Portkey Pricing
Portkey pricing bills every tier by a 'recorded log' its own pricing page never once defines.
Portkey's real numbers, verified straight from its own pricing page: Open Source is free to self-host with no request limit, Developer is free with 10,000 recorded logs a month, Production is $49 a month for 100,000 logs plus $9 per additional 100,000 up to 3 million, and Enterprise has no published price at all. No tier caps seats or users, a real structural difference from most of this category. This page verifies every figure, works out the real overage math in dollars, and covers the one question Portkey's own comparison table leaves open: what a single recorded log actually counts, and what your own customer never sees no matter which tier you're on.
Verified against Portkey's own pricing page
What Portkey actually costs
Portkey's pricing runs four tiers, and every figure below comes from Portkey's own current pricing page. Open Source costs $0, self-hosted on your own infrastructure, with no request limit stated anywhere, but only a basic dashboard, community support, and the core gateway features: the universal API, retries and timeouts, routing, and automatic fallbacks. Developer costs $0 a month too, hosted on Portkey's own cloud, and includes 10,000 recorded logs a month with three-day log retention and thirty-day metric retention, alongside prompt management capped at three templates, a playground, versioning, variables, and simple caching.
Production costs $49 a month and is the first tier billed against real usage rather than a hard cap: 100,000 recorded logs included, then $9 for every additional 100,000 up to a stated ceiling of 3 million, with retention stretched to thirty days for logs and ninety days for metrics. It also adds role-based access control, service account API keys, semantic caching, and LLM and partner guardrails, none of which appear on Developer. Enterprise has no dollar figure published anywhere on Portkey's site, quoted only through a sales conversation, and covers 10 million or more recorded logs, configurable retention, private cloud or VPC hosting, custom guardrail hooks, SSO, and signed SOC 2, GDPR, and HIPAA coverage.
Four tiers, one meter
The four Portkey tiers, in plain numbers
Straight from Portkey's own pricing page, not a third party estimate.
Open Source: $0, self-hosted
No request limit stated anywhere. Universal API, retries, routing, fallbacks, and a basic dashboard, run entirely on infrastructure you own and support yourself.
Developer: $0 a month, hosted
10,000 recorded logs a month, three-day log retention, three saved prompt templates. Exceeding the log allowance stops new logs from recording; requests keep routing normally.
Production: $49 a month
100,000 recorded logs included, then $9 per additional 100,000 up to a stated 3 million ceiling. Thirty-day log retention, role-based access control, and guardrails all start here.
Enterprise: custom pricing
No published number. 10 million or more recorded logs, configurable retention, private cloud or VPC hosting, SSO, and signed SOC 2, GDPR, and HIPAA coverage.
The number every paid tier bills against, undefined
What actually counts as one recorded log?
Every Portkey tier past Open Source bills against the same single meter, the recorded log, and Portkey's own pricing page states the allowance and the overage rate for each tier plainly. Its cost management documentation describes real-time cost tracking per request and aggregated reporting across time periods, but neither page, nor any other Portkey documentation checked while researching this page, states outright whether one recorded log means one incoming request, one underlying model call counted separately from a retry, or one step inside a longer agent workflow routed through Portkey's gateway.
That ambiguity compounds fastest for exactly the workloads this category exists to serve. A single customer message handled by a multi-step agent, a router step, a tool call, a retry after a provider timeout, and a guardrail check attached to the final response, could plausibly generate several recorded logs against Portkey's allowance for what the end user experienced as one exchange. Whether that scenario counts as one log or five changes a Production workspace's real monthly bill by a meaningful multiple, and Portkey's own page leaves no way to answer it without running a real workload and watching the count climb against the invoice.
The structural difference from most of this category
No tier caps seats. Every tier meters logs instead
Portkey's own comparison table lists no user limit and no workspace cap anywhere, Open Source through Enterprise.
Unlimited seats, every tier
Open Source, Developer, Production, and Enterprise all skip a seat count entirely. A team can add engineers freely without triggering a plan change, something PromptLayer and several other tools in this category cap explicitly.
Role-based access control from $49
RBAC and service account API keys ship on Production, not held back for an Enterprise-only conversation the way access controls are gated on some comparable pricing tables.
