Accomplish: The Open Source AI Desktop Agent That Actually Does Things

By Prahlad Menon 4 min read

Accomplish: The Open Source AI Desktop Agent That Actually Does Things

Most AI tools talk about doing things. Accomplish actually does them.

Accomplish (formerly Openwork) is an open-source AI desktop agent that runs locally on your machine. It can manage files, create documents, browse the web, and automate repetitive tasks — all without a subscription and without your data leaving your computer.

The Difference Between Chatbots and Agents

Here’s the distinction that matters:

  • Chatbot: “I can help you organize your Downloads folder. Here’s how you would do it…”
  • Agent: Actually organizes your Downloads folder

Accomplish is in the second category. Point it at a folder, tell it what you want, and it moves files, renames them, creates documents, or researches topics on the web. You approve each action, but the AI does the actual work.

Why This Matters Now

Claude just launched Cowork (their enterprise AI coworker). It’s powerful but expensive, runs on Anthropic’s servers, and requires sending your files to the cloud.

Accomplish is the open-source answer:

  • Runs locally — your files never leave your machine
  • Bring your own API — use OpenAI, Anthropic, Google, xAI, or run completely free with Ollama
  • MIT licensed — fork it, modify it, do whatever you want
  • No subscription — pay only for the API calls you make (or nothing with local models)

The Browser Advantage

Most local AI tools hallucinate when you ask them to research something. They can’t actually look things up — they’re just guessing based on training data.

Accomplish has a built-in browser engine. You can tell it:

“Go to this documentation page, read it, and summarize the key points into a file in my Project folder.”

It actually navigates to the URL, reads the content, and creates the file. No guessing.

What It Actually Does

CapabilityExample
File Management”Organize my Downloads folder by file type and date”
Document Creation”Write a summary of these meeting notes”
Browser Tasks”Research this topic and save the findings”
Custom Workflows”Every Monday, compile my weekly status report”
Calendar Integration”Create calendar entries from these meeting notes”

The key differentiator is that it executes these tasks — it doesn’t just tell you how to do them.

Running It for Free with Ollama

You don’t need to pay for API access:

  1. Install Ollama
  2. Run ollama run llama3 in your terminal
  3. Open Accomplish and select “Local” in settings

Now you have a completely private, completely free AI agent running on your hardware.

How It Works Under the Hood

Accomplish is built with Electron and React. The interesting part is that it spawns OpenCode CLI using node-pty to execute tasks. API keys are stored securely in the OS keychain.

The architecture is clean:

  • Main Electron process handles system integration
  • React UI for the interface
  • OpenCode CLI for task execution
  • Sandboxed folder access (you choose what it can touch)

The Privacy Model

Every action is:

  1. Shown to you before it runs
  2. Requires approval (you can see exactly what it’s about to do)
  3. Logged so you can review what happened
  4. Stoppable at any time

Your files stay local. The only external communication is with your chosen AI provider using your own API key.

Supported Providers

  • Anthropic (Claude)
  • OpenAI (GPT)
  • Google AI (Gemini)
  • xAI (Grok)
  • DeepSeek
  • Amazon Bedrock
  • Azure Foundry
  • OpenRouter
  • LiteLLM
  • Ollama (completely local, no API needed)
  • LM Studio (completely local, no API needed)

Getting Started

  1. Download from accomplish.ai (macOS or Windows)
  2. Drag to Applications
  3. Connect your AI provider (or Ollama for local)
  4. Choose which folders it can access
  5. Start asking it to do things

The whole setup takes about 2 minutes.

My Take

I’ve been watching the “AI coworker” space closely. Most tools are either:

  • Cloud-based with privacy concerns
  • Local but can only chat, not act
  • Powerful but require significant setup

Accomplish hits a sweet spot: it’s local-first, genuinely agentic (it acts, not just chats), and trivially easy to install. The browser integration is the killer feature — most local agents can’t actually research anything.

If you’ve been looking for something like Claude Cowork but don’t want the enterprise pricing or the cloud dependency, this is worth trying.

Links: