docs: hal0-web v0.8 rebase — factual sync with v0.8.2b4 live system

src/content/docs/blog/whats-new-in-the-v0-7-line.mdx

@@ -14,6 +14,12 @@ tags:
 
 import { Aside } from '@astrojs/starlight/components';
 
+<Aside type="note" title="Historical post">
+This post covers **v0.7.3-beta.1**, a prior beta line. The current line is
+**v0.8.2b4** — see [What's new in the v0.8 line](/blog/whats-new-in-the-v0-8-line)
+for the latest features.
+</Aside>
+
 <Aside type="caution" title="Pre-release">
 hal0 is pre-1.0. **v0.7.3-beta.1** is a *beta* — the current development line,
 not a GA release. Expect breaking changes between cuts. See the

src/content/docs/blog/whats-new-in-the-v0-8-line.mdx

@@ -0,0 +1,105 @@
+---
+title: "What's new in the v0.8 line: declarative config, voice, and Hermes autonomy"
+date: 2026-07-04T12:00:00
+authors: hal0
+excerpt: >-
+  Declarative Stacks replace the flag soup, voice (TTS + STT) lands
+  end-to-end on the NPU, Hermes owns its own config, and hal0-api can
+  finally drop root. A look at v0.8.2b4, the current beta line.
+tags:
+  - release
+  - stacks
+  - voice
+  - hermes
+  - beta
+---
+
+import { Aside } from '@astrojs/starlight/components';
+
+<Aside type="caution" title="Pre-release">
+hal0 is pre-1.0. **v0.8.2b4** is a *beta* — the current development line,
+not a GA release. Expect breaking changes between cuts. See the
+[releases page](/releases) for the stable-vs-next split, and the
+[full changelog](/changelog#v0.8.0-beta.1) for the complete list.
+</Aside>
+
+`v0.8.2b4` is the current beta line. The headline is **declarative
+configuration** — Stacks replace the launch-flag soup with a single source
+of truth — plus **voice** on the NPU, **Hermes autonomy**, and the
+long-awaited ability to run `hal0-api` without root. Everything below
+traces to the [changelog](/changelog#v0.8.0-beta.1) — no roadmap promises,
+just what landed.
+
+## Stacks: declarative config, finally
+
+The slot config tangle — base flags, profile overrides, per-slot
+`extra_args`, and the diverging launch commands they produce — is now a
+single `StackConfig`. A `StackApplyEngine` plans changesets, applies them
+atomically with rollback, and converges the live slot set in one pass.
+Content-hash drift detection tracks when reality diverges from intent,
+and seeded stacks (saber / forge / pi) are derived from the bench so
+you start from a known-good baseline.
+
+![Stack config editor](/screenshots/dashboard-overview.png)
+
+Flag provenance is no longer guesswork. Every slot exposes its resolved
+command at `GET /api/slots/{name}/resolved`, and the Edit drawer badges
+each flag with its source — base, profile, or `extra_args`. No more
+hunting through three files to figure out which one set `--ctx-size`.
+
+## Voice: TTS + STT on the NPU
+
+The voice stack works end-to-end. `voice_wire` is fixed, Open WebUI Call
+mode is wired through the NPU trio, and STT auto-provisions alongside the
+FLM slot. The NPU occupancy grid now carries voice activity alongside
+inference — when Call mode is active, you see it breathe.
+
+## Hermes owns its own config
+
+Hermes is no longer pinned to a hal0-rendered `config.yaml` snapshot.
+A `config-set` overlay layers hal0's keys over Hermes's self-managed
+config, so Hermes can self-migrate across versions while hal0 guarantees
+the integration keys stay right. A real upgrade path (`hal0 agent upgrade
+hermes`) replaces the old tear-down-and-pray cycle.
+
+Two systemd units now: `hal0-agent@hermes` (the core runtime, loopback
+only) and `hermes-gateway` (the messaging platform gateway — Telegram,
+Discord — on `0.0.0.0:8644`). The gateway is a first-class service with
+its own lifecycle.
+
+## hal0-api drops root (opt-in)
+
+Set `HAL0_USER=hal0` and the API runs unprivileged. A declarative
+ownership table (auditable via `hal0 doctor perms`) plus a narrow `sudo`
+seam (`hal0-slotctl`) keeps the unprivileged API in control of slot units
+and container lifecycle while the process itself can't write config it
+shouldn't. Slot containers stay rootful — the container is the sandbox
+boundary. Default `HAL0_USER=root` is unchanged, so existing installs
+don't break.
+
+## The quality-of-life layer
+
+The less flashy changes that operators live with every day:
+
+- **Capability-based fallback.** When a slot's `model.default` isn't
+  locally servable, `load()` falls back to the best registered model
+  matching the slot's capability — no more dead slots because a file
+  moved.
+- **q8_0 KV cache everywhere.** Main and MTP-draft KV caches are now
+  `q8_0` across slots — near-lossless and keeps fused FlashAttention
+  on AMD HIP.
+- **Breaking rename — `hal0/primary` → `hal0/chat`, `hal0/flm` → `hal0/npu`.**
+  Virtual model names now map 1:1 to their resolution chains. Slot-name
+  back-compat is intentionally kept.
+- **Chat-template render-validation.** A broken template can no longer
+  ship silently — every template in the catalog is validated at build time.
+
+## Try it
+
+```sh
+curl -fsSL https://hal0.dev/install.sh | bash
+```
+
+Already running an older line? `hal0 update --channel stable` (or
+`nightly`), and `hal0 update --rollback` reverts atomically. Full notes
+on the [releases page](/releases).

