Skip to content

Pricing

Clear pricing for running AI in production.

Your providers bill you for AI usage. INFRO bills you for routing, monitoring, and control. 0% markup on model spend.

Connect the provider accounts you already have and keep paying them directly at your own rates. What you pay INFRO is a flat subscription for the layer in front of them: one API across every provider, retries and failover, a record of every request and what it cost, budgets, and policy over who may call what. Every plan includes a 14-day trial, set up with you when your early-access account opens.

Plans

Switch to annual for about 2 months free

Developer

For your first production AI application.

$35/month

Billed monthly · annual saves $72

The whole control plane, with

  • 100k requests a month
  • 1 seat · 3 projects
  • 7 days of request history · 24 hours of prompt capture
Request Developer trial

Startup

For small teams running growing workloads.

$79/month

Billed monthly · annual saves $168

Everything in Developer, plus

  • 500k requests a month
  • 3 seats · 10 projects
  • 14 days of request history and prompt capture
Request Startup trial

Team

For production teams

For teams that need deeper observability.

$199/month

Billed monthly · annual saves $408

Everything in Startup, plus

  • 2M requests a month
  • 10 seats · 30 projects
  • 30 days of request history and prompt capture
  • Custom provider cost rates
  • OpenTelemetry export
Request Team trial

Business

For company-wide oversight and reconciliation.

$599/month

Billed monthly · annual saves $1,200

Everything in Team, plus

  • 10M requests a month
  • Unlimited seats · unlimited projects
  • 90 days of request history and prompt capture
  • Provider cost reconciliation
  • Priority support
Request Business trial

14-day trial · Card required · Cancel before day 14 and you won't be charged · Set up with our teamPrices in USD, excluding tax. Charged automatically when the trial ends.

Enterprise

For organizations with procurement, security review and their own terms.

Custom pricing

Invoiced quarterly or annually. 0% markup on model spend.

Everything in Business, including SSO and SCIM, plus

  • Self-hosted or hybrid: In your own cloud, or hybrid with our hosted console.
  • Contractual SLA: Availability committed in the contract, with remedies.
  • A named engineer: A shared channel and agreed incident response times.
  • Custom request volume: Agreed with you, past the 10M a month Business includes.
  • 365 days of history: Up to 365 days of request history; Business keeps 90.
  • Your MSA and DPA: Your paper reviewed, and your security questionnaire answered in writing.
Contact salesWhat an agreement changes

Two separate invoices from two separate companies. Your providers charge you directly for inference at your own rates; INFRO charges a subscription for the gateway, the history and the controls around those requests, and takes 0% of your model spend. Nothing INFRO sells is priced per token.

Your providers bill you for AI usage
Requests run on the provider accounts you connect, so your contracts, committed-spend discounts and enterprise agreements keep working exactly as they do today. Their invoice comes to you, at your rates, and INFRO takes 0% of it. There is no arrangement under which we could take a share — it is structural, not a promise.
INFRO bills you for routing, monitoring and control
A flat subscription for the gateway that routes and retries your requests, the history that records every one of them, the cost attribution and budgets on top, and the policy, roles and audit log around them. Priced on gateway requests and what an organization needs — never on tokens, and never as a share of what a model cost.
Costs are estimated, and yours to correct
Every request carries an estimate of what your provider will bill for it, from their published rate by default. On Team and above you enter the rates you actually negotiated and every figure in the console is recomputed from those instead — which changes our arithmetic, never your provider's invoice.
Ceilings that stop, not warn
Set a monthly cap per organization, project, member or key against that estimate, with alerts at your own thresholds — or have requests refuse at the limit rather than send an email after the overspend. Those are budgets you set against your providers' spend, and they are separate from the plan allowance below.
Credentials are used, never held for us
A credential you connect is encrypted before it is stored, bound to your organization and that one connection, and never returned by any endpoint — there is no reveal, for anyone. It authenticates your requests to your provider and is used for nothing else.
Same model, same weights
A model id resolves to the model it names. INFRO does not substitute a smaller or quantised variant, shorten the context window or degrade output, and it never buys or resells inference. Your request reaches your provider as you wrote it.

