# 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 →](/contact) research-analyst.vio YAML · 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. ## 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. 1 [HUB AUTHORS](/hub) 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. → 3 [HELIX DEPLOYS](/helix) Resolves supported dependencies, applies policy, transforms the composition, and prepares a supported target. → 4 SUPPORTED 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 definition Agents and skills Memory and connectors Policies 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. [Join the waitlist →](/contact)[Talk to us →](/contact)