What the meter doesn't show
Portkey's logs tell you what the gateway routed. They tell your customer nothing
Every number on Portkey's pricing page, the recorded log count climbing toward a plan's allowance, the per-request cost breakdown, the guardrail violation flag on a flagged response, describes what happened inside your own gateway and what it cost your own team to route it. Removing seat limits, which Portkey does across every tier, only widens who on your own team can see that same workspace; it adds no second, scoped view for a customer sitting outside your company. Even Enterprise's private cloud and VPC hosting changes where the gateway physically runs without changing who the dashboard was built for: the same engineers with access on shared infrastructure keep that same access on a dedicated deployment, just on infrastructure their own company controls.
If you sell an AI agent inside a product other companies pay for, that gap surfaces the moment a customer asks what the agent actually did for their account this month, a question no recorded log or cost breakdown answers on its own. AiAgRe's Node SDK reads the same kind of agent events Portkey's gateway already routes and logs, through the same LangChain, LlamaIndex, or CrewAI integrations, and tags each one with an organization identity and a customer identity at the point of ingestion. That turns into white-labeled dashboard components your customer sees inside your own product, scoped so tightly that one customer's numbers never reach another customer's view. It doesn't replace Portkey's own gateway routing or its cost tracking, and most teams run both: Portkey for what the gateway did with the request, AiAgRe for what that same request was worth to the person paying for it.
When a recorded log stops being the real question
Three questions Portkey's pricing page was never built to answer
None of these show up in a log count, a cost breakdown, or a guardrail flag. All three show up once real customers start asking.
Your customer asks what their account got, not how many logs recorded
A customer deciding whether to renew wants their own deflection rate and cost per resolution, scoped to their own traffic, not a walkthrough of a gateway workspace they can't log into. A clean guardrail record reassures your engineers; it says nothing to the person deciding whether to keep paying.
Unlimited seats for your team, zero seats for theirs
Every Portkey tier drops the seat cap entirely as a workspace scales. None of them add a scoped view for a customer outside your company, because Portkey's workspace model was built for your own engineers, not for a second tenant layer underneath them.
The overage rate tracks volume. Nothing tracks value
The $9-per-100,000 overage rate answers what a gateway's traffic costs as it grows, on a unit Portkey's own page never fully defines. It says nothing about what that traffic was worth, the same gap a fuller look at AI agent monitoring tools as a category runs into past Portkey alone.
FAQs
Portkey pricing: frequently asked questions
Common questions from teams pricing Portkey against their own log volume before they commit to a tier, or before they add a second layer on top.
How much does Portkey actually cost?
Four tiers, verified straight from Portkey's own pricing page. Open Source costs $0 to self-host, with no request limit at all, but only a basic dashboard and community support. Developer costs $0 a month too, hosted on Portkey's own infrastructure, and includes 10,000 recorded logs a month with three days of log retention and thirty days of metric retention. Production costs $49 a month, includes 100,000 recorded logs, and bills $9 for every additional 100,000 logs up to a stated ceiling of 3 million, with retention stretched to thirty days for logs and ninety days for metrics. Enterprise has no published number, quoted only through Portkey's sales team, and covers 10 million or more recorded logs with configurable retention, private cloud deployment, and SOC 2, GDPR, and HIPAA coverage none of the three lower tiers mention.
What's included on the free Developer plan, and where does it actually stop?
The Universal API and key management, prompt management capped at three saved templates, a playground, prompt versioning and variables, simple response caching, deterministic guardrails, and community support, all before touching a single paid feature. The number that stops a growing team isn't the three-template cap, most teams write far fewer reusable prompts than that early on, it's the 10,000 recorded logs a month paired with three-day log retention. A team debugging an issue a customer reported four days ago has already lost the trace by the time anyone goes looking for it, regardless of whether the month's 10,000-log allowance was ever exceeded.
What exactly is a "recorded log", and does Portkey ever define it?
No, and neither Portkey's pricing page nor its cost management documentation ever settles it in plain language. The pricing page bills every tier against recorded logs, and the documentation describes real-time cost tracking per request and aggregated reporting across time periods, but nothing public states whether one log means one incoming request, one underlying model call after a retry, or one step inside a longer agent workflow. That gap matters more here than it would for a simple chatbot: a single customer message routed through a multi-step agent, with a retry on a timed-out provider call and a guardrail check attached, could plausibly generate several recorded logs for what the end user experienced as one exchange. Budgeting against the log allowance without knowing the multiplier is a real risk, and Portkey's own site currently gives no way to resolve it without running a workload and watching the count climb.
