Frequently asked questions.

An intelligent operating system for your business. You codify how the business runs. The OS does the work, across the tools you already use. The same AI you use today, resident in the runtime. Every action is policy-bounded. Every change is traceable.

Software that adapts as you teach it, instead of running fixed code. The runtime builds itself from how the business actually works, and reshapes when things change. RARS stands for reasoning-assisted recursive system. Poliglot OS is built this way.

The AI lives inside the runtime, not next to it as a service the app calls. Same models you already use today; different place in the stack. The AI reasons against the live state of the business and executes programs the runtime itself enforces.

Poliglot is the company: a Public Benefit Corporation chartered to amplify human agency, together with the platform and infrastructure behind the runtime. Poliglot OS is the product: the AI, and the runtime it inhabits.

Agent frameworks wire together stateless LLM calls. Every step is a fresh inference, plans drift, coordination is handled by prompts, and verification is an afterthought. Poliglot OS takes a different approach: one resident AI with shared context plans against your typed operating model, where every candidate action is bounded by the formal constraints of your business logic. The plan itself is produced by the AI, but once it's written, the runtime executes it deterministically. AI reasoning, human approvals, and system actions compose into one executable program held to a single set of I/O contracts. No message-passing between specialist agents. No prompt-based guardrails. Enforcement is structural.

A matrix is a versioned, composable specification of a business domain: the concepts, rules, actions, and systems of record it contains. Codifying your operating model means writing your HR, finance, procurement, or compliance logic as a matrix instead of leaving it scattered across people, spreadsheets, and SaaS tools. Matrices compose into a single connected operating model that Poliglot OS executes against.

A neurosymbolic runtime combines neural networks (LLMs) with a symbolic reasoning system. Poliglot OS's probabilistic layer handles language, judgment, and planning. The symbolic layer runs deterministic programs against your typed operating model, enforces validation, and tracks provenance. The two layers are interleaved in the same execution loop, so the AI plans and the runtime enforces. Neither side has to do the other's job.

Most agent frameworks reason and act in the same loop: reason, take a step, observe, reason again. Every step is a fresh inference, which means plans drift and multi-step work compounds error. Poliglot OS inverts that. While the AI is exploring the graph, traversing the ontology, and gathering context, it loops. When it decides to act, it composes a complete program of interdependent stages (deterministic calls, AI reasoning, human approvals) and hands it to the runtime. The runtime executes the program as written. The AI does not re-infer at each step.

Every action Poliglot OS can take has an explicit input/output contract validated against your business constraints. That includes an LLM reasoning step, a human approval, and a deterministic API call. All are held to the same contract. The contract guarantees the output structure regardless of how the result was produced, so downstream consumers see the same shape whether a model or a service produced it.

When AI writes code, you don't scrub the reasoning chain. You read the diff. Poliglot OS gives you the same experience for business operations. After a session runs, the changes Poliglot OS made to your operating resources (contracts, procurements, reconciliations, budgets) are surfaced as a structured diff, with per-statement provenance for every addition and retraction. Full execution traces are still there if you need them. Most of the time, you won't.

Role-based access control grants permissions unconditionally: if your role allows it, the action goes through. That is dangerous for AI. Poliglot OS enforces situational access control: at execution time, policies evaluate the live state of the business. Does the procurement have a completed risk assessment, a licensed engineer sign-off, and budget headroom? If any condition fails, the operation is denied regardless of the actor's role.

Every change in the system, down to a single field, is recorded as a structured observation with full provenance: who changed it, through which process, under what authority, and when. That includes actions taken by humans, by Poliglot OS, and by matrix-provided agents. Accountability isn't a forensic investigation across logs. It's a direct lookup against a unified audit surface.

Poliglot OS is model-agnostic. You can Bring Your Own Model (connect a provider account you hold directly), or use our Managed Model Router, which routes inference to supported providers on your behalf. You can standardize on one provider across your workspace or mix per matrix. The runtime does not care which you choose.

Poliglot ships in two deployment tiers today, with a third on the roadmap. Public cloud is shared, multi-tenant, currently running in US regions, with AES-256 at rest and TLS 1.2+ in transit. Private cloud (business and enterprise plans) gives you a dedicated tenant with its own logical database, object storage, and per-tenant encryption keys. Customer-VPC deployment is next on the roadmap. Poliglot does not train or fine-tune AI models on your workspace data; this commitment is absolute and applies to every tier, and we do not ship proprietary models. We collect aggregated operational telemetry (action counts, error rates, latency) to run and improve the Service. During the Beta period, public cloud workspaces are subject to expanded operational analytics so we can evaluate functionality and better support customers; this ends when public cloud exits Beta. Full detail is on the Security & Privacy page.

Public beta opens May 2026. Private cloud follows in June 2026. Customer-controlled deployment is on the roadmap after that. You can request early access through the waitlist.

Poliglot OS is a horizontal runtime, not a vertical product. The capability it provides is executing your operating model, whatever domain it comes from. Any organization with complex, cross-system workflows can codify a matrix for its shape. Early pilot conversations are clustered in regulated and operations-dense domains, where the cost of an unaccountable action is the business itself, but the platform is not industry-bound.

Request early access through the waitlist for the public beta. Pilot slots are limited and prioritized by domain fit. For ongoing work: join the Discord, follow the GitHub organization, and watch for the open-source runtime launch in summer 2026.