What each plan is for, then every capability, row by row. Everything on this table exists today — a comparison is exactly the place a reader assumes what is listed is available now, so nothing on a roadmap appears on it.

  1. Developer$35/mo

    Operate and debug a first production application: route it over your own provider accounts, see every request, and know what each one cost.

    Start here unless you already know you need more than one seat, more than three projects, or more than a week of request history.

  2. Startup$79/mo

    Work on it together: several people, more applications and more volume, all reading the same request history and the same costs.

    Move up from Developer when more than one person needs console access, or one application is no longer the whole story.

  3. Team$199/mo

    Run several production workloads at once: longer history, traces in your own observability stack, and cost estimates computed from the rates you actually pay rather than list prices.

    Move up from Startup when you need estimates accurate to your own provider agreement, or traces in your own collector.

  4. Business$599/mo

    Run AI at company scale: unlimited people and projects, and estimated spend checked against your providers' own cost APIs so finance can see the variance rather than take an estimate on trust.

    Move up from Team when the estimate now has to be reconciled against the providers' own billing, and headcount should stop being a limit.

Team — $199 a month

Scale

Gateway requestsPer calendar month. Retries and provider failover inside one request do not add to the count.
2M
Team membersActive members plus outstanding invitations. Gateway keys are not seats.
10
ProjectsSeparate keys, budgets, model policies and attribution per project.
30
Provider connectionsYour own accounts with OpenAI, Anthropic, Bedrock, Vertex, Azure and the rest.
Unlimited
RoutesA stable model name your application calls, with ordered or weighted targets.
Unlimited

Gateway and reliability

One API across every providerAn OpenAI-compatible surface, plus images, audio and async video.
Included
Routing, retries and fallbackPriority or weighted targets, timeouts, circuit breaking, and failover mid-outage.
Included
Routing rules and failure rulesIf-rules on a route by metadata, feature, end user, key, project, prompt size, streaming, images, tools or the hour; retry, fail over or stop for each kind of failure.
Included
Several keys per provider, standby keys and alternate hostsActive keys share the traffic; standby keys and up to five alternate hosts per key take over on failure.
Included
Image, video, speech and transcriptionEach model's settings checked before the call; video as a job with signed webhooks; finished media copied to INFRO storage.
Included
Response cacheOpt-in, exact-match cache for chat completions. A cached answer never reaches your provider.
Up to 7 days
Console PlaygroundEvery capability on one of your connections, with Copy cURL. Its history stays in your browser.
Included
Scoped gateway keysPer application and environment, with their own limits, so a provider credential is never copied into a service.
Included
Encrypted provider credentialsSealed per organization and connection. No endpoint returns one, for anyone.
Included

Observability and cost

Request historyStatus, model, provider, route, tokens, latency and estimated cost, per request. This is metadata; prompts and completions are the separate row below.
30 days
Estimated provider costWhat your provider will bill for each request, itemised. Never a charge from INFRO.
Included
Cost attributionGrouped by project, key, model, provider or route.
Included
Budgets and alertsAt organization, project, member and key level, reporting or enforcing.
Included
Usage and request APIsGET /v1/usage and GET /v1/requests, on any gateway key.
Included
Prompt and response capture, off by defaultPrompts and completions, stored only if you turn it on and deleted at the end of the window. Switching it off is available on every plan and never needs an upgrade.
Up to 30 days
Custom provider rates for accurate cost trackingEnter the rates you agreed with a provider and every estimate is recomputed from them. Your provider agreements apply on every plan — this changes INFRO's estimate, not what you are charged.
Included
OpenTelemetry trace export to your own collectorTraces pushed to your own collector rather than only read in ours.
Included
Provider commitment trackingCommitted-spend agreements tracked against reported cost, with a forecast to the end of the term.
Not included
Traffic changesA new model snapshot, longer answers, a lost prompt cache or more fallbacks, with the estimated weekly change in provider cost. Per route and workflow from Team.
Included
Unit economicsCost per run of a workflow, from the metadata you already attach.
Included
Cost centers, shared-cost allocation and monthly statementsGroup spend by team or product line and close each month with a statement.
Included
FOCUS 1.4 exportCost and usage as FOCUS rows, in JSON or CSV, for FinOps tools that read the standard.
Included
Provider cost reconciliation against your providers' own billingConnect a read-only billing credential and INFRO compares its own estimate against what the provider actually reported, day by day, and states the variance. Available for providers that publish a cost API.
Not included

