Stage 1: Machina Core ✓
Local orchestrator, planning and execution modules, tool registry, session memory, action logs, multi-agent coordination, security foundation, and MCP protocol support. Delivered in v0.8 and v0.9.
The Age of Neural-Native Computing
MachinaOS does not replace your kernel. It adds a higher semantic layer above the host OS where goals, plans, approvals, tools, and timelines are first-class runtime objects.
Is MachinaOS available today?
MachinaOS is an MVP with a live private demo. The current release is v1.4. Installers are not publicly listed: evaluators can request source access and a signed desktop build (Windows or Linux) through the contact page.
The runtime supports local execution with a local model provider such as Ollama. Selecting a cloud model or network tool introduces external requests; local-first does not mean that every workflow is offline.
Portfolio products operate independently today. Their integration through MachinaOS is a development direction, not a prerequisite for trying the existing demos.
The Mission
Synchronize intent and execution
MachinaOS takes natural-language goals, normalizes them into structured plans, applies policy checks, executes approved steps through adapters, and returns auditable outcomes with full event traces.
Development Stages
Stage 2: Machina Shell ✓
Native desktop application (Windows and Linux), command palette, visual workflow studio, file explorer, process monitor, 17-view interface, and system status bar. Shipped in v0.9; signed with auto-update since v1.0.
Stage 3: OS Layer
Background services, persistent policy engine, semantic file features, organization-wide observability, and shared agent mesh for team deployments.
Engineering Principles
Explicit tool boundaries
The model never pretends to act. Every operation is grounded in a real invocation and a structured response.
Deterministic logs
All meaningful actions are recorded so runs can be audited, explained, and reproduced. The immutable audit log uses SHA-256 hash chains to detect tampering.
Human approval for risk
Destructive or high-impact actions trigger policy checks and require explicit confirmation. Consensus workflows gate sensitive tool calls across multiple agents.
No AI theater
Avoid vague claims. Ship practical capabilities with clear constraints and reversible mutations.
Inspectability over opacity
Every plan, step, tool result, event, and agent message is accessible and searchable. Nothing useful happens invisibly.
Local-first by default
Plans, secrets, logs, agent definitions, and workflow chains all live on your machine. Cloud services are opt-in LLM providers, not infrastructure dependencies.
Origin Story
Built from a real engineering problem
MachinaOS started from a simple frustration: AI assistants that could talk about what they would do, but not actually do it in a governed, auditable way. Every early AI agent framework required writing orchestration glue code, managing prompt chains in Python, and accepting that the "execution" was largely model-generated text rather than real tool calls with real outputs.
The goal was a runtime layer — something that sat above the OS and treated intent, plans, approvals, and tool execution as first-class objects, with a native desktop shell and a human in the loop by design. From v0.8 to v1.4, with 3,200+ passing tests, that runtime now exists. MachinaOS is the result.
Roadmap
v0.8 – v0.9 — Runtime and Studio
Chain execution runtime, planner, Visual Workflow Studio with step-by-step debugger, natural-language chain generation, and MCP as both client and server.
v1.0 – v1.3 — Desktop and Governance
July to September 2026: signed desktop app with auto-update, enterprise identity and signed policy packs, two-key approvals for critical actions, governed MCP connections, and local (LM Studio) plus frontier models out of the box.
v1.4 — Current
October 2026: governed workflows that pause for approval, a kill switch, watchers and webhooks, BPMN export, streamable-HTTP MCP, and an optional LangFlow connector. 17 views, 80+ native tools, 7 LLM providers, 50+ templates.
Next — OS Layer and Public Availability
Background services, a policy engine running independently of the UI, semantic file features, and a shared agent mesh for small teams. Public availability, licensing, and documentation will be announced; get in touch to be notified.