Define the complete AI system in .vio.

A portable AI system definition independent from its runtime.

.vio composes signed, versioned prompts, skills, memory, agents, models, connectors, policies, evaluations, and deployment requirements without erasing each artifact’s identity.

Join the waitlist →

research-analyst.vioYAML · SIGNED

manifest:
  name: Research Analyst System
  version: 1.0.0
  description: Agent for research,
    synthesis and reporting
  author: data-architect
  created: 2025-05-18
  contract_version: 1.0

frameworks:
  - crewai
  - langgraph
  - adk

assets:
  prompts:       /prompts/
  skills:        /skills/
  memory:        /memory/
  policies:      /policies/
  observability: /observability/

models:
  default:   openai:gpt-4o
  reasoning: anthropic:claude-3.5-sonnet
  embedding: openai:text-embedding-3-large

connectors:
  - name: vector_db
    env:
      VECTOR_DB_URL:     ${VECTOR_DB_URL}
      VECTOR_DB_API_KEY: ${VECTOR_DB_API_KEY}
  - name: search_api
    env:
      SEARCH_API_KEY:    ${SEARCH_API_KEY}

signing:
  algorithm: ed25519
  signer:    viorant-ci
  signature: a1b2c3d4...97fe

targets: [local, docker, cloud, self-hosted]

Framework-neutral definition

Keeps framework choices explicit; compatibility follows published support.

Multi-model management

Best-fit models per capability and cost.

Signed artifacts

Cryptographically signed & verifiable.

Provenance chain

Full lineage and change history.

Connectors via env vars

Secure, portable integrations.

Target requirements

Names the capabilities a supported deployment target must provide.

From scattered AI assets to one governed contract.

THE PROBLEM

Scattered prompts, skills, memory, and tools

Assets live in docs, code, and wikis, hard to find, reuse, and trust.

Local-to-prod drift

Different configs and models between environments create risk.

No signing or provenance

Changes aren't tracked, signed, or attributable.

Framework and cloud lock-in

Tied to one stack or provider, making portability impossible.

Hard to review AI stacks

Complex, non-standard configs are difficult to read and audit.

⇢

THE VIO FILE SOLUTION

One declarative file

Everything in one place. Human-readable. Versioned. Verifiable.

One definition across supported targets

Environment-aware requirements keep target differences explicit.

Signed artifacts and traceable lineage

Cryptographic signing plus provenance for audits and compliance.

Portable across supported targets

Helix and tested adapters prepare target-specific outputs from the same definition.

Human-readable and reviewable

Easy to diff, review, and reason about AI systems.

One file. Complete control.

Prompts

System, user, and agent prompts.

Skills

Functions, tools, and capabilities.

Memory

Vector stores, RAG and state.

Models

Named models and supported providers.

Agents

Behavior, routing, and orchestration.

Policies

Guardrails, approvals, and access control.

Observability

Logs, traces, metrics, and alerts.

How .vio fits AI Deployment Infrastructure.

One contract. Many environments. Clear responsibilities.

1HUB AUTHORS

Author and manage .vio system definitions in a local-first workspace, within the capabilities of the current release.

→

2.VIO DEFINES

The portable system definition composes agents, prompts, models, policies, connectors, and other artifacts.

→

3HELIX DEPLOYS

Resolves supported dependencies, applies policy, transforms the composition, and prepares a supported target.

→

4SUPPORTED TARGETS

Use only deployment targets named in the current compatibility matrix.

✓ Cloud✓ On-prem✓ Edge✓ Local✓ Self-hosted

SUPPORTING LAYERS

Viorant Cloud (verify current status)

A managed infrastructure option only where the current support matrix confirms availability and scope.

Sentinel (roadmap)

A proposed operational layer whose availability and implemented scope require product confirmation.

DEPLOYMENT TARGETS

Cloud On-prem Edge Local Self-hosted

Built for engineers. Designed for trust.

Human-readable

Clear, declarative, and easy to review.

Auditable

Every decision can be explained and verified.

Governed

Policies, approvals, and guardrails built in.

Portable

Use the same definition across supported targets.

Signed artifacts

Cryptographically signed and verifiable.

Modules & registry

Reusable modules with versioning and lineage.

Environment-variable secrets

No secrets in code. Secure by design.

Observability declared

Logs, traces, metrics, and alerts defined up front.

Multi-framework + multi-model management.

One system definition. Supported frameworks. Named models.

FRAMEWORKS

⬡ CrewAI⬡ LangGraph⬡ ADK+ More frameworks

VIO FILE (.VIO)

System definitionAgents and skillsMemory and connectorsPolicies and observability

MODELS & PROVIDERS

◆ OpenAI◆ Anthropic◆ Llama◆ Gemini+ More providers

How it works.

Declare the system

Define agents, tools, memory, models, policies, and observability in .vio.

⇢

Version and sign assets

Artifacts and modules are versioned and cryptographically signed.

⇢

Helix resolves dependencies and policies

Helix resolves dependencies and validates policies.

⇢

Prepare a supported target

Helix transforms the definition for explicitly supported environments.

Questions, answered.

01What is .vio?+

.vio is a portable definition that composes signed, versioned AI artifacts into a complete system for review, verification, and deployment to supported targets.

02Does it lock me into Viorant?+

No. .vio is portable and designed to remain useful without Viorant.

03How is Hub different from Helix?+

Hub is the optional authoring workspace. Helix transforms a .vio definition for explicitly supported deployment targets.

04What is Viorant Cloud?+

Viorant Cloud is our managed infrastructure option for running workloads.

05What is Sentinel?+

Sentinel is the observability, security, and control panel for deployed assets. Monitor, investigate, and enforce with confidence.

06How does signing work?+

Trust can attach and verify artifact evidence within its implemented proof scope. Verification does not prove broader correctness or compliance.

07Can I run on my own infrastructure?+

Customer-controlled deployment depends on a currently supported target and its documented limitations.

.vio gives the complete AI system
a portable definition.

The source of truth for every AI system you build.

Join the waitlist →Talk to us →

Viorant builds AI Deployment Infrastructure.