chore: remove stale docs and sync runtime defaults

.github/ISSUE_TEMPLATE/voice_pipeline.yml

@@ -85,7 +85,7 @@ body:
     id: llm_backend
     attributes:
       label: LLM backend + model
-      placeholder: "genie-ai-runtime / phi-4-mini-instruct-q4_k_m.gguf"
+      placeholder: "genie-ai-runtime / Qwen3-4B-Q4_K_M.gguf"
     validations:
       required: false
 

ARCHITECTURE.md

@@ -8,7 +8,7 @@ This repository should be understood as the Rust agent runtime that sits above:
 - GenieOS, the custom L4T and system image layer
 - `genie-voice-runtime`, the external voice runtime for wake/VAD/STT/TTS/audio
 - `genie-home-runtime`, the future AI-native home automation runtime
-- `genie-ai-runtime`, the future Jetson-only LLM inference runtime
+- `genie-ai-runtime`, the external Jetson-first LLM inference runtime
 
 It sits below:
 
@@ -89,9 +89,9 @@ This repo can keep transitional implementations while those layers are still for
 
 The current repo still contains pragmatic adapters used to ship on Jetson now.
 
-| Current adapter | Long-term replacement | Notes |
+| Current adapter | Target boundary | Notes |
 | --- | --- | --- |
-| `genie-ai-runtime` OpenAI-compatible client (default on Jetson) | `llama.cpp` client (selectable fallback) | Both backends ship behind the `LlmClient` facade; per-deployment selection via `[services.llm].backend` in `geniepod.toml`. Backend identity surfaces in `/api/health`, startup logs, and `genie-ctl status`. |
+| `genie-ai-runtime` OpenAI-compatible client (default on Jetson) | external `genie-ai-runtime` service | `llama.cpp` remains a selectable fallback/development backend. Both backends ship behind the `LlmClient` facade; per-deployment selection is via `[services.llm].backend` in `geniepod.toml`. Backend identity surfaces in `/api/health`, startup logs, and `genie-ctl status`. |
 | In-repo voice pipeline under `crates/genie-core/src/voice/` and `voice_loop.rs` | [`genie-voice-runtime`](https://github.com/GeniePod/genie-voice-runtime) | Keep current code as a transitional Jetson bring-up path. New wake/VAD/STT/TTS/audio ownership should move to the external runtime. GenieClaw should consume transcripts and issue speak commands. |
 | Home Assistant provider | `genie-home-runtime` MCP/API client | Keep HA-specific behavior behind `ha/` and tools/home boundaries. |
 | Actuation safety in `genie-core` | final safety in `genie-home-runtime` | Keep current safety as an agent-side guard and confirmation layer. |
@@ -281,7 +281,7 @@ The clean architecture path is incremental:
 1. Make boundary language consistent in docs and config.
 2. Keep Home Assistant and LLM backends behind narrow adapter traits (LLM side resolved via the `LlmClient` facade in `crates/genie-core/src/llm/`).
 3. Move physical actuation authority downward into `genie-home-runtime` when it exists.
-4. Move Jetson model-server specialization downward into `genie-ai-runtime`.
+4. Keep Jetson model-server specialization in `genie-ai-runtime`.
 5. Move voice/audio pipeline ownership downward into `genie-voice-runtime`.
 6. Keep GenieClaw focused on agent policy, memory, skills, tools, channels, and household interaction.
 

CODEBASE.md

@@ -43,7 +43,9 @@ The most important runtime path is:
 
 1. A request enters through the web UI, CLI, REPL, voice loop, or Telegram adapter.
 2. `genie-core` builds prompt context from conversation history and household memory.
-3. The LLM client talks to the local `llama.cpp`-compatible server.
+3. The LLM facade talks to the configured OpenAI-compatible backend. Jetson
+   deploys default to `genie-ai-runtime`; development configs can still use a
+   local `llama.cpp` server.
 4. If the model emits a tool call, the tool parser and dispatcher execute the tool.
 5. Tool results may be returned directly or summarized, depending on the tool.
 6. Conversation state and extracted memory are persisted to SQLite.
@@ -62,7 +64,7 @@ Around that path:
 | `Cargo.lock` | Locked dependency graph. |
 | `Makefile` | Main developer and Jetson deploy entry points: build, test, release, cross-compile, and deploy. |
 | `Dockerfile` | Multi-stage container build for the local dev/runtime image. |
-| `docker-compose.dev.yml` | Dev stack for `genie-core`, `genie-api`, and a local `llama.cpp` server. |
+| `docker-compose.dev.yml` | Dev stack for `genie-core`, `genie-api`, and a local OpenAI-compatible model server. |
 | `README.md` | Product-level overview and repo orientation. |
 | `GETTING_STARTED.md` | Local dev, Docker, and Jetson bring-up guide. |
 | `ARCHITECTURE.md` | Higher-level system architecture narrative. |
@@ -110,8 +112,11 @@ The primary runtime. This is where most product behavior lives.
 | Path | Purpose |
 | --- | --- |
 | `crates/genie-core/src/llm/mod.rs` | LLM module exports. |
-| `crates/genie-core/src/llm/client.rs` | Raw TCP/OpenAI-compatible chat client for the local inference server, including streaming and compatibility fallbacks. |
-| `crates/genie-core/src/llm/retry.rs` | Retry and graceful fallback layer around the base client. |
+| `crates/genie-core/src/llm/openai_compat.rs` | Raw bounded OpenAI-compatible HTTP client used by local and optional provider backends. |
+| `crates/genie-core/src/llm/genie_ai_runtime.rs` | Adapter for the default Jetson `genie-ai-runtime` backend and its request hints. |
+| `crates/genie-core/src/llm/llama_cpp.rs` | Adapter for the legacy/development `llama.cpp` backend. |
+| `crates/genie-core/src/llm/openai_compatible.rs` | Generic OpenAI-compatible provider adapter with bearer-token support. |
+| `crates/genie-core/src/llm/provider.rs` | Optional provider planning and limited-context readiness checks. |
 
 #### Home Assistant Integration