diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 00000000..868c91af
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,154 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Overview
+
+This is the official Speechmatics documentation site, built with Docusaurus 3 and deployed to [docs.speechmatics.com](https://docs.speechmatics.com) via Vercel on push to `main`.
+
+## Commands
+
+```bash
+npm start              # Start local dev server (also generates Jobs API reference)
+npm run build          # Production build
+npm run lint           # Run Biome linter
+npm run lint:fix       # Auto-fix lint issues
+npm run spellcheck     # Run cspell spell checker
+npm run typecheck      # TypeScript type checking
+
+# API reference generation
+npm run build:jobs-spec       # Generate Batch/Jobs API spec from YAML
+npm run build:jobs-api-ref    # Generate full API reference docs
+
+# Redirect management
+npm run sync-redirects        # Sync redirects.json → vercel.json
+npm run clear                 # Clear Docusaurus build cache
+```
+
+Node version: 22.11.0 (see `.nvmrc`). The `MIXPANEL_PROJECT_TOKEN` env var is needed for production builds but optional locally.
+
+## Architecture
+
+**Framework**: Docusaurus 3.9.2 with TypeScript. Content is Markdown/MDX.
+
+**Key configuration files**:
+- `docusaurus.config.ts` — site metadata, plugins, navbar/footer, Algolia search
+- `sidebars.ts` — imports 8 product-specific sidebar files and composes them
+- `sidebar-generator.ts` — custom sidebar generator that filters API endpoints and normalizes labels
+- `biome.json` — linter/formatter config (replaces ESLint/Prettier)
+- `cspell.json` — spell checker (en-US dictionary + `custom-words.txt`)
+
+**Content directories** under `/docs/`:
+- `api-ref/` — auto-generated from OpenAPI specs (do not edit manually)
+- `speech-to-text/` — Batch and Realtime transcription docs
+- `text-to-speech/` — TTS product docs
+- `voice-agents/` — Voice agents docs
+- `voice-agents-flow/` — Flow platform docs
+- `deployments/` — On-prem, container, and cloud deployment guides
+- `integrations-and-sdks/` — Integration and SDK guides
+- `get-started/` and `guides/` — General guides
+
+**API specs** in `/spec/`:
+- `batch.yaml` — Batch/Jobs API (OpenAPI)
+- `realtime.yaml` — Realtime API (OpenAPI)
+- `flow-api.yaml` — Flow API (AsyncAPI)
+
+**URL redirects**: Defined in `/scripts/redirects/redirects.json`, synced to `vercel.json` via `npm run sync-redirects`. A post-build plugin (`plugins/check-redirects.ts`) validates all redirect targets exist.
+
+**Custom plugins** in `/plugins/`:
+- `source-loader.ts` — Webpack loader that lets MDX files import raw `.py`, `.sh`, `.yaml`, `.js` content
+
+## Writing style (for .md/.mdx files)
+
+Apply when editing documentation.
+
+### Core writing principles
+
+- **American English**: use en-US spelling and grammar consistently. ✅ *diarization, color* ❌ *diarisation, colour*
+- **Sentence case** for all titles and headings. ✅ *Language identification* ❌ *Supported Formats and Limits*
+- **Prioritize lowercase**: capitalize only proper nouns and product names. ✅ *Refer to the channel diarization guide for the Batch API.* ❌ *For troubleshooting Containers, please refer to the On-Prem Documentation.*
+- **Keep it concise**: avoid redundant language. ✅ *Languages* ❌ *Speechmatics supported languages*
+- **Paragraphs**: max 3 sentences each.
+- **Active voice, second person, present tense** in descriptions.
+- **Human-centric**: write for readers with varying levels of technical depth.
+- **Jargon only when needed**: provide inline definitions or cross-links.
+- **Empathy and clarity**: explain 'why' before 'how'.
+- **Enterprise-ready**: use numbered sequences, modular sections, and cross-references to deeper specs.
+
+### Glossary
+
+**Batch** — the Speechmatics API for transcribing pre-recorded audio files; asynchronous, returns full transcript on completion.
