Skip to content

dsh-provider

Verified

@aistupidlevel/dsh-provider · v0.1.0 · MIT

AI Stupid Level Smart Router for DeepSeek Harness: adds an `aistupidlevel` provider route whose routing modes send each request to the model scoring best for the task, with your own provider keys.

Install

dsh plugin add @aistupidlevel/dsh-provider

Confirm the layer applied with dsh --profile default --dump-config — see the install guide.

Source

Tags

Creators

Readme

AI Stupid Level for DeepSeek Harness

Adds the AI Stupid Level Smart Router to DeepSeek Harness as a provider route, aistupidlevel.

Instead of a fixed model you pick a routing mode. Each request goes to the model that scores best for that mode on AI Stupid Level's continuously re-run benchmarks, among the providers you connected (OpenAI, Anthropic, Google, DeepSeek, Kimi, GLM). The router calls that model with your own provider keys and adds nothing to token prices.

Model Sends each request to
auto-coding the best model on the coding benchmark
auto-tooling the best model on the tool-use benchmark (suits agent work)
auto-reasoning the best model on the multi-turn reasoning benchmark
auto-best the best combined score
auto-task one of the above, chosen from the request itself
auto-value the best score per dollar among near-top models
auto-fastest-quality the fastest of the strong models

Install

dsh plugin --profile web add @aistupidlevel/dsh-provider

Replace web with your profile. The bundle supports DeepSeek Harness 0.2.0-rc.2 and 0.2.1-rc.2.

Set up

  1. Create a router key at https://aistupidlevel.info/router/keys (it starts with aism_), and add your provider keys in the router settings on the same site.

  2. Make the key available to DeepSeek Harness as ASL_ROUTER_KEY:

    export ASL_ROUTER_KEY=aism_...
    
  3. Choose AI Stupid Level and a routing mode on the Models page, or make it the default in your profile's cordis.patch.yml:

    - id: agent-default-model
      config:
        provider: aistupidlevel
        model: auto-coding
    

How it works

The bundle adds the aistupidlevel route to DeepSeek Harness's own pi-ai adapter (@deepseek-ai/dsh-llm-pi-ai) as an OpenAI-compatible gateway at https://aistupidlevel.info/v1. There is no code of its own: streaming, tool calls and retries are the adapter's. It does not change your default model.

If you configure other pi-ai providers yourself: a later layer that sets llm-pi-ai providers replaces the list this bundle adds, so copy the aistupidlevel block from cordis.patch.yml into your own list.

To pin one model instead of a routing mode, add its id (for example claude-opus-5-5) to the route's models list in your own layer. The router's live model list is at GET https://aistupidlevel.info/v1/models.

What this bundle does (disclosure)

  • Network calls: when you select the aistupidlevel route, DeepSeek Harness sends your prompts, tool definitions, tool results and your router key to https://aistupidlevel.info/v1. The router forwards each request to the AI provider it routes to, using your own key for that provider. Model usage is billed to your provider accounts.
  • Nothing else: the bundle contains only a configuration layer: no code, tools, hooks or telemetry.
  • Prompt logging: prompts are not stored by AI Stupid Level unless you enable prompt logging for a key in your account. See https://aistupidlevel.info/privacy.

Text and tool calls only: the router does not accept image input.

Licence

MIT