MAAX Studio
The execution layer for operational AI. Missions, gates, and ledgers as first-class primitives — running in your cloud, under your governance.
MAAX Studio runs AI work as missions, not pipelines. Every step is a candidate action that passes through command gates and is recorded in a mission ledger. The model is one component; the system holds the goal, the context, the gates, and the audit trail.
At a glance
What MAAX Studio is
A governed execution environment for agentic AI workflows. Built on the MAAX Runtime — the underlying architecture that coordinates agents, gates, context resolution, and evaluation.
Core primitives
- Missions — explicit goals with acceptance criteria, owners, and escalation rules.
- Project context graph — the structured knowledge each mission draws from.
- Goal-grounded generation — every model call carries mission, context, and acceptance criteria.
- Command gates — independent checkpoints between candidate output and downstream action.
- Mission ledger — full audit and regression replay of every decision.
- Human-governed delivery — explicit escalation paths, not implicit overrides.
Runtime architecture
Missions, acceptance criteria, owners, and escalation rules — the durable contract every model call is measured against.
Structured, permissioned retrieval from your systems with provenance attached to every field entering a model call.
Policy-driven routing across OpenAI, Anthropic, Google, and open-weight models based on capability, cost, and data class.
Independent policy, structure, task-quality, and safety gates that must pass before any downstream write.
Append-only record of every candidate action, gate verdict, escalation, and outcome — the substrate for audit, replay, and evaluator training.
Automated evaluators that run on a schedule and gate model or prompt promotion against real historical missions.
Failed gates land with the correct human or queue with full context, SLA, and reopen semantics.
Cockpit surface for mission health, gate calibration, ledger inspection, and evaluator management.
Capability areas
Policies compile into runtime gates. Data-class rules, tenant scopes, and action-type permissions cannot be bypassed by prompt engineering.
Every action carries mission ID, gate verdicts, model provenance, and cost — replayable end-to-end from the ledger.
Missions declare intent; the router picks the model. Swapping vendors is a config change validated by evaluator replay.
Mission health, escalations, and gate calibration surface in one console — designed for the operators who own the system in production.
Who it's for
- Teams shipping AI-native products who need a runtime they can audit.
- Operators replacing ungoverned agent stacks with a governed execution layer.
- Engineering leads who want missions, gates, and ledgers as first-class objects, not custom code.
How it differs
MAAX Studio is not a chat UI, an automation builder, or a generic agent framework. It is an execution OS: the system holds the mission and verifies every action. The model is replaceable; the governance is the product.
| Concern | Agent framework | MAAX Studio |
|---|---|---|
| Unit of work | Loop of model calls | Mission with acceptance criteria |
| Policy enforcement | Prompt discipline | Independent runtime gates |
| Audit trail | App logs (if any) | Append-only mission ledger |
| Model swap | Rewrite prompts + tools | Config change + evaluator replay |
| Failure handling | Retry the loop | Escalate with full context |
Security & deployment posture
- Runtime deploys inside your VPC on AWS, GCP, Azure, or Cloudflare — never a shared multi-tenant control plane for governed workloads.
- Data plane, secrets, and model keys stay under your existing IAM and KMS.
- SSO / SCIM through your IdP (Okta, Entra ID, WorkOS, Google) for the operator console.
- Ledger tables designed for retention policies, legal hold, and export to your data warehouse.
- Source-available option for teams that require code review as part of procurement.
Cyryx makes no compliance certification claims. MAAX Studio is engineered to make certification programs (SOC 2, ISO 27001, sector-specific frameworks) achievable on your infrastructure; the certification is yours to hold.
Where MAAX Studio fits the stack
- Above your model vendors — they supply capability; MAAX supplies governance.
- Alongside your agent frameworks — they compose steps; MAAX gates and audits them.
- Above your workflow engines — they schedule; MAAX defines missions and outcomes.
- Under your product surfaces — they render; MAAX guarantees what they render is grounded and governed.
Roadmap themes
Directional themes shared publicly; specific dates are shared under early-access agreements only.
- Runtime hardening — deeper policy-engine integration, richer gate primitives, and stronger evaluator DSL.
- Operator console — mission health, gate calibration, and evaluator workflows aimed at day-2 operability.
- Ecosystem — first-class adapters for the major agent frameworks and enterprise policy engines.
- Governance packs — reference configurations aligned to common regulatory frames, without claiming certification on our behalf.
Questions we get from procurement
Q.Is MAAX Studio a framework, a platform, or an application?
It is an execution environment — the runtime that sits between an operator's intent and the model. Missions, gates, ledgers, and evaluators are first-class objects, not code you assemble from a framework. The application surfaces those primitives; the runtime enforces them.
Q.How does it differ from LangChain, LangGraph, AutoGen, or CrewAI?
Those are agent frameworks — libraries you use to compose loops of model calls. MAAX Studio is the layer above them: a runtime that treats each agent step as a candidate action passing through independent command gates and landing in an append-only ledger. You can run agents built in any of those frameworks under MAAX governance.
Q.Which models can it run?
Any. Model routing is a first-class runtime concern. Missions declare requirements (capabilities, cost ceiling, data class); the runtime routes to the eligible model per call. Swapping a model is a routing change validated by evaluator replay, not a rewrite.
Q.Where does it run?
In your cloud of record. MAAX Studio deploys as a self-contained runtime on your VPC (AWS, GCP, Azure, or Cloudflare) with your data plane. Cyryx does not operate a managed multi-tenant SaaS for governed workloads — governance depends on the runtime living where your data does.
Q.Is it available today?
MAAX Studio is in active development with a small cohort of early-access customers. New engagements start through Cyryx Solutions, which builds on the runtime and graduates teams to a standalone MAAX deployment as it stabilizes.
Q.How does licensing work?
Runtime license per environment plus a support tier. Customer data never leaves your VPC, and the runtime is deliverable as source under a commercial license for teams that require it. Pricing is contract-based; no self-serve tier during early access.
Status
MAAX Studio is in active development at Cyryx Labs. Early access engagements are coordinated through Cyryx Solutions. The underlying frameworks are published openly by the Cyryx Applied AI Lab.