
dsh plugin add: Install, Verify, Remove Harness Bundles
dsh plugin add installs a bundle into a profile, but installed does not mean effective. Verify with dsh --dump-config, pack a tarball, or remove cleanly.
Modellix
LLM API — Browse Modellix Blog guides and insights by category.

dsh plugin add installs a bundle into a profile, but installed does not mean effective. Verify with dsh --dump-config, pack a tarball, or remove cleanly.

Swap DeepSeek Harness web_search and web_fetch providers: install dsh-modellix, add a Modellix key, and both native tools run through tool.modellix.ai.

An agent harness turns a language model into an agent: loop, tools, context, and the model source most explainers skip. Examples and what it does for you.

Set ANTHROPIC_BASE_URL to https://llm.modellix.ai, never /v1: Claude Code settings.json and the harness anthropic-messages provider, plus the 401 auth traps.

What 'openai-completions' means: an OpenAI compatible API is a protocol contract, not a switch. Changing only the URL breaks; the cheat sheet shows why.

One Modellix API key powers LLM, Design, and Web in DeepSeek Harness. Where it lives depends on how you add it: $DSH_HOME/.credentials.yaml or MODELLIX_API_KEY.

DeepSeek Flash: what the September 10 rename changed, both rate cards side by side, and the ratio where the new model costs more.

The latest GPT model is GPT-6 Astra (gpt-6-astra): what ~openai/gpt-latest points at now, what the switch cost, and when to pin.

MiniMax M3 API guide: model ID, 1M-token context, official per-million pricing, both base URLs, working curl and SDK code, and M3 vs M2.7.

Decode DeepSeek Harness provider errors: UNKNOWN_MODEL means the models list is the catalog, MISSING_CREDENTIAL means no key, only a 401 marks a key invalid.

Generating images inside DeepSeek Harness requires a plugin: the Design module adds image, video and audio output, and the billing boundaries most guides skip.

See what one DeepSeek Harness agent run costs: query GET /v1/logs for per-request cost, token buckets, and session attribution, then convert per 1M tokens.