeScript gateway (MIT) with a hosted control plane for observability, prompt management, and governance. It advertises 1,600+ models across 45+ providers.
Language
TypeScript (Node.js)
License
MIT gateway + commercial platform
Form factor
OSS gateway + hosted SaaS
Best for
Hosted observability & governance
Pros
- 1,600+ models across 45+ providers
- Simple and semantic caching
- Documented MCP gateway
- Mature observability, analytics, and prompt management
Cons
- RBAC (Production tier), SSO/SCIM, VPC deploy, granular budgets are paid
- Conditional routing is rule/parameter-based, not semantic
- No documented ensemble; log/metric retention is tiered by plan
What is LiteLLM?
LiteLLM is an open-source Python SDK and proxy that exposes 100+ LLM providers through one OpenAI-compatible API, with budgets and virtual keys in open source.
LiteLLM is an open-source Python SDK and proxy server that exposes 100+ LLM providers through one OpenAI-compatible API. Its core is MIT-licensed, with a paid Enterprise tier for identity, audit, and advanced guardrail features.
Language
Python
License
MIT (core) + commercial enterprise/
Form factor
SDK + proxy server
Best for
Python teams, broad provider access
Pros
- Broadest provider coverage (100+) in OpenAI format
- Ships as both an SDK and a proxy
- Virtual keys, budgets, and spend tracking in open source
- Semantic caching and an MCP gateway in open source today
Cons
- Python/Uvicorn runtime; key & budget features require PostgreSQL
- No semantic routing or ensemble per its own routing docs
- SSO/SAML, RBAC, SCIM, and audit logs are paid Enterprise
Portkey vs LiteLLM: feature comparison
The two converge on routing basics, caching, and an MCP gateway, then diverge on form factor (hosted platform vs Python SDK) and where each draws the free/paid line.
| Feature | Portkey | LiteLLM |
|---|---|---|
| Core & runtime | Open-source TypeScript gateway (Node.js) + a separate hosted control plane | Python; SDK + proxy on Uvicorn; key & budget features need PostgreSQL |
| Provider coverage | 1,600+ models across 45+ providers | 100+ providers in OpenAI format |
| Routing | Load balancing, fallbacks, retries, conditional routing (metadata/params/path), circuit breaker | Simple-shuffle, latency, least-busy, rate-limit-aware, cost-based, custom; fallbacks & retries |
| Semantic routing | — Rule/parameter-based only | — Not documented |
| Ensemble / fusion | — Not documented | — Not documented |
| Caching | Simple + semantic caching (Redis) | Exact + semantic (Qdrant, Redis, Valkey, S3, GCS) |
| Guardrails | 20+ deterministic + LLM-based + partner checks; tiered Basic/Partner/Pro/custom | Presidio PII + hooks in OSS; moderation & per-key scoping are Enterprise |
| Observability | Logging, tracing, 21+ metric dashboard, OpenTelemetry-compliant; retention tiered | Prometheus in OSS, Langfuse, OpenTelemetry, Datadog; some team export Enterprise |
| Budgets & governance | Model Catalog: cost/token budgets, rate limits, workspaces — granular budgets Enterprise | Virtual keys, per-key/user/team budgets, spend tracking in OSS (needs PostgreSQL) |
| MCP gateway | ✓ Documented (auth + access control for remote MCP) | ✓ In OSS (access control by key/team) |
| Self-host / VPC | OSS self-hosts; managed VPC/hybrid is Enterprise | Docker/K8s (Helm beta); no managed-VPC product |
| Enterprise identity | RBAC from Production tier; SSO/SAML/OIDC & SCIM Enterprise | Global roles in OSS; SSO/RBAC/SCIM/audit Enterprise |
Pricing comparison
Both have a free open-source core and paywall advanced governance — but along different lines.
Portkey's gateway is free (MIT); the hosted platform is tiered — a free Developer tier with capped logs, a paid Production tier (from about $49/month) that unlocks RBAC and service accounts, and an Enterprise tier for SSO/SCIM, VPC/hybrid deployment, granular budgets, and custom retention. LiteLLM's core is free (MIT) including budgets and virtual keys; its Enterprise license (custom-priced) adds SSO/SAML, org and team RBAC, SCIM, audit logs, and enterprise guardrails. In short, Portkey gates breadth-of-platform features by tier, while LiteLLM gates enterprise identity and audit.
When to use Portkey vs LiteLLM
Choose Portkey for breadth and a hosted governance UI; choose LiteLLM for a Python SDK and open-source budgets.
Choose Portkey if you…
- Want the widest provider catalog behind one API
- Want a hosted observability, analytics, and prompt-management UI
- Are comfortable with tiered pricing and tiered log retention
Choose LiteLLM if you…
- Want a Python SDK as well as a proxy
- Want OSS budgets and virtual keys without a managed service
- Want semantic caching and an MCP gateway in the open-source build
Bottom line
Choose Portkey for the widest catalog and a hosted governance UI; choose LiteLLM for a Python SDK plus open-source budgets and virtual keys.
For the broadest provider catalog and a hosted observability-and-governance UI, Portkey is the stronger pick; for a Python SDK with OSS budgets and virtual keys — plus semantic caching and an MCP gateway in the open-source build — LiteLLM fits better. If you're evaluating open-source AI gateways more broadly, AISIX is another option worth a look: a Rust, Apache-2.0 gateway with semantic routing and ensemble built in. See AISIX vs LiteLLM.
Frequently asked questions
Related comparisons
AISIX vs LiteLLM · LiteLLM alternatives · All AI gateway comparisons
Ready to get started?
For more information about full API lifecycle management, please contact us to Meet with our API Experts.
