[Sync] replace the flat layout with the unified stow tree

Supersedes the old flat .config/ layout (last published 2026-06-28) with the private repo's structure: one shared base plus per-host overlays. - packages: common/ gui/ lw/ fl/ wm/ plus install.sh and bin/ tooling (dotsync, reconcile-hyde.sh) - new README covering the layout, deploy order and the HyDE dependency - current HyDE waybar rig (layouts/, cava), pi agent extensions, claude/ config, tmux, presenterm, aichat roles - fish: kp (keepassxc-cli + fzf picker, db path from $KP_DB) and bind_M_n_history (alt+1..9 recalls the nth history entry) - drops cruft that should never have been tracked: the duplicate top-level .pi/ copy, btop.log, zellij config.kdl.bak, fish_variables - .pi/agent/auth.json is gitignored; auth.json.example ships instead Host-specific work sessions and the personal backlog stay in the private tree. Endpoint locators in the llamacpp/whisper guides are placeholders ($SERVER, <own-domain>) — the guides themselves stay, since they are the useful part.
2026-08-13 02:53:06 +02:00
parent f62cb40499
commit 1e39e0cb4b
314 changed files with 12306 additions and 10772 deletions
@@ -0,0 +1,82 @@
+# LW HOST OVERLAY — full-file copy of common/.config/aichat/config.yaml plus the `lw` client
+# (this laptop's own llama.cpp server at 127.0.0.1:11343; see lw/.config/llamacpp/README.md).
+# Stowed with --override so it shadows the common file on lw only. When the common config
+# changes, mirror the change here.
+# see https://github.com/sigoden/aichat/blob/main/config.example.yaml
+keybindings: vi
+editor: nvim
+model: local:Qwen3-Coder-30B-Instruct-UD-Q3_K_XL
+
+# Sessions: persist REPL sessions and keep more history before summarizing.
+# The default compress_threshold (4000) summarizes far too early for 24k+ windows.
+save_session: true
+compress_threshold: 16000
+
+# REPL prompts show live context usage (needs max_input_tokens, set per model below)
+left_prompt:  '{color.green}{?session {session}{?role /}}{role}{color.cyan}{?rag @{rag}}{color.reset}> '
+right_prompt: '{color.purple}{?session {consume_tokens}/{max_input_tokens} }{color.reset}'
+
+# NOTE: temperature/top_p are intentionally unset — the LAN server applies tuned
+# per-model sampling via --jinja (e.g. GLM 0.6/0.95); a global value would clobber it.
+
+# max_input_tokens = real ctx-size (from the router) minus output headroom.
+clients:
+# This laptop's own CPU server (llamaserver abbr; works everywhere, no network needed).
+# Speeds are benched decode t/s — see lw/.config/llamacpp/README.md roster.
+- type: openai-compatible
+  name: lw
+  api_base: http://127.0.0.1:11343/v1
+  models:
+  - name: gemma-4-E2B-it-UD-Q4_K_XL
+    max_input_tokens: 3000             # ctx 4096 · ~6 t/s — fastest, quick Q&A
+  - name: Granite-4.0-H-Tiny
+    max_input_tokens: 6000             # ctx 8192 · ~4.8 t/s — speed AND brains
+  - name: Qwen3-1.7B
+    max_input_tokens: 6000             # ctx 8192 · ~4.8 t/s — thinking, snappy
+  - name: Qwen3-4B-Instruct-2507
+    max_input_tokens: 3000             # ctx 4096 · ~3.9 t/s — daily driver
+  - name: Qwen2.5-Coder-3B-Instruct
+    max_input_tokens: 6000             # ctx 8192 · ~3.3 t/s — small coder
+  - name: Qwen3-30B-A3B-Instruct-2507-UD-IQ3_XXS
+    max_input_tokens: 3000             # ctx 4096 · ~2.7 t/s — quality when you can wait
+  - name: Jan-v3-4b
+    max_input_tokens: 3000             # ctx 4096 · ~2.6 t/s — agentic tune
+  - name: gemma-4-E4B-it-UD-Q4_K_XL
+    max_input_tokens: 3000             # ctx 4096 · ~2.5 t/s — quality (vision once mmproj added)
+# LAN access to the fl box (fast; only reachable on the home network)
+- type: openai-compatible
+  name: local
+  api_base: http://192.168.0.204:11343/v1
+  models: &lan_models
+  - name: Qwen3-Coder-30B-Instruct-UD-Q3_K_XL
+    max_input_tokens: 30000            # ctx 32768
+  - name: Qwen3-Coder-Next-UD-IQ3_XXS
+    max_input_tokens: 128000           # ctx 131072
+  - name: Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS
+    max_input_tokens: 22000            # ctx 24576
+    supports_vision: true
+  - name: Qwen3.6-35B-A3B-Thinking
+    max_input_tokens: 22000            # ctx 24576
+    supports_vision: true
+  - name: Qwen3.5-9B-UD-Q6_K_XL
+    max_input_tokens: 30000            # ctx 32768
+    supports_vision: true
+  - name: gemma-4-26B-A4B-it-UD-IQ4_XS
+    max_input_tokens: 22000            # ctx 24576
+    supports_vision: true
+  - name: gemma-4-E4B-it-UD-Q8_K_XL
+    max_input_tokens: 62000            # ctx 65536
+    supports_vision: true
+  - name: GLM-4.7-Flash-UD-Q4_K_XL
+    max_input_tokens: 22000            # ctx 24576
+  - name: gpt-oss-20b
+    max_input_tokens: 62000            # ctx 65536
+  - name: gpt-oss-20b-low
+    max_input_tokens: 62000            # ctx 65536
+# Remote access via duskadiy.com (reachable from anywhere; requires an API key —
+# $DUSKADIY_API_KEY, from the gitignored conf.d/secrets.fish). The LAN endpoint above is keyless.
