Portkey
4.3
Portkey bundles an AI gateway with observability, prompt management and guardrails — the operational layer a team assembles once an LLM feature reaches production. Routing, retries, budget limits and output checks live in one configuration rather than scattered through application code.
- Commercial use OK
- Free tier
- Enterprise contract
- Pricing
- $49/mo
- Free tier
- Free tier
- Commercial use
- Permitted
- Japanese UI
- Not confirmed
- Enterprise contract
- Available
What it is good for
- LLM features in production that need budgets and limits enforced
- Routing between providers by cost, latency or availability
- Applying output guardrails consistently across features
- Managing prompts outside the application deploy cycle
Strengths and trade-offs
Strengths
- Gateway, observability, prompts and guardrails in one place
- Budget and rate limits enforced centrally
- Provider routing and automatic retries
- Prompt changes without an application deploy
Trade-offs
- Another dependency in the request path
- Overlaps with tooling teams may already run
- Configuration can drift away from the code it governs
Editors' review
The operations layer you assemble by hand otherwise
Every team shipping LLM features builds the same things eventually: retries, a budget cap so one bug does not produce a five-figure bill, somewhere to change a prompt without a deploy, and checks on what the model returns. Portkey sells that assembled.
Central budget enforcement is the feature that justifies it fastest. An unbounded retry loop against a frontier model is a genuinely expensive failure mode, and application-level limits are easy to forget.
Two cautions. It is in the request path, so it is now part of your availability. And prompts managed outside the repository can drift from the code that depends on them — decide early where the source of truth is.
Verdict
Worth it once LLM features are in production and spend is real. Premature before that.
Compiled from official documentation and public sources, with our own editorial judgement. We do not publish figures we cannot source.