dsh-our-free-model: Frontier Model Access in DeepSeek Harness With No Login

What Does It Do?

Our Free Model is a DSH plugin for DeepSeek Harness that gives you frontier models such as Muse Spark 1.3 and MiMo V2.6 with no login, no sign-up and no API key. It removes the usual account, credential and quota friction by keeping the model list, availability probing and the calling entry point inside the plugin itself. Maintained by zouyuxuan122 under the MIT license, it mounts purely as a plugin with no kernel changes, no build step and zero dependencies, and its model list is refreshed from upstream rather than frozen as a snapshot.

Core Features

• Zero-configuration access: no account, key or backend quota setup is needed; models appear in the input box selector as soon as the plugin is enabled.
• Real availability filtering: models that upstream names but the gateway explicitly refuses to route are dropped from the dropdown and only logged in settings with a stated reason; 5xx, 429 and timeout errors are not treated as verdicts on a model, and region-gated models are isolated in a region-limited group.
• Thinking strength as a real budget: Light / Balanced / Deep map to actual output token budgets, doubled across all three tiers for models whose thinking cannot be turned off, with each model card printing the cap that tier will send.
• Local-only usage dashboard: token heatmap, total curves, output speed and time-to-first-token sampled per call, with nothing uploaded anywhere.
• OpenAI-compatible forwarding port: other local tools can call these models with a base URL and key, behind the same trust checks the kernel uses.
• In-app upgrade and hot reload: download, checksum, backup, atomic replace and read-back in one click with automatic rollback on failure, taking effect without restarting the app.

How to Use This Plugin?

After enabling it in DSH, the model selector in the input box gains an Our Free Model group; pick a model there and start chatting, while region-gated models stay isolated under a region-limited suffix. The settings page under Settings → Our Free Model has six sections: the model list shows availability, context window, the output cap each thinking tier actually sends and measured first-token latency, plus a single-call benchmark button; the announcement center receives pushes from the repository owner and shows urgent ones as a full-screen popup; the usage dashboard and local forwarding sections handle statistics and external calls; plugin settings toggle the master switch, region-limited visibility, probe interval and default output cap. A five-page onboarding appears on first launch and stays dismissed afterwards. In compositions without a browser UI the plugin still starts and serves models, since it treats llm as its only hard dependency, and the dashboard mounts once a web server appears.