Best DeepSeek Harness Model and Provider Plugins (2026)
Best DeepSeek Harness (dsh) model and provider plugins: Codex/ChatGPT connectors, fallback routing, adaptive model selection, and cost-tracking wallets.
DeepSeek Harness (dsh) already handles model configuration through Settings → Models — DeepSeek native, a directory of built-in providers (Anthropic, OpenAI, Bedrock, Vertex, Azure, Codex), and any OpenAI-compatible custom endpoint. The plugins in this category don't replace that mechanism; they sit on top of it — importing extra logins, routing between models automatically, or tracking what it's all costing. By stars, the top picks are dsh-codex-connect and deepseek-harness-wallet. Full list and install commands below.
What "provider" means here, and what it doesn't
It's worth being precise before the list, because this category name invites a wrong assumption. dsh's actual provider configuration — Provider IDs, Base URLs, API protocol, model lists, the Fetch available models button — lives in $DSH_HOME/settings.yaml and is driven entirely through the Web UI's Models page, as covered in Setting Up Your DeepSeek API Key and Models and Using OpenAI, Anthropic, or Any OpenAI-Compatible API. None of the plugins below reimplement that.
What they add instead: OAuth import shortcuts (so you don't hand-type an API key for a provider that already has a native login flow), fallback and routing policies (so a failed or overloaded model doesn't kill your session), and cost visibility (because none of that is exposed by default once you have multiple providers configured).
It also helps to know how dsh resolves credentials before you decide whether a plugin's OAuth-import trick is worth trusting: the documented order is your inherited process environment, then $DSH_HOME/.credentials.yaml, then the invoking directory's .env, then $DSH_HOME/.env. Anything a plugin writes into that chain sits alongside credentials you configured by hand — it doesn't get a separate, more privileged storage path.
The list
This ranking uses only the curated awesome-dsh-plugin collection (source=awesome) — not the auto-discovered plugins also indexed on the full Models & Providers category page, which holds 38 plugins as of August 2026. Stars are a snapshot as of August 2026.
| # | Plugin | Stars | What it does |
|---|---|---|---|
| 1 | dsh-codex-connect | 7 | ChatGPT OAuth + Codex models, opt-in search/image tools |
| 2 | deepseek-harness-wallet | 6 | Multi-provider balance, per-session cost, low-balance alerts |
| 3 | Qwen-MM-Plugins | 4 | Qwen multi-modal plugin support |
| 4 | dsh-llm-fallbacks | 3 | Role-based LLM retry & fallback strategies |
| 5 | dsh-github-login | 3 | Visual GitHub device-flow login, no terminal |
| 6 | dsh-model-router | 2 | Deterministic keyword routing + allowlisted switching |
1. dsh-codex-connect — 7 stars
Category: Models & Providers · Repo: franksong2702/dsh-codex-connect
Connects ChatGPT OAuth and OpenAI Codex models to dsh, with opt-in search and image tools. dsh's own provider directory lists Codex as needing native OAuth rather than a plain API key (see OpenAI-Compatible API guide for that distinction) — this plugin is one community path to wiring that login in.
dsh plugin --profile web add dsh-codex-connect
2. deepseek-harness-wallet — 6 stars
Category: Models & Providers · Repo: feibi-mochi/deepseek-harness-wallet
A multi-provider wallet chip in the Web UI: official DeepSeek balance, per-session cost and token counts, third-party token totals, a recharge shortcut, and low-balance alerts. If you've added more than one custom provider and lost track of what each session is actually costing, this puts that number where you can see it without checking each provider's dashboard separately.
dsh plugin --profile web add github:feibi-mochi/deepseek-harness-wallet
3. dsh-llm-fallbacks — 3 stars
Category: Models & Providers · Repo: btspoony/dsh-llm-fallbacks
Role-based LLM retry and fallback strategies — when a configured model errors out or rate-limits, this defines what it falls back to next instead of failing the turn outright. A narrower, more configurable alternative to dsh-model-router below if you specifically want retry policy rather than request-time routing.
dsh plugin --profile web add github:btspoony/dsh-llm-fallbacks
4. dsh-model-router — 2 stars
Category: Models & Providers · Repo: superboy911/dsh-model-router
Deterministic keyword routing, allowlisted model switching, and an isolated image_gen channel for dsh — the model routing decision is rule-based (keywords in the request) rather than adaptive, which makes behavior predictable if you're running this in a team setting where "which model answered this" needs to be traceable.
dsh plugin --profile web add github:superboy911/dsh-model-router
5. llm-adaptive — 2 stars
Category: Models & Providers · Repo: dylan121322/llm-adaptive
Adaptive model routing with per-request complexity classification and automatic provider routing — the complexity-aware counterpart to dsh-model-router's rule-based approach above. Worth comparing both if you're trying to route cheap requests to a smaller model and only escalate to a larger one when the task actually needs it.
dsh plugin --profile web add github:dylan121322/llm-adaptive
6. dsh-codex-auth — 2 stars
Category: Models & Providers · Repo: suntianc/dsh-codex-auth
A narrower alternative to dsh-codex-connect above: it reuses the login you've already completed in the Codex CLI as an openai-codex LLM route in dsh, and adds GPT Auth controls to the DSH Web settings. If you already use Codex CLI day to day and don't want to authenticate twice, this is the smaller-surface option. Published to npm.
dsh plugin --profile web add dsh-codex-auth
Also in this category: Qwen-MM-Plugins (4 stars, Qwen multi-modal support), dsh-github-login (3 stars, visual device-flow GitHub login synced into the gh CLI), and dsh-everything-oauth (1 star, imports local Codex, Grok, Claude, OpenCode, and CC Switch logins into dsh's Settings in one shot — the broadest of the OAuth-import plugins on this list, at the cost of trusting one plugin with all of them). Browse the full set on the Models & Providers category page.
FAQ
Do I need one of these plugins to add a custom model provider?
No. Adding any provider — DeepSeek native, a built-in directory provider, or a fully custom OpenAI-compatible endpoint — works through Settings → Models with no plugin required. See DeepSeek Harness API Key Setup for that flow.
What do the OAuth-import plugins actually save me?
Providers that need native authentication rather than a plain API key — Codex (OAuth), Bedrock (AWS credentials), Vertex (ADC project), Azure (api-version) — aren't satisfied just by pasting a key into the API Key field. Plugins like dsh-codex-connect, dsh-codex-auth, and dsh-everything-oauth exist to wire that native login flow in, sometimes reusing a login you already completed in a different CLI tool.
Is model fallback/routing something dsh does natively?
Not as a configurable policy — dsh routes to whichever model a session or request is explicitly pointed at. dsh-llm-fallbacks, dsh-model-router, and llm-adaptive are all community plugins filling that gap with different strategies (retry-on-failure, keyword rules, complexity classification).
Can DeepSeek's native chat model accept images?
No — per the official docs, DeepSeek's own chat-completions route is text-only and can't be made to accept images through any configuration. If you need vision, you'd add a multi-modal provider like Qwen-MM-Plugins or a custom provider with input: [text, image] declared, as covered in Using OpenAI, Anthropic, or Any OpenAI-Compatible API.
Next steps
For the full mechanics of adding a provider and the settings.yaml fields the Web UI doesn't expose, read Use OpenAI, Anthropic, or Any OpenAI-Compatible API with DeepSeek Harness. If cost is your main concern, see Is DeepSeek Harness Free?. For the rest of the catalog, browse /plugins.