dsh-tool-web-enhanced

Configuration reference

Existing stock keys keep identical names and defaults; SearXNG keys are additive.

The stock fetch* keys are kept ONLY for drop-in config compatibility — web_fetch is NOT modified by this plugin (it is re-registered verbatim from the stock package). Setting them has no enhanced behaviour.

Key	Type	Default	Description
search	boolean	true	Register web_search.
fetch	boolean	true	Register web_fetch (unchanged).
searchMaxResults	number	8	Source cap for both native and SearXNG sections.
searchTimeoutMs	number	30000	Cooperative budget for web_search (incl. the SearXNG call).
fetchTimeoutMs	number	30000	Cooperative budget for web_fetch.
fetchMaxOutputChars	number	200000	Output cap for web_fetch.
searxngUrl	string	http://127.0.0.1:8080	SearXNG base URL; empty/falsy disables SearXNG.
searxngEnabled	boolean	true	When false, skip the SearXNG section entirely.

Note on searxngEnabled: it defaults to true. With searxngUrl pointing at http://127.0.0.1:8080, a running SearXNG instance is expected. If SearXNG is not deployed, every web_search performs one extra (silently-failed) connection attempt to the local JSON API before falling back to native-only results — harmless, but avoidable. Set searxngEnabled: false when you are not using SearXNG to skip that attempt entirely.

RAG configuration

Key	Type	Default	Description
rag.enabled	boolean	true	Enable the RAG section + rag_index tool.
rag.storePath	string	'' (auto)	SQLite store path; empty → /storages/rag/rag.db.
rag.embeddings.provider	string	auto	auto / deepinfra / local. auto → deepinfra when a key is present, else local.
rag.embeddings.apiKeyEnv	string	DEEPINFRA_TOKEN	Env var name holding the DeepInfra key.
rag.embeddings.apiKey	string	''	Literal DeepInfra key (wins over apiKeyEnv).
rag.embeddings.deepinfraModel	string	BAAI/bge-m3	DeepInfra embedding model.
rag.embeddings.deepinfraBaseURL	string	https://api.deepinfra.com/v1/openai	DeepInfra OpenAI-compatible base URL.
rag.embeddings.localModel	string	Xenova/bge-small-en-v1.5	Local transformers.js model.
rag.databases[].name	string	—	Database (section) name.
rag.databases[].path	string	—	Directory of Markdown files to index.
rag.databases[].topK	number	5	Results returned per database.

Local embeddings (the default when no key is configured) download a small ONNX model (~34MB) on first use, via the optional @huggingface/transformers dependency. The DeepInfra path uses DEEPINFRA_TOKEN (or rag.embeddings.apiKey). RAG errors degrade silently — the section is simply omitted from the output.