A privacy-first desktop AI agent that runs locally, thinks deeply, and remembers everything. 30+ tools · 8+ providers · One orchestrator.
Ork is a cross-platform desktop AI agent that runs entirely on your machine. It orchestrates intelligent agents across multiple LLM providers, remembers what it learns, and coordinates specialized subagents — all without sending your data to the cloud.
All agent execution, file access, and LLM routing happens locally. Your code never leaves your machine.
Route to Anthropic, OpenAI, Groq, Ollama, and more — switch providers at runtime without restarting.
Local embeddings + vector search. Ork learns from every session and recalls knowledge across projects.
Spawn specialized subagents with scoped tools. Coordinate teams of agents working together on complex tasks.
Every interaction follows a governed agentic loop — prompt, reason, act, observe, repeat. Each action is committed to a tamper-evident session chain.
Automatic retry on tool failures, context overflow recovery, and auto-continuation when responses are truncated.
Inject messages mid-loop to redirect the agent without interrupting. Guide complex tasks in real-time.
SHA-256 hash-chained append-only log. Every action is a block linked to the previous — tamper-evident governance with Merkle checkpoints.
Ork doesn't just chat. It learns, remembers, and gets smarter over time. A local embedding model indexes knowledge into a semantic store that persists across sessions.
Each persona is a self-contained personality pack — custom instructions, tools, permissions, and MCP servers. Switch between specialized agents for different tasks.
Clean code expert. Development, refactoring, TDD.
System design master. Planning, trade-offs, scalability.
Data insights specialist. Analytics, visualization, patterns.
Deep investigation agent. Multi-source synthesis, documentation.
Witty content creator. Writing, brainstorming, storytelling.
Security-focused auditor. Vulnerability detection, compliance.
Infrastructure automation. CI/CD, deployment, monitoring.
Lead qualification. Rule-following precision, validation.
General-purpose assistant. Balanced capabilities for any task.
Meta-agent that creates and validates new personas.
Go beyond single-agent workflows. Ork spawns specialized subagents in isolated processes, coordinates teams with shared task lists, and orchestrates complex multi-step workflows.
Spawn specialized child agents in isolated processes. Each gets scoped tool access — an explore agent can't write files. Choose the model independently per subagent.
Multi-agent teams with shared task lists, inter-agent messaging, and coordinated workflows. Agents communicate through a message bus, report results, and hand off tasks autonomously.
Understand how you work. Ork tracks usage patterns and generates self-contained analytics reports — tool usage, session patterns, provider breakdown, response times, and language breakdown.
Built-in tools
LLM providers
Agent personas
Binary size
30+ built-in tools organized by domain. Extend with MCP servers for custom integrations.
Built for performance and privacy. Minimal footprint, maximum capability.
A native desktop experience. Fast, focused, and private.