Connect DeepSeek Harness to Token Factory

Corvex Token Factory connects to DeepSeek Harness (dsh) through an OpenAI-compatible provider. Add your API key and provider configuration, then start the Web UI with a Token Factory model.

Prerequisites

A Token Factory API key. See Authentication.
DeepSeek Harness installed following its official instructions.
A model from the Token Factory catalog. This guide uses zai-org/GLM-5.3.

Set your API key

Run these commands in your terminal, replacing sk-corvex-... with your API key. DSH_HOME gives Token Factory its own DSH settings folder.

export CORVEX_API_KEY="sk-corvex-..."
export DSH_HOME="$HOME/.dsh-corvex"
mkdir -p "$DSH_HOME"
chmod 700 "$DSH_HOME"

Add the provider configuration

Create $DSH_HOME/corvex.patch.yml with the following contents. apiKeyEnv tells DSH to read the key you set above.

- id: llm-pi-ai
  config:
    providers:
      corvex:
        apiKeyEnv: CORVEX_API_KEY
        api: openai-completions
        baseURL: https://api.tokenfactory.corvex.cloud/v1
        compat:
          supportsDeveloperRole: false
          maxTokensField: max_tokens
        models:
          - id: zai-org/GLM-5.3
            contextWindow: 393216
            maxTokens: 393216
            input: [text]
- id: agent-default-model
  config:
    provider: corvex
    model: zai-org/GLM-5.3

If you already configured other providers, keep them in the providers section when adding this one.

Start the Web UI

Open your project directory and start DSH:

cd /path/to/your/project
dsh --profile web --patch "$DSH_HOME/corvex.patch.yml"

DSH opens the Web UI at http://127.0.0.1:3080. Start a new session with zai-org/GLM-5.3 selected. Use Add workspace to choose another project.If you normally start DSH with a different command, use it in place of dsh. In a new terminal, repeat the commands in step 1 before starting DSH.

To run a single task from the command line instead:

dsh --profile headless --patch "$DSH_HOME/corvex.patch.yml" \
  "Explain the structure of this project."

Token limits

The conversation, reasoning and answer share the model’s token limit. The configuration sets contextWindow and maxTokens to 393216; DSH leaves room for the conversation when deciding how many tokens the model can generate. A smaller maxTokens value can stop the model before it finishes reasoning.

If a request fails

DSH automatically summarizes older messages to make room and retries when a request exceeds the context limit. You do not need to edit the conversation. A 429 error means requests are temporarily limited. This can follow heavy usage or repeated failures. If the response includes Retry-After, wait that long before trying again. Waiting alone will not fix a request that keeps failing. If DSH cannot continue, contact Token Factory support with the error message and request ID, if available. See Errors for details.

Get Started

Models & Pricing

Integrations

API Reference

Security & Compliance

Reference

Connect DeepSeek Harness to Token Factory

Prerequisites

Token limits

If a request fails

​Prerequisites

​Token limits

​If a request fails

Prerequisites

Token limits

If a request fails