Private early access

Request a 14-day trial.

Evaluate INFRO on your own provider account and your own traffic. Card required at checkout; cancel before day 14 and you won’t be charged.

What happens after you submit

Right away
An automatic confirmation email
Sent from infro.io to the address you entered, so you know the request was recorded.
Review
A person reads your request
We email you when an account can be opened for you. Accounts open in batches during private early access, so we do not commit to a date.
Account opens
You start the 14-day trial
Name your organization, choose a plan and enter a card at checkout — the clock starts when that checkout completes, not when you submit this form or when the account opens. Then connect a provider account you already hold, create a gateway key, and invite your team.
At the end of the trial
Charged automatically on day 14, unless you cancel
The plan and interval selected at that moment are charged to your card at the end of day 14. Cancel before day 14 and you won't be charged. Your providers bill you directly for model usage throughout; INFRO takes none of it.

Want to see it with the team first? Request a demo

Want a price for your workload? Request a workload estimate

Reviewing before you commit? The API documentation and plans are public.

What the trial account includes.

What you can use when your account opens, followed by the capabilities reserved for enterprise agreements or planned for later.

Enterprise — Available on an enterprise agreement rather than self-serve.On the roadmap — Planned after v1. No date committed.

Model catalog: Text, image, video, and audio models with each vendor's reference rate, browsable now at /providers.
API documentation: The full request and response contract for every endpoint, published in full so it can be reviewed before you write any code against it.
Customer console: The full customer console at dash.infro.io, on your own organization's keys, requests, spend and routing. Signed in — an account comes with early access.
Provider connections: Connect provider accounts you already hold. Each credential is encrypted when saved, scoped to your organization and never shown again; applications use INFRO gateway keys instead.
Text — OpenAI-compatible: POST /v1/chat/completions with the request shape the OpenAI SDKs already send, including streaming and tool calling.
Image, video, and audio: One request shape per modality across every model, with video and long audio running as async jobs.
Async jobs and webhooks: Submit long renders, get a job id immediately, receive a signed webhook on completion.
First-party SDKs: Python, TypeScript, and Dart clients covering every modality on one key. The OpenAI SDKs work against the text endpoint without them.
Media re-hosting: Generated media is copied to INFRO storage so output URLs do not expire underneath your application.
Health-aware routing: Named routes use priority or weighted targets over your own connections, with health checks, retries and failover.
Automatic failover: A route that times out or 5xxs before the response starts is retried elsewhere without your client seeing it.
Routing rules: If-rules on a route send matching requests to their own targets, by metadata, feature, end user, key, project, prompt size, streaming, images, tools or time of day. Failure rules decide per failure kind whether to retry, move on or stop.
Alternate hosts and key spreading: A connection can list more API hosts for the same key, tried when the first cannot be reached. Requests that name a model directly are spread across a provider's keys, with standby keys used only on failure.
Usage and cost per request: Every request records its model, latency, token or unit counts, and estimated provider cost, queryable by project and key.
Request traces and payload logs: Per-request timing detail, with prompt and completion logging off by default and a retention window you choose.
Usage, request and FOCUS exports: GET /v1/usage and GET /v1/requests on any gateway key, the request log and usage series as CSV, plus a FOCUS v1.4 Cost and Usage dataset as JSON or CSV, so AI spend lands in your FinOps platform beside the rest of your cloud cost.
OpenTelemetry export: Spans pushed to your own collector from the request path, with the endpoint and headers set in the console. Included from the Team plan.
Console playground: Run text, image, video, speech and transcription on one of your connections or through a gateway key, and copy the call as cURL. The Playground's history stays in your browser.
Budgets and spend ceilings: A hard cap per organization, project, member, and key, with alerts at your own thresholds.
Spend attribution: Group cost by model, project, key, or an end-user id you attach to the request.
Negotiated provider rates: Enter the rates your contracts actually carry, per model and optionally per provider, so estimated cost uses them before the vendor's list rate. Included from the Team plan.
Provider cost reconciliation: Daily comparison of estimated spend against the cost reported by OpenAI's and Anthropic's own cost APIs, using a separate admin credential, with the likely cause of a persistent gap and a suggested rate you can apply. Included on the Business plan.
Provider commitment tracking: Record a committed-spend agreement with a provider and track it against reported cost, with a forecast to the end of the term. Included on the Business plan.
Spend anomaly and drift alerts: Each scope's usual day is learned from its own history; a day that departs from it, a budget forecast to be exceeded, cost estimates drifting from the provider's own figures, unpriced traffic and failing connections become findings in the console. Routed to email on every plan, and to Slack and signed webhooks from the Startup plan.
Summary, weekly summary and monthly report: One page for leaders: this budget cycle's estimated provider spend against the budget and its projection, reliability, governance and attribution as four status lights with their reasons, and what needs a decision. Emailed to owners and finance members every Monday, with a printable monthly report on the 2nd; members choose for themselves.
Project access for members: An owner or admin limits a developer or analyst to some projects: they read those projects' requests, analytics, jobs and findings and manage the gateway keys in them, and the gateway refuses everything organization-wide.
Saved views and links to a view: Any console screen with its period, project, environment and grouping kept under a name, for the member who saved it or shared with the organization, and a link to the view as it is. Opening a view reads the figures again through the reader's own role.
Microsoft Teams and PagerDuty alert channels: Findings posted to a Microsoft Teams channel through a Teams workflow, or opened as PagerDuty alerts through an Events API v2 integration, beside email, Slack and signed webhooks. Included where alert channels are.
Traffic changes: Changes in your traffic that come with no error: a model answered by a new snapshot, longer answers, a lost prompt cache, more fallbacks, answers cut off by the length limit, slower responses or a new error mix. Each finding names the likely cause and the estimated weekly change in provider cost. INFRO sees request metadata, not answers, so it does not judge answer quality.
