Is DMN free?
Yes. There's no paid tier and no account, and the app you install stays free. DMN doesn't sell AI, so there are no credits to buy and no markup: your agents run on your own Claude or Codex plan, billed by those providers as usual. DMN's job is to make that plan go further. The about page says why.
Does my code leave my machine?
No. Indexing, search, notes and lessons all run on your computer, and the engine only accepts connections from it. Your agents still talk to their own AI providers, as they do without DMN. Privacy and network lists every connection DMN makes, and what each sends.
Which agents does it work with?
Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Continue, Warp and any MCP client, through the skill, MCP or both. In the app's panes you can also run Aider and Cursor's CLI. See Connect your agents.
Which systems?
Windows 10 or 11, 64-bit. Mac and Linux builds are on the roadmap. On Windows, DMN can move a project into WSL and run agents' commands there.
Do I need a GPU?
No. DMN uses an NVIDIA GPU when it has the CUDA runtime, and the CPU otherwise. On the CPU, indexing runs at below-normal priority, so your build and your editor go first; searches run at normal speed.
Which languages does it understand?
It parses 41 languages into its structure graph (which function calls which). Search by meaning and by words works across the project.
Can I use it without the app?
Yes. The installer also puts the engine and the dmn command line on your PATH. dmn start runs the engine, and dmn skill-init --apply or dmn mcp-init --apply connects your agents. The app adds the workspace: panes, the Fleet panel, approvals and the rest.
Is the source public?
Not yet. DMN is free to use; the source isn't published.
Is it for teams?
It's built for one person running several agents at once. The notes in a project's .dmn/memories folder are plain markdown, so you can commit them to share with others on the project.
Does it slow my agents down?
A search takes milliseconds on a CPU, and saves the agent the requests it would have spent reading files. The benchmarks measured fewer requests and fewer tokens on code the model couldn't place from memory, and say where DMN doesn't help.
How does it compare with other tools?
The comparisons cover Claude Code's desktop app, Warp, and Conductor, Superset and BridgeMind.
Something else?
Write to support@getdmn.com.