btspoony/dsh-llm-fallbacks
Role-based LLM retry & fallback strategies.
dsh-llm-fallbacks implements role-based LLM retry and fallback strategies for DeepSeek Harness: when an agent's requests keep failing (retries exhausted, auth errors, quota exceeded, HTTP 429), it walks a configured provider/model fallback chain for the current role, so the step/turn continues on the target model instead of being interrupted by model problems. Configuration lives in a Settings card (web profile) and works identically in the dsh-tui terminal profile.
Install
dsh plugin --profile web add dsh-llm-fallbacksnpm dsh-llm-fallbacks 0.3.5 verified 2026-09-01 (maintainer btspoony; README EN primary, repo btspoony/dsh-llm-fallbacks). Automatic provider/model fallback chains — when LLM requests keep failing (retries exhausted, auth errors, quota exceeded, 429 rate limiting), the plugin switches provider/model along the chain for the current role and the current step/turn continues on the target model. Works in both web and dsh-tui profiles; README documents dsh plugin --profile web add dsh-llm-fallbacks and the dsh-tui variant.
Compatibility
DSH 0.1.2-alpha.3; web + dsh-tui profiles; node >= 22; pnpm >= 10.
Details
- Repo: btspoony/dsh-llm-fallbacks
- Category: Models & Providers
- Stars: 20
- Version: npm dsh-llm-fallbacks 0.3.5
- Last push: 2026-09-01
- First seen: 2026-08-07
Recent updates
Role-based fallback chains; failure triggers (auth/quota/429); turn continues on target model; web + dsh-tui settings card.
FAQ
- When does the fallback trigger?
- When retries are exhausted, auth fails, quota is exceeded, or the API rate-limits (429) — the plugin switches along the chain for the current role.
- Does my task get interrupted?
- No — the current step/turn continues on the target model after the switch.
- How do I install it?
- dsh plugin --profile web add dsh-llm-fallbacks (npm 0.3.5) — dsh-tui has its own variant.
Alternatives
btspoony/dsh-advisor · Saktawdi/dsh-ha-orchestrator · dylan121322/llm-adaptive