What happens once the free Developer plan goes over its 10,000 log allowance?
Portkey states this one plainly, more plainly than most vendors state their own overage behavior: "Exceeding this limit doesn't affect your requests; only logs beyond the limit are not recorded." The agent keeps running and every response still reaches the user. What stops is the record of what happened. A support agent that goes over its allowance on day twenty of the month runs the rest of that month with no trace history at all, which means the one week it would matter most, right as a customer escalates a complaint about something the agent did, is exactly the week with the least visibility into what the agent actually did. This sits close to the same blind spot covered on the Datadog LLM Observability pricing page, where a span cap produces the identical silent gap under a different vendor's numbers.
How does Production's $9-per-100,000 overage actually add up?
Production's base $49 covers the first 100,000 recorded logs. Everything past that bills at $9 per additional 100,000, up to a stated ceiling of 3 million logs a month. Worked out in real numbers: a workspace logging 500,000 events a month pays $49 plus four blocks of overage at $9 each, $85 total. One logging a million events pays $49 plus nine blocks, $130 total. One pushing the full 3 million logs Portkey lists as Production's ceiling pays $49 plus twenty nine blocks, $310 total. What Portkey's own page never states is what happens to a workspace that needs a 3,000,001st log in the same month: whether it hard-stops at the same silent non-recording behavior Developer uses, keeps billing overage past the stated ceiling, or forces an Enterprise conversation. Budget for reaching that number before hitting it, not after.
Does any Portkey tier cap the number of seats or users?
No, and it's worth stating directly since most of the vendors in this category do. Portkey's own comparison table lists no seat count, no user limit, and no workspace cap on Open Source, Developer, Production, or Enterprise: every tier is metered purely by recorded logs, never by headcount. Role-based access control ships starting on the $49 Production tier too, not held back for Enterprise the way it is on the PromptLayer pricing page's comparable tier. The tradeoff runs the other way from a seat-capped tool: a five-person team generating heavy log volume can outgrow Portkey's pricing fast even while comfortably under any seat count, while a fifty-person team with light, low-volume agent traffic could stay on Production a long time without ever adding a person to the bill.
Is Portkey cheaper than Langfuse or Datadog LLM Observability?
At the free tier, Langfuse's Hobby plan includes fifty thousand units a month against Portkey Developer's 10,000 recorded logs, though a Langfuse unit and a Portkey log count different underlying events and don't convert one for one. Past free, Portkey's $49 Production tier undercuts Langfuse's $199 Pro tier and Datadog LLM Observability's usage-based pricing beyond its free fifteen-host edition on sticker price alone, but Langfuse's Pro plan publishes three-year data retention where Portkey's Production tier caps logs at thirty days. The full unit-by-unit Langfuse breakdown lives on the Langfuse pricing page, and the fuller category view, including where Datadog and five other tools fit, lives on the AI agent monitoring tools comparison.
Does any Portkey plan show pricing or ROI data to my own customers, and can I run AiAgRe alongside it?
No plan does, Open Source through Enterprise, and yes, running both is the normal setup rather than a workaround. Every number Portkey's dashboard renders, the recorded log count climbing toward a plan's allowance, the per-request cost breakdown, the guardrail violation flags, lives inside a workspace built for the team paying Portkey's own bill. Removing every seat limit, as Portkey does, widens who on your own team can log in; it does nothing to add a second, scoped view for a customer sitting outside your company. AiAgRe's Node SDK reads the same kind of underlying agent events Portkey's gateway already routes and logs, tags each one with an organization identity and a customer identity at the point of ingestion, and turns that into white-labeled dashboard components your own customer sees inside your product, showing deflection rate and cost per resolution scoped so tightly that one customer's numbers never reach another's view. Installing it doesn't touch how Portkey counts a recorded log; the two SDKs read the same agent, one billed and displayed to your team, the other billed and displayed to the customer paying for it.
Already routing every request through Portkey? Now show your customer what it did.
Request access and we'll walk through how AiAgRe's embed tokens map onto the trace data Portkey's gateway already produces.