ReWrite — Features To Add

Forward-looking backlog. Not committed to a release. Add items as they come up; move to a phase plan when picking one up.

Open

1. Plugin-managed local whisper.cpp server (desktop)

Goal: let a desktop user run fully on-device transcription with whisper.cpp without ever opening a terminal during normal use. Plugin spawns the server, plugin tears it down, pipeline talks to it via the existing openai-compatible adapter under the hood.

Scope: desktop only. Mobile profile is unaffected and continues to use remote or openai-compatible providers. Guard everything with Platform.isDesktop and lazy-require Node modules inside the guard, the same pattern src/secrets.ts uses for safeStorage.

Obsidian policy: clean. Confirmed against Obsidian's Developer policies and Plugin guidelines on 2026-05-24: no explicit prohibition on child processes or launching user-supplied binaries. Standard transparency rules apply (disclose in README, no obfuscation).

Binary + model location: user-supplied absolute paths in settings, not a fixed location under <vault>/.obsidian/plugins/rewrite-plugin/. Two reasons: (a) people often keep binaries in ~/bin/ or a tools dir and would rather not duplicate, (b) large model files (150 MB-1.5 GB) inside .obsidian/ would inflate vault sync (Obsidian Sync, Syncthing, iCloud, etc.). Plugin only reads paths, never copies or downloads.

Staged delivery:

Phase A (the "C" option) — on-demand, button-driven

User experience:

New settings section "Local whisper.cpp server (desktop)". Fields:
- Binary path (absolute, file picker + text input)
- Model path (absolute, file picker + text input)
- Port (default 8080, integer)
- Extra args (optional, advanced)
Start / Stop buttons in settings. Status indicator (stopped / starting / running / crashed).
When running, transcription provider auto-resolves base URL to http://127.0.0.1:<port> if the user selects a new "Local whisper.cpp" transcription provider, or the user can still point openai-compatible at it manually (no special-casing in pipeline).

Build:

New module src/whisper-host.ts: start(), stop(), status(), onLog(cb). Uses lazy-required child_process.spawn, net (port-in-use probe), fs (existence checks).
Health check after spawn: poll GET / or GET /v1/models on the port for up to ~5s before declaring ready.
Orphan handling on startup: if a previous Obsidian crash left a process bound to the configured port, surface that in status rather than silently failing or killing an unrelated process.
Cross-platform: handle .exe suffix on Windows, chmod +x is the user's problem (document it).
Capture stdout/stderr, ring-buffer to ~1 MB, show in a "View log" disclosure inside settings.
Stop on plugin unload regardless of how the server was started.

Phase B — auto-start lifecycle

Once Phase A is solid:

Add "Start automatically when Obsidian opens" toggle (default off).
Add "Stop when idle for N minutes" toggle (default off; useful for the large-v3 user who doesn't want 1.5 GB resident all day).
README updates: setup walkthrough with binary + model download links, troubleshooting (port in use, antivirus blocking, model load failure).

Touch points: src/main.ts (lifecycle hook for stop-on-unload only), new src/whisper-host.ts, src/settings/tab.ts (new section), src/types.ts (settings shape), src/platform.ts (capability probe for Platform.isDesktop + FileSystemAdapter), README.md (new section + disclosure of process-spawn behavior).

Risks / things to watch:

Process supervision has long-tail bugs (zombies, orphans, signal handling differences across platforms). Phase A's button-driven model is forgiving; Phase B's auto-lifecycle is where these bite.
Antivirus may quarantine whisper-server.exe on Windows on first run. Document it; don't try to work around it.
The plugin must never spawn anything the user didn't explicitly configure. No discovery, no PATH lookup, no "auto-download." Path is supplied → spawn that exact file.

Done

Added an optional listModels(config, signal) method to both TranscriptionProvider and LLMProvider interfaces. Implementations: OpenAI / Groq / Mistral (shared via the OpenAI-shaped adapter), Anthropic, Gemini, Deepgram. Skipped (text-only fallback): openai-compatible (URL-specific, list-shape varies), AssemblyAI, Rev.ai, Web Speech. Settings tab renders a hybrid Setting per model: dropdown of cached models + Refresh button (when the provider supports listing) plus an always-canonical text input that wins for "Custom" model names not yet in the dropdown. Cache lives at GlobalSettings.modelCache.{transcription,llm}[providerId] = { ids, fetchedAt }. No auto-fetch on settings open; the user clicks Refresh once their key is set. Errors surface via Notice with the provider-attributed ProviderError message.

Act on existing text in a note

Added a { kind: 'text'; text: string } source variant to the pipeline (skips transcription, same path as paste). Three entry points: rewrite-plugin:process-text command, an editor-menu item "ReWrite with template...", and a new "From note" tab in the main modal. All three resolve text from the active editor (selection if non-empty, otherwise whole note body) and open a lightweight template quick-picker before running. Setup-card gating split into voice (full check) vs text (LLM-only check) via isProfileConfiguredForText and a purpose param on renderSetupCard; the modal's per-tab rendering picks the right gate. Helpers live in src/ui/text-source.ts (resolution + runTextPipeline) and src/ui/template-picker.ts (the quick-picker modal). Shipped without the per-invocation "Replace source" checkbox; can revisit if users ask.

Collapse API key settings: per-profile only, hide rarely-changed fields under "Advanced"

Removed the "Global API keys" section and the by-family apiKeys map. Each profile now owns its own transcription and LLM keys directly on transcriptionConfig.apiKey / llmConfig.apiKey. The per-profile fields are no longer framed as "overrides"; they're the only slot. "Transcription language" and "LLM max tokens" moved into a per-profile <details> "Advanced" disclosure. The resolver functions (resolveTranscriptionApiKey, resolveLLMApiKey) and family helpers (transcriptionProviderFamily, llmProviderFamily) are gone, along with the ProviderFamily type. secrets.json.nosync now only contains profile:{kind}:{side} IDs.

6.8 KiB Raw Blame History