DSH Provider Config
Field-tested LLM provider configuration templates and rate-limit retry best practices for DeepSeek Harness (DSH). Currently focused on SenseNova (商汤 Token Plan), structured to extend to any OpenAI-compatible provider.
Why this project
DSH configures LLM providers via ~/.dsh/settings.yaml. The most critical — yet most overlooked — piece is retryPolicy. It decides:
- Whether a task auto-recovers when hitting rate limits (HTTP 429 /
RATE_LIMIT) - How long retries continue before giving up, and whether it can hang forever
- Whether persistent provider failures surface an error you can actually diagnose
This project distills these settings into ready-to-use templates and explains what every parameter means and the tradeoffs.
Contents
| Path | Description |
|---|---|
config/sensenova.yaml |
Ready-to-use SenseNova provider template (copy into ~/.dsh/settings.yaml) |
docs/retry-policy.md |
Retry mechanics in depth: normal vs always, exponential backoff, jitter, retry-count tradeoffs |
docs/troubleshooting.md |
Rate-limit troubleshooting: telling short bursts from persistent problems |
Quick start
- Open the template:
config/sensenova.yaml - Read the retry mechanics:
docs/retry-policy.md - Copy what you need into your
~/.dsh/settings.yaml - Restart DSH, and enjoy automatic retry on rate limits
Configuration template
See config/sensenova.yaml. The template contains no real API keys — keys are referenced exclusively through the environment variable SENSENOVA_API_KEY.
Retry policy at a glance
- Recommended:
mode: normal+maxRetries: 15+retryableCodes: [RATE_LIMIT, TIMEOUT, SERVER, TRANSPORT] - Exponential backoff
initialDelayMs → maxDelayMswith jitter to avoid thundering-herd - Full tradeoff analysis in
docs/retry-policy.md
License
MIT — see LICENSE.
Security
This project never contains real API keys. Keys are referenced only by environment-variable name (e.g. SENSENOVA_API_KEY); templates contain nothing but the env-var name. See SECURITY.md.