Response cache: Opt-in, exact-match cache for chat completions. An identical request inside the time-to-live you set is answered from the cache and never reaches your provider, so your provider does not bill it. Every plan; the longest time-to-live depends on the plan. No semantic matching.
Organizations and roles: Owner, admin, developer, and analyst roles with keys and console access scoped to each.
Environments: Production, staging and development to start, and your own beside them. A connection or route serves one, several or every environment; each gateway key belongs to exactly one, and never reaches a connection outside it.
Key limits and rotation: Limit a gateway key to chosen providers or IP ranges, and rotate it with an overlap window so the old key keeps working while you deploy the new one.
Model allowlists: Approve which models a project may call; anything else returns a policy error instead of a line item.
Guardrails: Pattern-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 your own terms, plus optional classifier checks on your own connection to a decision model's provider. Detection can miss things. Included from the Business plan.
Jev decisions on your TypeSafe account: POST /v1/systemone runs TypeSafe's Jev on your own TypeSafe connection, with TypeSafe's request body and model names: key scope and spend limits checked before TypeSafe is called, the version that answered and an estimated cost on every response, and a request record for each decision.
Decisions with Jev: Saved questions your code calls as POST /v1/decisions/{slug}, answered by Jev on your own TypeSafe connection or by any other decision model you connect: an outcome, a confidence band from your thresholds, pins decided before any model call, a deadline with a fallback, and batches of up to 500 items. Tool-call gating, triage, evals, labeling and control loops are templates of it.
Rerank: POST /v1/rerank in Cohere's shape, served by Jev on your own TypeSafe connection or by any other decision model you connect: each passage judged against the query, ordered by the probability it is relevant.
Prompt classifier on routes: A route can have Jev, or any other decision model you connect, pick a class for each prompt; route rules send each class to its own targets, and an unsure or unreachable classifier falls back to a class you choose.
Classifier guardrail checks: Yes/no checks for prompt injection, jailbreaks, harmful, self-harm and off-topic requests, answered by Jev on your own TypeSafe connection or by any other decision model you connect, with a block and a flag threshold each. Prompts only; the text is never stored. Included with guardrails, from the Business plan.
Decision models from every provider: Decision policies, rerank, classifier guardrails, the route classifier and evals on the decision model of your choice, on your own account with its provider: OpenAI's GPT-6 Luna, Cloudflare's Clef, Perplexity, Inception, Liquid AI, Upstage, Microsoft-Decision-1 on Azure, or any of them through OpenRouter. OpenAI's SDK and TypeSafe's SDK both work against INFRO.
Eval monitors: Score a share of served chat answers against your rubric in the background with Jev or any other decision model you connect, recorded per request and summarized per served model.
End-customer spend caps: One spend cap for every end customer of your product, per organization or per project, with overrides your backend sets through the management API. A refused request returns a 402 with the time the cap resets, for your app to show its user. Caps from the Team plan; the per-customer cost view on every plan.
Audit log: Member, key, and policy changes recorded with actor and timestamp, viewable in the console.
Management API: Configure an organization as code through /manage/v1 with a scoped, expiring management token: projects, gateway keys, connections, routes, budgets, alerts, cost centers, commitments, guardrails and end-customer caps. Read access from the Startup plan, every write from Team.
MCP server for AI assistants: Connect Claude Code, Cursor, VS Code, Windsurf or Codex to an organization with a management token: read spend, requests, traces, findings and configuration, and make the changes the token's scopes allow. Prompts, credentials and key secrets never pass through it.
Assistant sign-in for the MCP server: Connect Claude.ai, Claude Desktop, ChatGPT and other assistants to the MCP server by signing in from the assistant, with your own role in one organization you choose, instead of pasting a management token.
Reference images, last frames, extend and video-to-video: Guide an image or video with reference images and their roles, end a video on a chosen frame, steer away from what you do not want, continue a finished video or restyle one — on the models and providers that take them, checked before any call.
Upscaling, lip-sync, sound effects and music: Upscale an image or a video, lip-sync a video or a still to a voice line, and make sound effects and music — on the models your own providers serve, each request checked against what the model takes before any call.
Cost estimates before a request: Ask what a request would cost on your own provider account before you make it: per image with resolution and quality tiers, per second of video, per character of speech, per token of text.
INFRO Studio: A workspace for images, video and audio made through your own provider connections: a project library, generation history and estimated provider cost per result, kept for a window your plan sets and deleted after it.
Deprecation and maintenance notices in the console: Catalog model deprecations, maintenance windows and product changes appear under the bell in the console, with each deprecated model's retirement date, its successor and what in your organization still uses it, and are emailed to owners and admins.
MCP gateway for your agents' tools: Register your own MCP servers once with their credentials, approve and pin each tool's definition, grant gateway keys access, and let agents call those tools through INFRO with guardrails, a call log on the same trace as their model calls, and kill switches checked on every call. INFRO executes no tool.
Provider Program: AI API providers apply to have INFRO integrate their API, follow the application on a private page, send test access safely, tell INFRO's customers about deprecations and incidents, and add a Connect to INFRO button to their docs. Free in both directions; it never changes routing.
Two-factor authentication and passkeys: Two-factor sign-in with an authenticator app and one-time recovery codes, passkeys and hardware security keys for sign-in or as the second step, a list of signed-in sessions you can end, and an organization setting that requires two-factor for every member. On every plan.
SSO and SCIM: SAML and OIDC single sign-on against your own identity provider, and SCIM 2.0 provisioning so your directory creates and deactivates INFRO members. Included from the Business plan. A domain is proved by DNS before any provider can sign anybody in.

