Frequently Asked Questions

Actualyze is an enterprise AI governance platform and AI gateway, one control plane between your applications and AI providers. Each call routed through it is checked for access, held to its budget, scanned for PII, routed, and recorded in an append-only audit trail. Team access and budget authority are separate: who may use a model and which budget pays are decided by different roles.

Actualyze is built for the teams that answer for enterprise AI. Platform engineering gets one governed path to every approved model without becoming the bottleneck. Security and compliance teams get guardrails on every request routed through it, SSO, and an append-only audit trail. Finance and FinOps teams get spend attributed to budgets, teams, and cost centers, with hard monthly limits and monthly statements.

An AI gateway is a proxy between your applications and model providers: one endpoint for many models, with routing, failover, and usage logging. Many gateways add budgets, guardrails, and central policy. Actualyze does that job and adds the structure an enterprise runs on: Team Owners approve what their own team deploys, Budget Managers own the budgets that pay for it, and neither overrides the other. Each model's approval requirement sets who signs off before a team can deploy it. Our AI gateway comparison sets Actualyze beside the other AI gateways, row by row.

You can start using Actualyze with a 30-day free trial from the Get started page, or book a demo to walk through your setup with the team. Pricing details are on the pricing page.

The hosted platform has been generally available since April 2026. On-prem deployment is coming in 2027 and will run on your own infrastructure, with weights and data inside your boundary. SOC 2 Type II is in progress, and the hosted platform carries a 99.9% uptime SLA with encryption in transit and at rest.

To connect an AI client or application to Actualyze, point it at Actualyze instead of your model provider. Most AI clients, SDKs, and apps let you set a custom base URL, so you swap in the Actualyze URL, add your Actualyze API key, and choose a model. Your code stays the same, and every call now runs through your access rules, budgets, and guardrails. Our Labs section has step-by-step setup guides for Claude Code, Codex, Cursor, GitHub Copilot, and other popular clients.

Actualyze works with any model served behind an OpenAI- or Anthropic-compatible API. Providers include OpenAI, Anthropic, Google, Amazon Bedrock, Azure OpenAI, and OpenRouter. vLLM or Ollama servers you already operate connect through a relay you run, which only makes outbound connections. Model Groups let the same model serve through any provider a caller may already use.

Actualyze governs AI coding tools the same way it governs any application: route them through it and every call gets your access rules, budgets, and guardrails. A client that lets you set an OpenAI-compatible or Anthropic-compatible base URL can connect, and our Labs section has the exact settings for Claude Code, Codex, Cursor, GitHub Copilot, OpenCode, and others.

Actualyze governs agents the same way it governs applications: every model call an agent makes runs through access rules, its budget, and guardrails. Tool-call controls cover MCP and ordinary function calling, blocking risky tool calls with allowlists, URL filtering, and parameter scanning, and scanning tool results before they reach the model. Give each agent its own API key and set a monthly limit on that key, so one runaway agent stops at its limit instead of draining a shared budget.

Actualyze routes model traffic through Virtual Models. A Virtual Model is a named set of models behind one model name that you define. For each request, a routing model Actualyze trains picks the cheapest model in the set strong enough to serve it, or leans toward quality if you set the Virtual Model that way. Actualyze keeps that routing current as new models arrive. When a model errors or is slow to start responding, the request fails over down an ordered list. Model Groups let the same model serve through any provider your callers can already reach, so one model name works across providers and another provider's route can take over when a route fails.

Actualyze controls AI spend through budgets. Every API key is bound to one team and one budget when it is created, and every call charges exactly that budget. When a budget reaches its monthly limit, the next call stops with a 402 before any model is contacted, and one alert threshold warns you before that point. A key's holder can also set a monthly limit on the key itself, and an optional workspace COGS ceiling stops spending at a limit on confirmed provider cost, with estimated spend shown beside it as exposure. The Control AI Cost use case shows how budgets, key limits, and chargeback fit together.

Actualyze charges every call to a budget and records which team made it, so you can see spend by budget, team, and cost center as it happens. Give each agent its own API key and you can track and limit its usage separately. You choose how costs are charged back, from passing provider costs straight through to adding a margin, and finance gets exportable ledgers and monthly statements for reconciliation.

Actualyze prices every call as it runs, at the route level: the specific model and provider endpoint that served it. It uses the most trusted rate it has, in order: a rate your team confirmed for that route or model, the seller's list price, then the catalog's reference price. Models on your own hardware carry a $0 cost basis, and you can still charge that usage back internally. If a call can't be priced, Actualyze flags it for your FinOps team instead of recording it as free, and you decide whether those calls are allowed or blocked.

Actualyze guardrails mask PII by default on prompts and responses, using a fixed catalog of built-in shapes (personal names and dates can be turned on), and each guardrail runs in monitor mode (log only) or enforce mode (mask). For agents, tool-call controls block risky tool calls with allowlists, URL filtering, and parameter and result scanning. Guardrails mask PII, not clinical health data. Data you forward to a SIEM carries metadata only, never prompts or responses. More detail is in the Secure AI Usage use case.

No. Actualyze doesn't store your prompt or response content. Each request is recorded as metadata: the request, the model selected, the route taken, and the cost attributed. Guardrail detections record what kind of data was found, never the content or the masked value, and Data Forwarding streams to your SIEM carry metadata only, never prompts or responses. For agent tool calls, the record holds the tool name, its parameters, and a summary of what it returned.

Actualyze records every governance action in an append-only, immutable audit trail, and a tamper attempt is itself recorded. Each request enters the trail as metadata only: the model selected, the route taken, and the cost attributed, never the prompt or response. Retention is unbounded until a Workspace Admin sets a window, and a declared regime such as SOX sets the minimum. Records export as CSV or JSON, or stream to your SIEM over OTLP through Data Forwarding.

Actualyze supports your HIPAA, SOX, and GDPR programs with the controls auditors ask about: an append-only audit trail with retention you can set to a regime's minimum, such as SOX's seven years, PII masking on prompts and responses, SSO, and model access granted through teams, with every call charged to a budget. Guardrails mask PII, not clinical health data, so HIPAA workloads still need their own PHI controls. The Prove Compliance use case walks through the audit trail, retention, and export.

Actualyze reduces shadow AI by making the sanctioned path the faster one. Teams get one endpoint for every approved model, access they request through the Model Catalog, and usage that is metered and attributed to a team and budget. You can verify your email domain and require SSO for its members. A readiness view lists who still signs in by a method enforcement would refuse, and monitor mode lets you audit before you enforce. Actualyze is not a network scanner, and it doesn't meter subscriptions bought outside it. The Eliminate Shadow AI use case lays out the full approach.

Actualyze supports SSO and SCIM directory sync. People sign in through your SSO identity provider (SAML), teams can be imported from directory groups, and SCIM directory sync deprovisions people who leave. Deactivating a member ends their sessions and revokes their API keys in that workspace.