Six DeepSeek Harness Plugins to Cut Token Usage and Organize Context
Six DeepSeek Harness Plugins to Cut Token Usage and Organize Context
A review of six plugins for DeepSeek Harness, covering free model access, memory organization, agent routing, and context monitoring.
Some plugins collect and store information, others expand AI-agent capabilities, and others help monitor their behavior. Each has its own purpose, so it is worth understanding what it does before installing it.
- dsh-our-free-model — 49.7k stars
Allows the use of modern models without registration, authentication, or API keys. Users only need to choose a model. Local port forwarding is also supported.
The free pool has rate limits, so confidential data should not be sent through it.
- OpenViking — 41.1k stars
Organizes memory, knowledge, and Skills as a filesystem. Instead of loading the full context, the agent receives short summaries and retrieves information only from the relevant sections.
The reported memory accuracy is 80–83%, while token savings are claimed to be between 34% and 91%.
The service requires prior configuration. Initial setup takes at least 30 minutes.
- Reasonix — 36k stars
An open-source AI coding agent that connects to DeepSeek Harness through ACP as a subagent.
It can analyze source code, edit files, run tests, and revert changes.
It supports two-level configuration. Users should note that if the plugin is disabled, the shell may simulate actions.
- routing-suite — 7k stars
Automatically selects a model based on the task. It first classifies the request and then routes it to the appropriate model.
The reported routing accuracy is 96%, with cache hit rates of 92–94%.
Installation takes three steps, and the plugin has strict requirements for the main system version.
- dsh-ops — 4,723 stars
Reduces the volume of returned results and hides some built-in tools. PDF and Word documents are loaded on demand.
This prevents system instructions from being overloaded with unnecessary information and reduces token usage.
However, the claimed savings have not yet been independently verified by third-party developers.
- dsh-context — 3,823 stars
Provides five ways to track context composition, token usage, cache state, and interactions between subagents.
The project also received a star from the official DeepSeek account.
It works only in view mode and does not save changes.
Why it matters
AnalysisThese plugins can lower AI-agent costs and improve context hygiene, but their advertised metrics should be treated cautiously until independently verified.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
DeepSeek Harness Introduces Creator Mode for Custom Plugins
DeepSeek Harness has added a Creator Mode that lets users build custom plugins, workspaces, tools, and interfaces. The developers also published the English and Chinese prompts used to create the demo plugins.
Eight Useful Plugins for DeepSeek Harness That Expand DSH Capabilities
A curated set of plugins adds plugin discovery, image generation, long-term memory, remote control, and context analysis to DeepSeek Harness.
Context Distillation & Token Squeezing
Methods for automatically removing excess words, syntactic noise, formatting spaces, and outdated messages before sending prompts to the model, optimizing budget and accelerating inference.
Codex Plugins for Beginners: The Complete Guide to the New Skill Ecosystem
A comprehensive guide to the Codex plugin ecosystem: understanding Skills vs. Plugins, connecting Marketplaces, GUI and CLI management, MCP server authorization, and packaging custom plugins.