Configuration

Plugin-row config fields (all have defaults; the vision* ones can also be edited live in Settings → Web Enhanced → Vision, whose saves override these base values):

key	default	meaning
`cronIntervalMs`	30000	Minimum delay before the cron scheduler fires; it arms one timer at the earliest pending run and no timer exists while nothing is scheduled
`balanceApiKeyEnv`	`DEEPSEEK_API_KEY`	Env var for the balance query API key
`balanceCacheTtlMs`	60000	Balance view cache duration
`balanceBaseUrl`	`https://api.deepseek.com`	Balance endpoint base URL
`balanceProviders`	`[deepseek-official]`	Model routes the balance line is shown for; a route with its own configured `baseURL` must also share the endpoint's host
`modelsDevUrl`	`https://models.dev/api.json`	Pricing blob fetched for the session-cost readout
`modelsDevCacheTtlMs`	21600000	How long one fetched pricing index stays fresh (6 h)
`modelsDevTimeoutMs`	10000	Pricing fetch timeout
`pricingProviderMap`	`{deepseek-official: deepseek}`	Model-route provider id → models.dev provider id
`skipDirs`	`[node_modules]`	Directories skipped by the file tree/search and by the mention pickers (`.git` is always skipped; the browse overlay does not apply the filter)
`readMaxBytes`	1 MiB	Text read cap (truncated with a marker beyond it)
`writeMaxBytes`	2 MiB	File write cap
`binaryMaxBytes`	5 MiB	Binary preview (base64) cap
`gitOutputMaxBytes`	256 KiB	Single git stream output cap
`gitMaxCount`	100	`git log` row cap
`gitWorkingMaxFiles`	300	Cap on the uncommitted file list, and with it how many untracked files are read to count their lines
`searchMaxDepth` / `searchMaxEntries`	8 / 200	File search depth and entry caps
`officeMaxBytes`	5 MiB	Office preview (docx/xlsx) file size cap
`browseMaxEntries`	500	Entry cap of one directory level in the mention browser
`pluginOpTimeoutMs`	300000	Deadline for one pnpm operation (update/remove)
`profileDir`	empty	Profile directory; empty walks up from this module. Only for a deployment whose profile is not an ancestor of the loaded plugin
`visionEnabled`	true	Master switch of the image-understanding integration
`visionPatchAdmission`	true	Wrap `llm.resolveModelInfo` so text-only models admit images past the send preflight and the `read_image` gate (reversible, unload-order safe)
`visionProvider` / `visionModel`	empty	Pin the DSH-configured provider/model used for transcription; empty = auto-detect the first image-capable models from all configured providers
`visionHarnessModels`	`[]`	User-selected DSH model pool `[{provider, model}, …]`, tried in order before the dedicated API; non-empty replaces auto-detection (the pinned pair still goes first)
`visionPrompt` / `visionMarker`	Chinese thorough-description prompt / `[图片内容描述]`	Transcription prompt and the marker the model sees instead of the image block
`visionBaseUrl` / `visionApiKey` / `visionEndpointModel`	empty	OpenAI-compatible VLM endpoint (e.g. DashScope compatible mode); key falls back to `visionApiKeyEnv` → `VISION_API_KEY` → `DASHSCOPE_API_KEY`. Empty base URL or model disables this source
`visionEndpointModels`	`[]`	Dedicated-endpoint model pool; the Settings tab fills it by fetching `/models` and multi-selecting. Transcription tries `visionEndpointModel` first (when set), then every pool model in order
`visionApiKeyEnv` / `visionAnonymous`	`VISION_API_KEY` / false	Env var for the endpoint key; `true` skips the Authorization header (free/local endpoints get a hard 20 s timeout cap)
`visionTimeoutMs` / `visionMaxTokens`	120000 / 4096	VLM request timeout and output cap
`visionAutoLocalOllama`	true	Probe `visionLocalOllamaUrl` at startup; when an Ollama is running, its first vision-capable model is prepended to the transcription chain (images stay on this machine)
`visionLocalOllamaModel` / `visionLocalOllamaUrl`	empty / `http://localhost:11434/v1`	Preferred Ollama model (empty picks the first `vl`/`vision` model) and its OpenAI-compatible base URL
`visionFallbackModels`	`[]`	Ordered fallback chain `{model, baseURL?, apiKey?, anonymous?, timeoutMs?}`; each entry may point at a different provider, keyless non-anonymous entries are skipped
`visionCacheLimit` / `visionCooldownMs`	200 / 60000	In-process transcription cache entries (SHA-256 of image bytes) and how long an endpoint that just failed (429/timeout) is skipped

dsh-web-enhanced

AI 分析

インストール

ドキュメント

Configuration