+# Same server, same model ids — reuse the list above via a YAML anchor.
+- type: openai-compatible
+  name: duskadiy
+  api_base: https://llm.duskadiy.com/api/v1
+  models: *lan_models
@@ -0,0 +1,4 @@
+# --host 127.0.0.1 deliberately: laptop roams onto untrusted wifi and the API has no key.
+# --cors-origins localhost: llama.cpp PR 25655 recommendation for same-machine use — without it
+# the default CORS reflects any Origin, letting a malicious page in the local browser call the API.
+abbr llamaserver "nice -n 19 llama-server --host 127.0.0.1 --port 11343 --cors-origins localhost --models-max 1 --models-preset ~/.config/llamacpp/config.ini"
@@ -0,0 +1,4 @@
+# lw host extras (sourced last by ~/.config/hypr/hyprland.conf)
+# No discrete GPU on this laptop — no nvidia.conf.
+
+exec-once = xremap ~/.config/xremap/config.yml # keyboard remaps (lw is the only xremap host)
@@ -0,0 +1,21 @@
+# lw hyprlock sizes — 1920x1080, scale 1.
+#
+# gui/.config/hypr/hyprlock.conf ships sizes tuned for wm's 3072x1728 logical panel; these
+# are those values × 0.625, which keeps the same proportions the layout was designed around
+# (clock ~13% of screen height, date ~2.5%, meta ~1.5%, logo ~16%).
+#
+# Caveat — lw is multi-monitor and hyprlock draws on every output. monitors.conf makes
+# $mainScreen (eDP-1) 1366x768, so on the built-in panel these read ~40% larger than
+# intended (clock ≈ 18% of its height). Sized for the 1920x1080 externals on purpose; if
+# the laptop panel is the one you actually lock on, use the 768-tuned set instead:
+#   $fsClock = 98  $fsDate = 20  $fsMeta = 12  $fsSmall = 11  $fsIcon = 18  $logoSize = 124
+$fsClock = 138
+$fsDate = 28
+$fsMeta = 16
+$fsSmall = 15
+$fsIcon = 25
+$logoSize = 175
+
+# No $y* overrides: the shared offsets in hyprlock.conf already spread the content to the
+# edges, and they land the same on both hosts because each host's sizes are the same
+# proportion of its screen. Override them here if lw ever needs to differ.
