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.

research-analyst.vioYAML · SIGNED
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
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.
Terminal demo: vio validates a .vio file and Helix deploys the AI agent

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. Supported targets. Clear responsibilities.

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.

→

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.

Deploy a .vio file from a terminal or CI with the Viorant CLI.

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.

1
Declare the system

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

⇢
2
Version and sign assets

Artifacts and modules are versioned and cryptographically signed.

⇢
3
Helix resolves dependencies and policies

Helix resolves dependencies and validates policies.

⇢
4
Prepare a supported target

Helix transforms the definition for explicitly supported environments.

Questions, answered.

01 What 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.

02 Does it lock me into Viorant?

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

03 How is Hub different from Helix?

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

04 What is Viorant Cloud?

Viorant Cloud is our managed infrastructure option for running workloads.

05 What is Sentinel?

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

06 How does signing work?

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

07 Can 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.