src/content/docs/docs/concepts/architecture.mdx

@@ -61,9 +61,10 @@ state changes to the dashboard footer. These are shared across all requests —
 the API is a thin layer over them.
 
 <Aside type="note">
-hal0 ships with **no built-in authentication or TLS**. The API is open on the
-local network by design. See [Security](/docs/concepts/security) for the
-reverse-proxy pattern you put in front of it.
+hal0 ships with authentication and HTTPS **off by default** — pass `--auth=basic`
+at install to bring up Caddy with basic_auth, bearer tokens, and automatic HTTPS.
+When left off, the API is open on the local network. See
+[Security](/docs/concepts/security) for both postures.
 </Aside>
 
 ## Slots: where models actually run

src/content/docs/docs/concepts/security.mdx

@@ -1,26 +1,28 @@
 ---
 title: Security posture
-description: hal0 ships no built-in auth or TLS — open on your LAN by design. Learn the threat model, the reverse-proxy pattern, and the MCP host and origin allowlists.
+description: Auth is opt-in at install time and off by default for trusted-LAN use. Learn both postures, the --auth=basic flag, the reverse-proxy pattern, and the MCP host and origin allowlists.
 sidebar:
   order: 70
 ---
 
 import { Aside, Card, CardGrid, LinkCard, Steps } from '@astrojs/starlight/components';
 
-hal0 makes one security decision explicit and central: it does **not** ship
-built-in authentication or TLS. The API binds `0.0.0.0:8080` and every endpoint
-is open to anything that can reach that port. This is a deliberate design choice
-(ADR-0012), and the rest of your security posture is built on top of it rather
-than inside it.
+hal0 makes one security decision explicit and central: authentication and TLS are
+**opt-in at install time** and **off by default**. Pass `hal0 install --auth=basic`
+to bring up Caddy as a reverse proxy with basic_auth, bearer tokens, and
+automatic HTTPS. Without it, the API binds `0.0.0.0:8080` and every endpoint is
+open to anything that can reach that port — the right posture for a
+trusted home LAN.
 
 <Aside type="caution">
-Do not expose hal0-api directly to the public internet. It has no auth gate —
-anyone who can reach the port can load models, run inference, drive agents, and
-read activity. Keep it on a trusted network, and put a reverse proxy in front of
-it for anything beyond a single trusted LAN.
+**Without `--auth=basic`**, do not expose hal0-api directly to the public
+internet — anyone who can reach the port can load models, run inference, drive
+agents, and read activity. Keep it on a trusted network, or install with
+`hal0 install --auth=basic` to enable Caddy with basic_auth, bearer tokens, and
+automatic HTTPS.
 </Aside>
 
-## Why no built-in auth?
+## Why auth is opt-in
 
 Earlier designs bundled an auth layer and a TLS-terminating proxy. That was
 removed (ADR-0012) because it duplicated, badly, what every operator already

src/content/docs/docs/getting-started/index.mdx

@@ -19,10 +19,10 @@ OpenAI-compatible `/v1/*` requests to the right slot, and serves the
 dashboard. There is no shared inference daemon to babysit.
 
 <Aside type="note" title="Current release">
-This documentation tracks the **hal0 v0.7.x beta** line — the container-runtime
-era, where each slot (`chat`, `embed`, `rerank`, `stt`, `tts`, `img`, `vision`,
-and `agent` — plus the NPU-conditional slots `stt-npu` and `embed-npu` when
-FastFlowLM is installed) runs in a dedicated podman container. See the
+This documentation tracks the **hal0 v0.8.x beta** line — the container-runtime
+era, where each slot (`agent`, `code`, `img`, `npu`, `rerank`, `tts`, and
+`utility`) runs in a dedicated podman container behind an OpenAI-compatible
+API on port 8080. See the
 [changelog](/changelog) for the exact current version.
 </Aside>
 
@@ -69,9 +69,11 @@ FastFlowLM is installed) runs in a dedicated podman container. See the
   />
 </CardGrid>
 