@@ -0,0 +1,64 @@
+
+# █▀▄▀█ █▀█ █▄░█ █ ▀█▀ █▀█ █▀█ █▀
+# █░▀░█ █▄█ █░▀█ █ ░█░ █▄█ █▀▄ ▄█
+
+# Set your monitor configuration here
+# See https://wiki.hyprland.org/Configuring/Monitors/
+# For a sample file, please refer to https://github.com/prasanthrangan/hyprdots/blob/main/Configs/.config/hypr/monitors.t2
+
+$mainScreen = eDP-1
+$leftScreen = DP-2
+$rightScreen = HDMI-A-1
+
+monitor = $mainScreen, 1366x768@60.14, 0x0, 1 # main screen
+monitor = $leftScreen, 1920x1080@60, -1920x0, 1 # left 
+monitor = $rightScreen, 1920x1080@60,1366x0, 1 # right
+
+# monitor=HDMI-A-1,1920x1080@60,-1366x0,1 # hdmi on left
+# monitor=HDMI-A-1,1920x1080@60,0x-768,1 # hdmi on top 
+# monitor=DP-2,640x480@59.94Hz,0x0,1 # vga
+
+# monitor = HDMI-A-1, 1366x768@60.00Hz, auto , 1, mirror, eDP-1 #mirror lapscreen 
+# monitor = DP-2, 1920x1080@60.00Hz, auto , 1, mirror, eDP-1 #mirror lapscreen jag3
+# monitor = eDP-1, 1920x1080@60.00Hz, auto , 1, mirror, HDMI-A-1 #main hdmi, lapscreen mirrors
+# monitor = eDP-1, 1920x1080@60.00Hz, auto , 1, mirror, DP-2 #main VGA, lapscreen mirrors
+
+# monitor = , preferred, auto, 1, mirror, DP-2 # dmz projector
+# monitor = eDP-1, 1920x1080@60, auto , 1, mirror, DP-2 #mirror vga test
+
+#Main Monitor
+workspace = 1, monitor:$mainScreen, default:true
+workspace = 2, monitor:$mainScreen
+workspace = 3, monitor:$mainScreen
+workspace = 4, monitor:$mainScreen
+workspace = 5, monitor:$mainScreen
+workspace = 6, monitor:$mainScreen
+workspace = 7, monitor:$mainScreen
+workspace = 8, monitor:$mainScreen
+workspace = 9, monitor:$mainScreen
+workspace = 10,monitor:$mainScreen
+
+# Second Monitor
+workspace = 11, monitor:$rightScreen, default:true
+workspace = 12, monitor:$rightScreen
+workspace = 13, monitor:$rightScreen
+workspace = 14, monitor:$rightScreen
+workspace = 15, monitor:$rightScreen
+workspace = 16, monitor:$rightScreen
+workspace = 17, monitor:$rightScreen
+workspace = 18, monitor:$rightScreen
+workspace = 19, monitor:$rightScreen
+workspace = 20, monitor:$rightScreen
+
+# Third Monitor
+workspace = 21, monitor:$leftScreen, default:true
+workspace = 22, monitor:$leftScreen
+workspace = 23, monitor:$leftScreen
+workspace = 24, monitor:$leftScreen
+workspace = 25, monitor:$leftScreen
+workspace = 26, monitor:$leftScreen
+workspace = 27, monitor:$leftScreen
+workspace = 28, monitor:$leftScreen
+workspace = 29, monitor:$leftScreen
+workspace = 30, monitor:$leftScreen
+
@@ -0,0 +1,2 @@
+# lw (laptop, smaller screen)
+font_size 9.0
@@ -0,0 +1,13 @@
+# run with `kitty --session ~/.config/kitty/music.conf`
+# lw override: smaller split resizes than the shared version, to fit the laptop screen.
+
+new_tab music
+layout splits
+cd ~
+
+launch --var window=first cava
+launch --location=hsplit pyradio
+resize_window taller 10
+focus
+launch --location=vsplit pipes-rs
+resize_window wider 20
@@ -0,0 +1,133 @@
+# Local LLM server — laptop (lw), setup & usage
+
+llama.cpp in **router mode** on the laptop. This is the **CPU** counterpart of the fl box
+(`fl/.config/llamacpp/`) — lw has **no discrete GPU**, so everything runs on the CPU and
+the only resource ceiling is **system RAM** (not VRAM). None of fl's VRAM/freeze rules apply.
+
+- **Box**: Intel Core i7-5600U (Broadwell 2015, **2 cores / 4 threads**, AVX2) · iGPU HD Graphics
+  5500 (no compute) · **15.49 GiB RAM** · DDR3L-1600 dual-channel: ~20 GB/s theoretical, **~8–10
+  GB/s effective measured** — the dense-decode bottleneck → prefer **small quants + MoE**.
+- **Endpoint**: `http://127.0.0.1:11343/v1` — **local only** (server binds `127.0.0.1`; deliberate:
+  the laptop roams onto untrusted wifi and the API has no key, unlike the stationary fl box).
+- **Files here** (`~/.config/llamacpp/`, stowed from the repo's `lw/.config/llamacpp/`):
+  `config.ini` (presets — section names ARE the API model ids), `bench.py` (CPU benchmark),
+  `bench-results.md` / `bench-history.md` (ledgers, generated by bench.py). The GGUF blobs
+  themselves live in `~/software/models/` and are not tracked.
