LCARS

Ask me anything about your vault. I'll search your notes and answer with citations.

Quick Note Title

Current key events

Search upcoming

Service Task Status

Time	Service	Task	Status	Summary	Result
Loading…

System status

Full reindex rebuilds both open and restricted vector stores — same as npm run index. Use this after changing classification rules or embedding settings. Incremental updates changed/deleted notes only. Save embedding settings below before indexing — unsaved form changes are not written to data/settings.json.

Classification & chunking

Folder rules, daily-note chunking, exclusions, and store paths are configured in one file (not editable here).

Vault

Vault path Obsidian vault name as shown in Obsidian → vault list Exclude paths one per line — merged with rules in rag-rules.json

Embeddings

Ollama URL

Embedding model Dimensions

Embed batch size

Chat

Local mode uses the Ollama chat model below and searches both stores. Claude mode uses the Anthropic API (open store only) — set the API key env var shown under Classification.

Chat model local mode only

Retrieval top K Min score Excerpt max chars

Min score drops retrieved chunks below the threshold (0–1). Try 0.55 to filter weak matches; set 0 to disable.

System prompt

Memory

Memory-first answers use a short instruct model (not the thinking chat model) over zz-LCARS Settings/Memory.md. Exact hits can skip the LLM entirely.

Enable memory-first answers Memory note path

Answer model Propose model

Memory top K Memory min score

Template exact / high-confidence hits (skip answer model) Skip vault RAG when memory answers

Changing embedding model or dimensions requires a full reindex. Classification rules live in config/rag-rules.json. HTTP port is set in config/default.json and requires a server restart.

Logs

Log file

Loading logs…

Heartbeat monitors

Each monitor gets a status card on the Service status page. Choose HTTP endpoint for Shortcuts, cron, and webhooks, or Obsidian note for sandboxed tasks that can write to the vault but cannot reach this server.

Loading…

Add monitor

Service name Reporting

Expected frequency

Description shown on the Status card

Slug: —

Obsidian note poller

Scheduled task that reads heartbeat lines from a vault note. Sandboxed tasks can report by replacing their line instead of calling the HTTP API. The note path is excluded from RAG indexing.

Scheduled jobs

LCARS-managed schedulers — in-process pollers run inside the RAG server; launchd jobs (vault indexer, daily note opener, podcast digest) use calendar schedules and are not editable here.

Job	Schedule	Last run	Last success	Status
Loading…

Meeting transcription on this host. Notes are written as ## Title, then ### Summary and ### Transcript under your Daily Note ## Notes section.

Daily note path Notes section

Whisper binary Whisper model

Whisper model path optional full path to ggml model Summarization model Ollama model for transcript cleanup and meeting summaries (separate from vault chat) ffmpeg binary Max upload (bytes)

Stale job detection

Used when you delete failed and hung Scribe jobs, and to classify jobs as safe to purge. Jobs actively processing on this server are not removed until they exceed the hung timeout.

Orphan timeout (minutes) Job still marked queued/running but no worker on this server — e.g. after a restart or crash. Default 2. Hung timeout (minutes) Upload or pipeline started on this server but no stage progress — e.g. stuck receiving or summarizing. Default 30.

Retention

Successful job retention (days) Full delete of completed Scribe sessions and their Events log entries after this many days (audio, transcript, and re-write checkpoints). Default 7. light-edits the diarized transcript via LLM before it is written to the note Summary prompt instructions for the summary bullets only

Removes failed jobs, hung jobs (using the orphan and hung timeouts above), and successful jobs past the retention period.

Connection

On iOS, set your Mac's Tailscale URL (e.g. http://mini:3847). Leave blank when using the built-in web UI on the same host. Used for all API calls (RAG, Scribe, events) and Scribe daily-note timestamps.

Server URL Tailscale hostname of your RAG box, including port. API key optional — matches RAG API_KEY env Timezone Used for daily note date and meeting headings

Uploads

Recent recordings and upload status. Upload starts from scratch, Resume continues from the last server checkpoint, Pause stops the current run, and Remove clears the local listing (non-completed sessions also delete the server job).

Loading…

Podcast digest

Tag items in any vault note with #podcast. The pipeline extracts content, writes a spoken digest via Ollama, synthesizes audio with Kokoro, and publishes a private RSS feed.

Pending items: 0

Feed:

No runs yet.

Podcast settings

Podcasts note path

Ollama chat model

Max words / episode Cap when packing items and in the script prompt — not a target to pad toward Max episodes / run Keep episodes

Produce hour (0–23) Daily launchd schedule on macOS — updates when you save Produce minute

Feed title Tailscale hostname hostname only, e.g. lcars.tail1234.ts.net — not a full URL

System prompts

Per source item: item summary, short section title, then episode script. Section transitions close each source and introduce the next. Table of contents, pauses, and paragraph merging are automatic.

Item summary — external sources YouTube videos and web articles Item summary — vault notes Inline #podcast text, [[wikilinks]], and vault files Episode script — system prompt Applied once per source item (not the whole episode); {wordLimit} is a cap per item Research script — system prompt Research write-ups only; opens with “You asked me to research…” then narrates results Section transitions — system prompt Recap + handoff between sources; TOC handoff to the first source Short section titles — system prompt 1–4 word labels for TOC, feed title, and transitions (e.g. Lisbon, deadlifts)

TTS

Python path Kokoro voice

YouTube & web

yt-dlp binary Min caption words

Web max chars Web timeout (ms)

Heartbeat slug Log retention (days)

Future: daily Notes fallback (not yet available) Append daily Notes recap when queue is light

Recent episodes

Open Podcasts.md in Obsidian

Episode	Duration	Items	Audio	Published
Loading…

Podcast log

Loading…

Full log available under RAG → Logs → podcast.log

Flagged Mail Monitor

Watches Apple Mail for flagged messages, summarizes each via Ollama, appends to your Daily Note ## Captures section, then unflags the message. Daily note dates and capture times use your Mac's system timezone. Enable the job and set its interval under Scheduled Jobs.

Captures section Heading in today's daily note (without ##). Dates and times use your Mac's system timezone.

Summarization model Ollama model for email summaries (separate from vault chat) Max email body (characters) Longer bodies are truncated before summarization Summary prompt instructions for the summary bullets only

LCARS

Service status

Tasks

Key Events

Current key events

Search upcoming

Event log

Settings

System status

Classification & chunking

Vault

Embeddings

Chat

Memory

Logs

Heartbeat monitors

Add monitor

Obsidian note poller

Scheduled jobs

Stale job detection

Retention

Connection

Uploads

Podcast digest

Working…

Podcast settings

System prompts

TTS

YouTube & web

Recent episodes

Podcast log

Flagged Mail Monitor

Connect to your Mac

Service status

Tasks

Key Events

Current key events

Search upcoming

Event log

Settings

System status

Classification & chunking

Vault

Embeddings

Chat

Memory

Logs

Heartbeat monitors

Add monitor

Obsidian note poller

Scheduled jobs

Stale job detection

Retention

Connection

Uploads

Podcast digest

Working…

Podcast settings

System prompts

TTS

YouTube & web

Recent episodes

Podcast log

Flagged Mail Monitor

Share Sheet handlers

Add handler