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#0.1.1

For example, into the web profile:

dsh plugin --profile web add github:orzgithub/dsh-ollama#0.1.1

#0.1.1 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