docs: audit and fix config reference, keybindings, README, and website (#743)

## Summary

- Fix `extraction.ocr.confidence_threshold` default (`0` -> `70`) and
description (controls annotation inclusion in spatial layout output, not
word dropping)
- Document `extraction.ocr.tsv` (spatial layout annotations) and
`extraction.llm_timeout` fields missing from config reference
- Add all 12 per-pipeline env vars (`MICASA_LLM_CHAT_*`,
`MICASA_LLM_EXTRACTION_*`) to env var table
- Add `[extraction.ocr]` section to example TOML config
- Document <kbd>ctrl+b</kbd> (background extraction) in keybindings
reference and documents guide
- Note that extraction results can be accepted without an LLM configured
- Fix README backup advice from `cp` to `micasa backup` (2 instances)
- Add Homebrew install method to website landing page (was missing
entirely)
- Add LLM chat feature to website feature grid (was absent)
- Add "when an LLM is configured" qualifier to website extraction
description

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: cpcloud

README.md

@@ -30,7 +30,7 @@ Your house is quietly plotting to break while you sleep -- and you're dreaming a
 - **Is the dishwasher still under warranty?** Appliance tracking with purchase dates, warranty status, and maintenance history tied to each one.
 - **The basement is leaking again.** Log incidents with severity and location, link them to appliances and vendors, and resolve them when fixed.
 - **Who did we use last time?** A vendor directory with contact info, quote history, and every job they've done for you.
-- **Where's the warranty card?** Attach files (manuals, invoices, photos) directly to projects and appliances. Stored as BLOBs in the same SQLite file -- one `cp` backs up everything. Full-text search (`ctrl+f`) finds documents by title, notes, or extracted text. When an LLM is configured, PDFs and images are automatically processed through an extraction pipeline (text extraction, OCR, LLM) that pre-fills form fields with structured data.
+- **Where's the warranty card?** Attach files (manuals, invoices, photos) directly to projects and appliances. Stored as BLOBs in the same SQLite file -- `micasa backup` backs up everything. Full-text search (`ctrl+f`) finds documents by title, notes, or extracted text. When an LLM is configured, PDFs and images are automatically processed through an extraction pipeline (text extraction, OCR, LLM) that pre-fills form fields with structured data.
 - **How much have I spent on plumbing?** Press `@` to chat with a local LLM about your data. It writes the SQL, runs the query, and summarizes the results -- all on your machine.
 
 ## Keyboard driven
@@ -69,7 +69,7 @@ micasa                # start fresh with your own house
 micasa --print-path   # show where the database lives
 ```
 
-> One SQLite file. Your data, your machine. Back it up with `cp`.
+> One SQLite file. Your data, your machine. Back it up with `micasa backup`.
 
 Need Nix or container install options? See the full [installation guide](https://micasa.dev/docs/getting-started/installation/).
 

docs/content/docs/guide/documents.md

@@ -124,6 +124,10 @@ with a JSON Schema constraint that produces structured database operations
 the document itself. The operations are validated against a strict allowlist
 before display.
 
+When no LLM is configured (or when OCR/text extraction alone is sufficient),
+you can still accept the text and OCR results without running the LLM step.
+The extracted text is saved to the document for full-text search regardless.
+
 **This feature is early-stage.** Results vary significantly by model, document
 type, and text quality. Invoices and quotes with clear line items tend to work
 best; complex multi-page documents or poor OCR output often produce incomplete
@@ -160,6 +164,7 @@ apply them. On error the overlay stays open showing which step failed. Press
 | Key | Action |
 |-----|--------|
 | <kbd>a</kbd> | Accept results (when done, no errors) |
+| <kbd>ctrl+b</kbd> | Background the extraction (continue working while it runs) |
 | <kbd>esc</kbd> | Cancel / exit explore mode |
 | <kbd>j</kbd>/<kbd>k</kbd> | Navigate steps (pipeline) or rows (explore) |
 | <kbd>h</kbd>/<kbd>l</kbd> | Navigate columns (explore) |

docs/content/docs/reference/keybindings.md

@@ -205,6 +205,7 @@ overlay has two modes. See the
 | <kbd>x</kbd>       | Enter explore mode (when operations are available) |
 | <kbd>a</kbd>       | Accept results (when extraction is done with no errors) |
 | <kbd>r</kbd>       | Rerun LLM step (when LLM step is complete) |
+| <kbd>ctrl+b</kbd>  | Background the extraction (continue working while it runs) |
 | <kbd>esc</kbd>     | Cancel extraction and close overlay |
 
 ### Explore mode

docs/layouts/index.html

@@ -91,7 +91,12 @@ <h2>Frequently <span class="typewriter accent" id="typewriter">asked</span> ques
 
       <div class="feature">
         <p class="feature-q">Where&rsquo;s the warranty card?</p>
-        <p class="feature-a">Attach files&mdash;manuals, invoices, photos&mdash;directly to projects and appliances. Stored in the same SQLite file. Full-text search finds any document by title, notes, or extracted text. PDFs and scanned documents are automatically OCR&rsquo;d and analyzed.</p>
+        <p class="feature-a">Attach files&mdash;manuals, invoices, photos&mdash;directly to projects and appliances. Stored in the same SQLite file. Full-text search finds any document by title, notes, or extracted text. PDFs and scanned documents are automatically OCR&rsquo;d so their text is searchable, and when an LLM is configured, that text can also be analyzed for richer summaries and structure.</p>
+      </div>
+
+      <div class="feature">
+        <p class="feature-q">How much have I spent on plumbing?</p>
+        <p class="feature-a">Chat with a local LLM about your data. It writes the SQL, runs the query, and summarizes the results&mdash;all on your machine.</p>
       </div>
 
     </div>
@@ -104,7 +109,11 @@ <h2>Frequently <span class="typewriter accent" id="typewriter">asked</span> ques
     <h2>Get <span class="accent">started</span></h2>
     <div class="install-steps">
       <div>
-        <p>Install with Go (1.25+):</p>
+        <p><a href="https://brew.sh">Homebrew</a> (macOS and Linux):</p>
+        <pre><code>brew install micasa</code></pre>
+      </div>
+      <p class="install-or">or with Go (1.25+):</p>
+      <div>
         <pre><code>go install github.com/cpcloud/micasa/cmd/micasa@latest</code></pre>
       </div>
       <p class="install-or">or grab a binary from the <a href="https://github.com/cpcloud/micasa/releases/latest">latest release</a></p>