Governance and access

Roles and permissionsOwner, admin, developer and analyst, with console access and keys scoped to each.
Included
Model allowlistsApprove which models an organization, project or key may call.
Included
EnvironmentsYour own environments beside production, staging and development. A key belongs to one; connections and routes serve several.
Included
Key limits and rotationLimit a gateway key to chosen providers or IP ranges, and rotate it with an overlap window.
Included
Two-factor sign-in and passkeysWith a list of signed-in sessions and an organization setting that requires two-factor sign-in.
Included
Management APIConfigure the organization as code through /manage/v1 with a scoped, expiring token.
Read and write
MCP server for AI assistantsClaude, ChatGPT, Claude Code or another assistant reads spend, requests, traces and findings with your role, and makes the changes you confirm. Each change is in the audit log; prompts and completions are never returned.
Read and change
MCP gatewayYour agents reach your own MCP servers with their gateway key: approved and pinned tool definitions, grants per key, guardrails on arguments and results, and every tool call on the trace. Turning a server or a tool off works on every plan.
Not included
Audit logMember, key and policy changes, with the actor and the time.
Included
Signed webhooksFor async job events, with secret rotation.
Included
Spend anomaly findingsEach scope's usual day is learned from its own history; a day that departs from it becomes a finding in the console, with the arithmetic stated. Organization, project, key, connection and model.
Included
Spend alerts to Slack and signed webhooksAnomalies, budget forecasts, cost drift and failing connections routed to Slack and to signed webhook endpoints, as well as email. Turning a channel off works on every plan.
Included
Spend anomalies per workflow and per end customerThe same detector run per workflow and per end customer — your 200 highest-spending end customers — from the metadata you already attach.
Included
Spend caps per end customer, set from the console or your backendOne rule for every end customer of your product, such as $5 a day, plus overrides for specific customer ids that your backend can set through the management API. A request over its cap gets a 402 your app can show its own user, with the time the cap resets. Seeing what each end customer costs works on every plan.
Included
SAML/OIDC single sign-onPeople sign in through your own identity provider. The domain is proved by a DNS record before the provider can sign anybody in, and an assertion can grant member or admin — never ownership.
Not included
SCIM directory provisioningYour directory creates and deactivates INFRO members over SCIM 2.0, so a seat disappears on the day somebody leaves. Removing people keeps working on every plan.
Not included
Guardrails: PII, secrets and your own termsPattern-based checks on chat completion prompts and responses: block, redact or flag email addresses, phone numbers, card numbers, IBANs, US Social Security numbers, IP addresses, API keys and a list of terms you supply. Pattern matching can miss things; it is not a classifier. Turning a policy off keeps working on every plan.
Not included
Priority support queueYour messages are answered ahead of the general queue, by the engineers who build the platform. No response time is committed — an enterprise agreement is where a contractual one is written.
Not included

Studio

INFRO StudioThe workspace for image, video and audio: storyboard, cast, Director, recipes, export to Final Cut Pro, DaVinci Resolve and Premiere Pro, and team review. Every generation runs on your own provider connections, with its estimated provider cost shown before it runs.
Included
Studio storageHow long Studio keeps uploads and results once an owner or admin turns its storage on. Separate from request payload logging.
30 days

The whole plan, for two weeks, on your own provider accounts and your own traffic. An evaluation of a control plane that cannot carry real requests is not an evaluation.

  1. 01

    Pick a plan

    Choose a plan and enter a card at checkout. You get that plan in full for 14 days, counted from when checkout completes — not from when your account was created. Switching plans mid-trial keeps the days you have left and bills the new plan when the trial ends.

  2. 02

    Connect a provider

    Your own OpenAI, Anthropic, Bedrock, Vertex or Azure account. The credential is encrypted, bound to your organization, and never returned by any endpoint.

  3. 03

    Change one base URL

    Your application keeps its own SDK and its own prompts. Requests start appearing with their route, provider, latency and estimated cost.

