Martty

DeepSeek Harness guide

Using DeepSeek Harness in Martty

Install Martty as a DeepSeek Harness profile, launch the Rust terminal client, manage models and permissions, and resume durable coding-agent sessions.

· OpenMA

Install the Martty profile

Martty's recommended path lets DeepSeek Harness manage the terminal package and its ACP adapter. You need Node.js 22.19 or newer.

npm install --global @deepseek-ai/dsh
dsh plugin --profile martty add martty@latest
dsh --profile martty

The first command installs DSH. The second creates or updates a dedicated profile and mounts Martty with its runtime dependencies. The final command starts the Host and launches the Martty client.

Understand what is running

DeepSeek Harness runs once in the Host process. Its ACP plugin exposes the session over standard input and output. Martty starts a separate Cordis client process and a Rust painter. The client does not spawn another agent, so model state, tools, skills, and permissions stay with the original Host.

Choose model, mode, and permissions

Type /model to select a model advertised by the agent, /agent for an Agent Preset, and /permission or shift+tab to choose a permission mode. Use /auth when authentication is required. These controls reflect ACP capabilities rather than a static list baked into Martty.

Work with long-running turns

  • Press enter during a turn to queue a follow-up.
  • Press ctrl+x to steer the active turn immediately.
  • Press esc to interrupt while preserving the draft.
  • Use /status and /session for connection and usage facts.

Resume later

Use /resume to select a saved JSONL session, or launch with --session-id. Martty restores the durable conversation alongside cached workspace, model, and permission information. Start cleanly with /new.

Preview before connecting an API

npm install --global martty
martty --demo

The demo renders the interface without a runtime or API key. When you are ready for real work, return to the DSH profile above. See the full DeepSeek Harness TUI reference for capabilities and architecture.