A CLI that treats AI search like infrastructure, not magic

What it does SearchCLI is the command-line front-end to Volcengine’s AI Search platform. It lets you onboard structured data, configure search and recommendation applications, run conversational retrieval, and tune query relevance — all through a stable, scriptable interface. The tool is explicitly designed for two operators: human developers and external AI agents, both using the same command surface.

The interesting bit The “Viking skills” architecture. SearchCLI exposes its capabilities as installable skill bundles (vs-search, vs-recommend, vs-chat, etc.) that external agents can discover and invoke. This isn’t just an API wrapper with CLI sugar; it’s an attempt to make search infrastructure legible to autonomous systems that need to plan, execute, and verify multi-step workflows without human hand-holding.

Key highlights

• Reviewable execution model: dry-runs, confirmation gates, and read-after-write verification built into the command design
• Agent-native output: --json flags everywhere, plus dedicated agent quick-start documentation
• Search tuning pipeline: automated query generation, plan/run/report cycle for text-similarity evaluation
• Credential hygiene: API keys go to local secure credential store, not plain config files
• Video-aware: explicit --type video handling with DataFieldConfig requirements, avoiding silent schema failures

Caveats

• Requires Volcengine cloud credentials (AK/SK); not a standalone search engine
• LLM features for relevance judging need a separate OpenAI-compatible endpoint configured
• Node.js 20+ and git are hard dependencies

Verdict Worth a look if you’re building agent systems that need to provision and iterate on search infrastructure programmatically. Skip it if you want a self-hosted search engine or aren’t already in Volcengine’s ecosystem — this is integration tooling, not a replacement for Elasticsearch or Meilisearch.