  4. 04

    Decide

    At the end of day 14 the plan you have selected is charged automatically. Cancel before day 14 and you won't be charged. If the charge fails, access continues for 3 more days while you update the card; after that it is suspended, and nothing is deleted.

Your AI providers bill you directly for model usage throughout the trial and after it, at your own rates. That part never runs through INFRO.

Each plan includes a number of gateway requests a month. Here is exactly what one is, when the count resets, what you are told on the way, and what happens when you pass it.

One request is one request
One inference request your application sends to the gateway counts once, whether it streams or not. Retries and provider failover inside a single request do not add to the count. One request your application made is one request on your allowance, however many upstream attempts it took.
What is never counted
Console and dashboard traffic, authentication, health checks, model-catalog reads and INFRO's own background jobs are never counted. A request refused before it reached a provider — a budget ceiling, a model your policy does not allow — is recorded in your request log so you can see it, and does not count against the allowance.
It resets on the 1st, UTC
The allowance resets at 00:00 UTC on the first of each calendar month, on monthly and annual subscriptions alike. An annual subscription buys twelve monthly allowances rather than one pooled yearly quota, so a quiet month never funds a busy one.
Past it, on a paid plan
Requests keep being served. The gateway never refuses production traffic for exceeding a subscription allowance. Nothing extra is charged. There is no overage rate, on any plan. The overage is counted and shown on the billing screen, which offers the larger plan. Owners, admins and the billing contact are also emailed once when the month reaches 80%, 100% and 150% of the allowance.
If it keeps happening
Sustained running above an allowance is a conversation about moving up a plan, not an automatic charge and not an automatic upgrade. We would rather move you up a plan than quietly bill you for the excess or, worse, start refusing your users’ requests over a quota of our own.
An allowance is not a budget
The plan allowance counts gateway requests and is what the subscription includes. A budget caps the estimated provider spend those requests will put on your providers’ invoice. Different units, different companies’ money, and neither one moves the other: passing an allowance does not touch a budget, and a budget you set never changes what INFRO charges.
During a trial, the same
A trial includes everything in the chosen plan, with the same limits as the paid plan: past the allowance requests keep being served and measured, and are never refused for being a trial. Requests during a trial are never charged per use. The subscription itself is charged at the end of the trial unless it is cancelled first.

Volume past the largest published plan is agreed rather than published: custom allowance, custom retention and invoicing, under an enterprise agreement. Talk to us about custom volume.

Six answers — requests, people, projects, request history, prompt capture, capabilities — and it names the smallest plan that covers all of them, and which one decided it. Model spend is an optional extra field; your providers bill that separately.

Which plan do you need?

What you need

picks the plan
Capabilities you need
Your model spendoptionaldoes not change the planWhat your providers bill you today. Add it to see the subscription beside it — the two are separate bills and are never added together.
  • $2 in / $10 out
  • $0.03
  • $0.084
  • $80

The plan that fits

INFRO Startup

$79

a month — smallest plan that covers 250,000 requests a month

Charged monthly

$79

every month, or $780 once a year

Request Startup trialCompare every plan →

Chosen by requests, people, projects, retention and the capabilities you ticked — the same entitlements the gateway enforces. Your monthly selection travels with the button above.

The questions behind the number.

What happens during the 14-day trial, and after it?

You get the full plan you chose for 14 days: connect your providers, route production traffic, and see every request. Creating your account does not start the clock: you name your organization, choose a plan and enter a card at checkout, and the trial starts when that checkout completes. Nothing is charged during the trial. At the end of day 14 your card is charged automatically for the plan and interval selected at that moment; cancel before then and you won't be charged, and access continues until the trial ends. Switching plans mid-trial keeps the days you have left and bills the new plan when the trial ends. A few days before it ends we email you with the charge date and amount. If that charge fails, access continues for 3 more days while you update the card; after that access is suspended and the subscription ends, so you are never charged later for time without access. Nothing is deleted.

Do I need a card to start a trial?

Yes. The card is entered on the payment provider's checkout and is charged automatically at the end of day 14, for the plan and interval selected then. Cancel before day 14 and you won't be charged. Each person gets one trial across every organization they create or own; if you have already had one, checkout says so and the plan is charged the same day.