-<Aside type="caution" title="No built-in auth or TLS">
-`hal0-api` binds `0.0.0.0:8080` and treats every request as authenticated
-by network reachability — the right posture for a trusted LAN appliance.
-If hal0 is reachable from anywhere you do not control, front it with a
-reverse proxy that owns auth and TLS.
+<Aside type="caution" title="Auth is opt-in — trusted LAN by default">
+`hal0-api` binds `0.0.0.0:8080` and, by default, treats every request
+as authenticated by network reachability — the right posture for a
+trusted home LAN. To safely expose hal0 beyond that, install with
+`--auth=basic` to enable a Caddy reverse proxy with HTTPS, basic_auth
+at the edge, and bearer tokens for the API. See
+[Security](/docs/concepts/security) for both postures.
 </Aside>

src/content/docs/docs/guides/connect-mcp.mdx

@@ -17,10 +17,16 @@ servers, mounted on the API as Streamable-HTTP sub-apps:
   server.
 
 <Aside type="caution">
-hal0 has **no built-in network auth** (ADR-0012). The MCP mounts are
-reachable to anything that can reach the API. Run hal0 on a trusted LAN
-or behind your own reverse proxy — do not expose `/mcp/*` to the
-internet.
+hal0 has **no built-in network auth by default** (ADR-0012). Without
+`--auth=basic`, the MCP mounts are reachable to anything that can reach
+the API. Run hal0 on a trusted LAN or behind your own reverse proxy —
+do not expose `/mcp/*` to the internet.
+
+When `--auth=basic` is active, the MCP mounts sit behind the same Caddy
+instance as the rest of the API — they inherit Caddy's TLS and basic_auth
+gating automatically. The host and origin allowlists below still apply
+at the MCP transport layer, but the outer auth gate is already handled.
+See [Security](/docs/concepts/security) for the two deployment postures.
 </Aside>
 
 ## Connect

src/content/docs/docs/guides/voice-stt-tts.mdx

@@ -12,8 +12,9 @@ hal0 exposes two voice directions through OpenAI-compatible endpoints:
 - **Speech-to-text** (`POST /v1/audio/transcriptions`) — `whisper-v3:turbo` served by
   FastFlowLM (FLM) on the XDNA NPU, co-loaded with the chat model in one FLM process
   (the "NPU trio": chat + STT + embed).
-- **Text-to-speech** (`POST /v1/audio/speech`) — Kokoro-82M ONNX container on CPU,
-  54 voices, MP3 by default (default voice: `af_heart`).
+- **Text-to-speech** (`POST /v1/audio/speech`) — two backends:
+  - **Qwen3-TTS** (ROCm, GPU): natural speech with voice cloning. Default on Strix Halo.
+  - **Kokoro-82M** (CPU, ONNX): lightweight, 54 voices, MP3 by default (default voice: `af_heart`). Runs without GPU.
 
 Both are children of the **`voice`** capability: `voice.stt` maps to the
 `stt` slot and `voice.tts` maps to the `tts` slot.
@@ -33,6 +34,15 @@ curl -X POST http://localhost:8080/api/capabilities/voice/stt \
   -d '{"enabled": true, "backend": "npu"}'
 ```
 
+</TabItem>
+<TabItem label="Text-to-speech (GPU, ROCm — Qwen3-TTS)">
+
+```sh
+curl -X POST http://localhost:8080/api/capabilities/voice/tts \
+  -H 'Content-Type: application/json' \
+  -d '{"enabled": true, "backend": "rocm", "provider": "qwen3-tts"}'
+```
+
 </TabItem>
 <TabItem label="Text-to-speech (CPU, Kokoro)">
 
@@ -49,7 +59,9 @@ The response is `{ "ok": true, "selection": { ...current selection... } }`.
 
 <Aside type="note">
 STT runs on the XDNA NPU via the FLM container — not on CPU. There is no CPU
-whisper.cpp backend in hal0. Kokoro TTS runs on CPU (ONNX). Neither uses ROCm.
+whisper.cpp backend in hal0. TTS has two options: Qwen3-TTS on ROCm (GPU, voice
+cloning, default on Strix Halo) and Kokoro-82M on CPU (ONNX, lightweight, 54
+voices). Neither STT backend uses ROCm.
 </Aside>
 
 ### How the NPU STT path works
@@ -96,10 +108,10 @@ As with transcription, `model` is required — a missing or empty `model`
 returns **400** (`request.missing_model`).
 
 <Aside type="note">
-Kokoro weights are operator-staged on the configured model store. The Kokoro
-toolbox is a containerized server managed by hal0 as a podman slot — no manual
-container management needed. It implements `POST /v1/audio/speech`,
-`GET /v1/models`, and `GET /health`.
+TTS weights are operator-staged on the configured model store. Both Qwen3-TTS
+(ROCm) and Kokoro (CPU) run as containerized servers managed by hal0 as podman
+slots — no manual container management needed. Each implements `POST
+/v1/audio/speech`, `GET /v1/models`, and `GET /health`.
 </Aside>
 
 ## See also