Enterprise

Real, but sold with an agreement rather than self-serve.

Self-hosted and hybrid deployment: INFRO in your own cloud account, or hybrid: the gateway in your infrastructure and the console hosted by INFRO. Scoped and agreed per contract; self-serve plans run on the hosted, multi-tenant service.
Contractual SLA: An availability commitment with remedies, agreed per contract. There is no published SLA before the gateway has an operating history.
Invoiced subscription and custom terms: The platform subscription invoiced quarterly or annually on your terms rather than charged to a card, with volume and retention agreed rather than taken from the published ladder. Model usage stays on your own provider invoices.
Named support and escalation: A named engineer, a shared channel, and agreed response times for incidents. Everyone else gets the same inbox the founders read.

On the roadmap

Deliberately after the first release. No date, no commitment.

Anomaly detection: Latency and error-rate changes surfaced without a threshold being configured. Spend anomalies are covered by spend alerts; route-health detection is part of the core routing system. This is everything beyond both.
Cost recommendations: Automatic analysis of where a workload could move to an equivalent model with a lower provider rate.
Spend change analysis: Attributing a month-over-month bill increase to request volume, output length, model mix, and new workloads.
Usage per end customer, for billing: Requests, tokens and estimated provider cost for each end customer, by day or month, project and model, as a CSV from the console or JSON from the API — for you to bill your own users from in your billing system. From the Team plan.
Regional processing: Pinning an organization or project to a region requires infrastructure we have not bought yet. We will not claim it before it is enforced.