Is anything deleted when a trial or a subscription ends?

No configuration is deleted. Your provider connections and their credentials, gateway keys, routes, budgets, model policies, projects and team all stay exactly as they are, and start serving again the moment a plan is active — nothing has to be rebuilt. What does not pause is normal retention: request history ages out of its window on the usual schedule, and any prompts and completions you captured are deleted when their window expires. So the configuration is preserved and the logs keep expiring on the clock they were always on.

Who pays the AI providers?

You do, directly. You connect your own accounts with OpenAI, Anthropic, AWS Bedrock, Google Vertex, Azure and the rest, and those providers invoice you at your own negotiated rates under your own contracts. INFRO is never in that transaction — there is no arrangement under which we could take a percentage of it, which is why the 0% is structural rather than a promise.

How does annual billing work?

The full year is charged once, up front, at the annual price — $1,980 for Team, for example, rather than $199 a month. That works out at $165 a month and saves $408 over the year. The saving is not the same percentage on every plan, so each card prints its own. The request allowance is still monthly and resets on the 1st of each calendar month: an annual subscription buys twelve monthly allowances, not one pooled yearly quota.

What counts as a request?

One inference request your application sends to the gateway counts once, whether it streams or not. Retries and provider failover inside a single request do not add to the count. One request your application made is one request on your allowance, however many upstream attempts it took. Console and dashboard traffic, authentication, health checks, model-catalog reads and INFRO's own background jobs are never counted. A request refused before it reached a provider — a budget ceiling, a model your policy does not allow — is recorded in your request log so you can see it, and does not count against the allowance. The allowance resets at 00:00 UTC on the first of each calendar month, on monthly and annual subscriptions alike.

What happens if I go past the request allowance?

Requests keep being served. The gateway never refuses production traffic for exceeding a subscription allowance. Nothing extra is charged. There is no overage rate, on any plan. The overage is counted and shown on the billing screen, which offers the larger plan. Owners, admins and the billing contact are also emailed once when the month reaches 80%, 100% and 150% of the allowance. Sustained running above an allowance is a conversation about moving up a plan, not an automatic charge and not an automatic upgrade. A trial works the same way: it includes everything in the chosen plan, with the same limits as the paid plan.

What if I need more than Business?

Volume past the largest published plan is agreed rather than published: custom allowance, custom retention and invoicing, under an enterprise agreement. Past the largest published allowance the honest answer is a conversation rather than a card: we agree the volume, the retention and the terms, and invoice rather than charge a card. The estimator on this page says so too — it will not recommend a plan that cannot serve your traffic.

What is the difference between request history and prompt storage?

Request history is metadata: status, model, provider, route, tokens, latency and estimated cost, for every request. That is what the retention window on each plan covers — 7 days on Developer, up to 90 days on Business. Prompts and completions are a separate store with its own window and its own switch: capture is off by default on every plan and has to be turned on deliberately, and turning it off again never requires an upgrade. The two windows are not always different — on Startup, Team and Business the capture window matches the history window, and only Developer is shorter (7 days of history, 24 hours of capture). Each plan card prints its own.

How do cancellation and downgrades work?

Cancel at any time from the payment provider's portal, linked in the console; it also holds your payment method and your invoices and receipts. Cancel during a trial and nothing is charged. Cancel a paid plan and it runs to the end of the period you paid for and then stops serving inference; the console, your history and your exports keep working, and nothing is deleted. Plan changes are made on the billing screen in the INFRO console, not in the portal: an upgrade applies immediately with a prorated charge for the rest of the period, and a downgrade applies at your next renewal, with no refund or credit for the current period. If a smaller plan has fewer seats or projects than you are currently using, we do not delete projects or remove people to make it fit: what you have keeps working, and you cannot create more until you are back under the new limit.

How do enterprise terms work?

Custom volume, custom retention, invoicing rather than a card, a contractual availability commitment, a named person, and a self-hosted or hybrid deployment if you need one. It starts with a conversation rather than a checkout, and it is still a software agreement. Model spend stays between you and your providers under every plan.

Request a 14-day trialCustom volume, invoicing or your own terms