dsh-ollama

dsh-ollama

English | 中文

A DeepSeek Harness plugin for local AI models via Ollama native protocol.

Features

• Ollama Native API — Uses /api/chat NDJSON streaming directly, no OpenAI-compatible bridge needed
• Model Management — Browse installed models, pull new ones, view details, remove
• Configurable Parameters — Adjust Ollama-specific params like Temperature, Top K/P, Mirostat, Seed, thinking mode, and more
• GUI Configuration — All settings managed through the DSH Plugins page, no manual config files

Installation

Install the plugin into a DSH profile (the Web profile is usually named web):

dsh plugin --profile <your profile, usually web> add github:orzgithub/dsh-ollama#0.1.2

For example, into the web profile:

dsh plugin --profile web add github:orzgithub/dsh-ollama#0.1.2

#0.1.2 pins the release tag — omit it to follow the repository's default branch. Restart DSH afterwards; the Ollama card then appears on the sidebar Plugins page.

Prerequisites

• Ollama installed and running (default: http://127.0.0.1:11434)
• DeepSeek Harness v0.17+

Configuration

All configuration is done through the DSH GUI:

1. Open the sidebar Plugins page and use the Ollama bundle card
2. Connection — Set your Ollama server address and test connectivity
3. Installed Models — View, inspect, and remove models
4. Pull Model — Download new models by name (e.g. llama3.1:8b)
5. Parameters — Tune Ollama-specific generation parameters (click "Ollama Params" button in the chat input area when an Ollama model is active)

Configurable Parameters

Parameter · Description Temperature · Sampling temperature (0–2) Seed · Random seed for reproducibility Top K / Top P / Min P · Nucleus sampling controls Repeat Penalty / Repeat Last N · Repetition penalty settings Frequency / Presence Penalty · Token penalty controls Mirostat / Mirostat Tau / Mirostat Eta · Mirostat adaptive sampling TFS Z / Typical P · Alternative sampling methods Max Tokens (num_predict) · Maximum output tokens Context Length (num_ctx) · Context window size Stop Sequences · Comma-separated stop tokens Thinking · Enable/disable Ollama thinking mode

License

GPL-3.0