+- **Batch** (capitalized) = the Speechmatics API product. ✅ *Submit your file to Batch to receive a complete transcript once processing is done.*
+- **batch** (lowercase) = generic grouped processing. ✅ *Queue multiple files for batch conversion over a more extended period.*
+
+**Batch job** — a request to the Batch API that processes a pre-recorded audio file and returns the full transcript once complete.
+
+**Diarization** — automatic segmentation of audio by speaker identity. Always use *diarization* (not synonyms) for consistency.
+
+**On-prem** — short for *on-premises* (plural, not *on-premise*); running Speechmatics technology within clients' own infrastructure.
+
+**Realtime** — the Speechmatics API for live transcription; processes streaming audio and returns results as speech is received.
+- **Realtime** (capitalized, no hyphen) = the Speechmatics API product. ✅ *You can stream audio to Realtime to receive instant transcription results.*
+- **real-time** (hyphenated) = general adjective for low-latency systems. ✅ *The system returns real-time results.*
+
+**Realtime session** — a continuous transcription process where audio is streamed and transcribed as received, with low latency.
+
+**Speech to text** — Speechmatics' core product group converting spoken audio to written text via ASR models.
+- **speech to text** (no hyphens) = open compound noun phrase. ✅ *Our platform supports speech to text and text to speech.*
+- **speech-to-text** (hyphenated) = compound modifier before a noun. ✅ *We offer speech-to-text conversion for streaming media.*
+
+**Voice agents – Flow** — platform for conversational AI agents. Use an en dash (–) between the terms, not a hyphen or em dash.
+
+### Content framework
+
+**Navigation item names**: prefer 1–2 words; keep under 3 words; sentence case only. ✅ *Language identification* ❌ *Virtual Appliance Usage Reporting*
+
+**Page titles**: may differ from nav item; keep under 5 words if possible; sentence case only.
+
+**Descriptions**: one-sentence benefit-led statement; second person, active voice, present tense.
+- ✅ *Learn how to transcribe pre-recorded audio files.*
+- ✅ *Enable the model to fetch data and take actions.*
+- ✅ *Take your first steps with the Speechmatics API.*
+- ❌ *Speechmatics can turn audio into text.*
+- ❌ *This page will give you comprehensive details on the API.*
+
+**Headings**: no hard word limit, but keep concise — use context to reduce wording. ✅ *Examples* ❌ *Realtime output examples*
+
+**Visual hierarchy**:
+1. **Page title (H1)** — main topic, e.g., *Quickstart*
+2. **Section heading (H2)** — key steps or concepts, e.g., *Realtime processing*
+3. **Subheading (H3+)** — finer details within a section, e.g., *Operating points*
+4. **Paragraph** — up to 3 sentences
+
+### Doc types
+
+**Overview** — create for every product or top-level item needing a definition. Principles:
+1. Define core concepts: list terms with one-sentence definitions and links to the API reference.
+2. Include key use cases (as bullets).
+3. High-level flow (optional): a simple diagram or numbered list.
+4. Point to next steps: link to Quickstart, API reference, or tutorials.
+
+**Quickstart** — guide users from zero to working code in under 5 minutes. Principles:
+1. Keep it concise and scannable.
+2. List actionable steps.
+3. Code-first: show snippets before explaining them.
+4. Offer multi-platform examples.
+5. Include troubleshooting tips and link to the error messages guide.
+6. Point to next steps: end with links to API reference, tutorials, or SDK guides.
+
+Tips: minimal prose (code over paragraphs; use prose only to clarify 'why'); call out setup quirks as inline notes; use the same section order across all Quickstarts.
+
+**Guide** — detailed workflows or feature sets. Principles:
+1. Organize into clear numbered steps, each covering a logical unit of work.
+2. Mix prose with code: introduce each step with a brief "why" (1–2 sentences), then show the code snippet.
+3. Highlight essential details using admonitions for notes, tips, and cautions.
+4. Include examples and expected results.
+5. Offer troubleshooting guidance: anticipate common errors and link to the error messages guide.
+6. Point to next steps: link to Quickstart, API reference, or deeper tutorials.
+
+**Admonitions** use Docusaurus syntax:
+```
+:::info
+This is an admonition
+:::
+```