Visual Workflow Studio
SVG canvas with drag-and-drop palette, smart argument builder for 30+ tools, zoom and pan, auto-assign, optimize, and one-click load of 50+ chain templates. Build multi-agent pipelines without writing code.
Local-first orchestration
Turn Repo Work Into Governed AI Workflows
Use MachinaOS to review code, audit security, onboard into unfamiliar repositories, and build repeatable agent workflows without giving up control. It turns a natural-language request into a visible plan, runs real tools on your machine, gates risky steps, and records an inspectable timeline.
MachinaOS is an MVP with a live private demo. Try a guided workflow, or request an evaluator build to run it on your own codebase. Current availability and execution options.
◆ Private repository, open doors. Evaluators receive the full source, the policy engine definitions, a SHA-256 audit-chain you can verify yourself, and the 3,200+ test suite. Request evaluator access.
See It Work
One loop, every time: you state the goal, MachinaOS shows the plan, a person approves anything risky, and every step is traced and verifiable.
-
1
Intent in Neural Link
Describe the goal in plain language. No prompt engineering, no scripts.
-
2
Visual plan in Studio
The request becomes an explicit graph of real tools, validated locally before anything runs.
-
3
Policy approval gate
Risky steps stop for a human. Blocked actions are refused before they reach the system.
-
4
Debugger and audit trail
Step through the run live, then verify the hash-chained audit log: nobody can quietly edit history.
The recent demos use Claude as the planning model. MachinaOS runs the same way on local models (Ollama, LM Studio) or any other provider: local or cloud, your choice.
Core Capabilities
Built up across releases v0.8 to v1.4: natural-language workflow generation, webhook triggers, a visual pipeline editor with a step-by-step debugger, chain composition, an inter-agent communication graph, MCP as both client and server, and a Neural Link that answers from real tool results.
Studio Debugger
Server-side debug sessions with Step Over, Continue, and Stop. Each canvas node animates in real time — amber pulse for running, green ✓ for success, red ✗ for failure — with per-step duration metrics in a resizable results tray.
Chain-to-Chain Composition
Treat any saved workflow chain as a reusable building block. Drag chain nodes from the Studio palette, embed them in larger pipelines, and execute recursively with depth limiting and cycle detection.
Agent Communication Graph
Live SVG visualization of inter-agent traffic across six communication channels — requests, negotiations, handoffs, broadcasts, consensus, and collaborations — with a unified activity feed and message threading.
MCP Integration
Full Model Context Protocol support in both directions: consume external MCP servers (memory, fetch, GitHub, SQLite, or your own over stdio, SSE, or streamable HTTP) and expose MachinaOS's native tools, resources, and prompt templates to Claude Desktop, Cursor, and other MCP clients, behind the same policy gate.
Enhanced Neural Link
Conversational command interface with streaming plan cards, RAG-augmented context, and natural-language delegation. Type your intent, watch the plan form, approve risky steps, and inspect every tool result.
NL Chain Generation — Describe It, Build It
Type a sentence in the Neural Link or click ✨ Generate in Studio. MachinaOS sends your description to the active LLM using native tool-calling — no JSON parsing, no prompt hacks — and returns a fully wired, versioned, executable workflow chain. Works with any configured provider, cloud or local.
Webhook Triggers
Any saved chain can be triggered by an external system via a stable HTTP POST endpoint. CI/CD pipelines, monitoring alerts, and scheduled jobs all work. The endpoint sits behind the same API token as the rest of MachinaOS, and the JSON body becomes the workflow's input. Executions appear in the same audit trail as manual runs, with a 20-run invocation history per webhook.
Core Architecture
Heuristic Intent Router
Fast deterministic routing for commands like file listing, git checks, and diagnostics before model fallback.
Planner and Executor Split
The planner proposes steps and assigns tools; the executor runs only approved actions with policy constraints and full event tracing.
Delegation and Multi-Agent Runtime
Specialized agents handle subtasks with tool-scoped permissions, capability scoring, health monitoring, auto-healing, and explicit execution traces.
Memory and Secrets Layers
Session, preference, and per-agent memory keep context stable. AES-256-GCM encrypted secrets vault keeps credentials off disk in plain text.
Security Center
Built-in vulnerability scanner, append-only audit log with SHA-256 hash chains, credential scrubber, and secrets vault in one panel.
Local-First Runtime
Planner, tool registry, memory, secrets vault, and audit log all run on your machine. Use Ollama for fully local LLM inference, or opt into cloud providers per workspace.
What It Does
From intent to governed execution
Machina turns a goal into a structured plan, runs it with real tools, and shows every step — approvals, events, and runtime state included.
Understand a project in minutes
Scan an unfamiliar workspace, detect stack clues, inspect key files, and explain where to start.
Debug with governed execution
Combine diagnostics, code search, Git context, and safe tool execution in one visible workflow.
Turn routines into reusable workflows
Capture repeatable engineering tasks as inspectable AI-native execution chains with 50+ built-in templates.
Run from your desktop
Signed Windows and Linux desktop app with a bundled Python runtime and auto-update. No prerequisites, no cloud dependency. Provided as an evaluator build.
Stay secure by default
AES-256-GCM encrypted secrets vault, automated credential scrubbing from outputs, and a tamper-evident audit log built into every execution.
Build visually, ship reliably
Design multi-agent pipelines on a visual canvas, assign agents automatically, optimize for parallelism, and debug step-by-step with live feedback.
How It Works
1 — Understand intent
Machina turns a goal into a structured task with identified scope and required tools.
2 — Build a governed plan
It selects tools, orders steps, and marks risky actions for human approval before execution.
3 — Execute and show its work
Machina runs the workflow while exposing task state, events, tool results, and runtime metrics.
System Snapshot
Inspectable runtime
Source and test suite available to evaluators on request
80+ Tools · 17 Views
18 tool domains, full desktop UI
7 LLM Providers
Ollama, LM Studio, OpenAI, Anthropic Claude, Google Gemini, OpenRouter, and any OpenAI-compatible endpoint
MCP Client + Server
Consume and expose Model Context Protocol
50+ Chain Templates
Plus agent and workflow DSL blueprints
Local-First
Plans, secrets, logs, and agents on your machine
Ready to evolve your workflow?
Run a practical intent-to-action loop with plan previews, approvals, tool outputs, and traceable results — on your machine, with your data.
Want to look under the hood? Evaluators get the full source, the policy engine definitions, a verifiable SHA-256 audit chain, the 3,200+ test suite, and a desktop build to run on their own codebase.
Request Evaluator Access View Pricing Preview