About Arrow
Arrow is a Windows desktop platform for building, running, and scheduling intelligent automation entirely on local hardware. Record any interaction once, compose it into a visual workflow graph, and execute it deterministically — with local AI, local vision, local OCR, and local speech.
What Arrow is
A visual automation platform
Workflows ("chains") are node graphs, not scripts. Recorded desktop sequences and browser flows are first-class nodes you connect, branch, loop, and nest.
A deterministic engine with embedded AI
The graph is a symbolic program. LLM nodes, conditional gates, decision inputs, and the orchestrator add fuzzy reasoning only at the seams.
An agent platform
Agent Mode turns chains into tools for a local model. The agent's "mind" is a system chain you design — up to an orchestrator root that routes a goal.
Everything runs on-device: bundled GGUF models via llama.cpp, ONNX vision, PaddleOCR, Piper TTS, and Vosk STT. Arrow works offline and in air-gapped environments.
Capability map
| Capability | What it does |
|---|---|
| Node-graph editor | Author, edit, expand, and run chains in a visual canvas. |
| Desktop recorder | Capture clicks, typing, scrolls, drags, clipboard, screenshots, and UI elements. |
| Web recorder | DOM-level browser capture via CDP; selector-based sessions. |
| Execution engine | Dependency-gated scheduler over typed nodes; branches, loops, sub-chains. |
| Local AI | llama.cpp GGUF and Ollama; chat, generate, and embed endpoints. |
| Semantic retrieval | ComoRAG probe-driven retrieval, embeddings, and Graph RAG. |
| Vision & OCR | Screen understanding, ONNX object detection, and PaddleOCR. |
| Speech | Piper neural TTS and Vosk STT for voice mode. |
| Agent mode | System-chain-driven agent; chat overlay, phone client, chains-as-tools. |
| Orchestrator | Goal loop over mini-brain chains with trace and synthesis. |
| Memory | In-run context, SQLite persistence, RAG, and consolidation. |
| Scheduling | Once, daily, weekly, interval — JSON-persisted, in-process runs. |
| Sandbox | Run automation inside an isolated Windows RDP session. |
| Agent export | Compile a system chain and its assets into a standalone .exe. |
The three pillars
Editor
The graph editor and every runtime surface: typed nodes, configuration dialogs, chain library, scheduler UI, and the agent chat.
Player
The execution engine and all input/output machinery: interpreter, desktop recorder, web recorder/replay, and grounding.
AI
Local inference and retrieval: a FastAPI server over llama.cpp and Ollama, embeddings, ComoRAG/GraphRAG, and the Laya decision engine.
Why Arrow
Local-first, offline
All inference, storage, and licensing work without a network. Nothing leaves the machine.
Deterministic, not guesswork
The graph is a program. AI is used only where fuzzy judgment adds value, with fail-closed defaults.
Small-model discipline
Every model interaction is shaped for 1–4B models: stepped protocols and single-token contracts.
Inspectable and reversible
Logs, memory panels, trace ports, blocked states, and rated outputs — and ESC always aborts.
How Arrow keeps up with frontier models
Arrow competes on orchestration, not raw model size — it gets more out of small local models than a single large model gets out of itself.
- Integrated coding agent — an Aider-style local-agent loop in the Code Node Studio writes, edits, and debugs Python inside the graph.
- Apps composed from nodes — sequence, web, LLM, vision, code, memory, and MCP nodes wired by ports into an application, not a script.
- Anti-detection by design — jittered clicks, curved mouse paths, variable timing, native CDP input, and a durable browser profile.
- Small models, big results — stepped protocols, single-token contracts, and a typed decision engine (Laya) let a 3B local model do what frontier models do.
- Learns from verified runs — verified goals freeze into learned chains and memory consolidates, so the library improves itself.
- Element-first grounding — resolve the live UI element before acting, surviving moved windows, DPI, and theme changes.
Ready to automate?
Turn desktop automation from simple macro recording into intelligent, visual, AI-powered orchestration — entirely on your machine.