+
+## How it runs
+
+Launched by the fish abbr **`llamaserver`** (defined in `lw/.config/fish/host.fish`;
+llama.cpp is pacman-installed, `llama-server` is on PATH):
+
+```
+nice -n 19 llama-server --host 127.0.0.1 --port 11343 --cors-origins localhost \
+  --models-max 1 --models-preset ~/.config/llamacpp/config.ini
+```
+
+- Router mode: each requested model loads on demand in a child process.
+- **`--models-max 1`**: one model resident; a new request LRU-evicts the previous. There's no
+  fl-style freeze risk here (no display-driving GPU) — this is purely the safest RAM posture on
+  15 GB with no co-residency swap. Raise to 2–3 only to keep small models hot together
+  (E2B+4B+1.7B ≈ 6 GB fits). `sleep-idle-seconds` per preset frees idle models either way.
+- **`nice -n 19`**: on 2 cores generation pins the whole CPU — niceness makes inference yield to
+  interactive apps (full speed when the desktop is idle).
+- **`--cors-origins localhost`**: same-machine profile from llama.cpp PR 25655 — the default CORS
+  reflects any Origin, so without this a malicious page in the laptop's own browser could call
+  the (keyless) API on 127.0.0.1. Browser-only enforcement: bench.py/CLI clients are unaffected.
+- `config.ini` here is the stowed repo copy (`lw/.config/llamacpp/`), so editing it edits the
+  repo (no deploy step).
+
+## RAM (the golden rule)
+
+There is no VRAM and no freeze risk. The constraint is plain: **N resident models must fit in
+system RAM together.** `bench.py` records each model's RAM footprint (MemAvailable delta) and
+the remaining free RAM — keep enough headroom for the desktop (and for co-resident models if you
+ever raise `--models-max` above 1). If RAM gets tight: shorten `sleep-idle-seconds`, use smaller
+quants, or reduce `ctx-size`. Note: the 4B/E2B presets are small enough to co-reside, but a big
+MoE (e.g. Qwen3-30B-A3B at ~12 GB) fills most of 15 GB → it runs **1-at-a-time** regardless.
+
+`threads = 2` (physical cores) on the dense presets — they measure at the practical bandwidth
+wall (~8–10 GB/s effective) and the 2026-07-26 sweep confirmed more threads don't lift them.
+The **compute-bound** pair runs `threads = 3` (same-run sweep: Granite 4.5→4.8, 30B 1.9→2.0;
+`threads = 4` LOSES to HT contention on both). Every preset sets `threads-batch = 4`: prefill
+is compute-bound on all models — benched +8% prefill / −9% TTFT with no decode cost, and the
+extra HT load only bursts during prefill.
+
+## Tuning cheat-sheet (CPU)
+
+- **Quant size is the speed dial**: decode is bandwidth-bound, so a smaller quant is directly
+  faster. Prefer Q4_K_M / Q5_K_M; F16/Q8 are usually quality-overkill AND slow on CPU.
+- **`threads`** = **physical** core count (`lscpu` → Core(s)/socket × Socket(s)). Hyperthreads
+  rarely help decode; oversubscribing can hurt. Set this per preset (default 4 is a guess).
+- **Model size**: stick to ≤ ~8B. **MoE** models (few active params, e.g. Qwen3-A3B) are ideal
+  IF they fit RAM — they decode at their *active* size while giving bigger-model quality. There
+  is no `n-cpu-moe` tradeoff here — it's all CPU already.
+- **MTP / speculative decode** (`spec-type = draft-mtp`): a clear win on the fl GPU (+50–100%) but
+  **TESTED AND REJECTED on this box** (2026-07-26): the draft lost the final CPU-pinned
+  (`gpu-layers-draft = 0`) same-run A/Bs on both gemmas — E2B 4.0 vs 4.9 t/s at 47% acceptance,
+  E4B 2.3 vs 2.4 at 62%. Batched verify triples the FLOPs per weight-read and 2 cores have
+  no spare compute, so spec decode can't pay here at any setting. MTP stays an fl-only trick.
+  (Only gemma-4 has drafts anyway; Qwen 2507/2.5 have no MTP path — that needs Qwen 3.5/3.6.)
+- **KV cache**: leave at default (f16). Quantized KV (`cache-type-*`) adds dequant overhead with
+  no VRAM to reclaim → usually slower on CPU. (This is the opposite of the fl config.)
+- **iGPU / Vulkan**: the pacman build **does** have a Vulkan backend and by default auto-offloads
+  layers to the HD 5500 — that's why every preset pins `n-gpu-layers = 0`. The iGPU shares the
+  same DDR3, is slower than the cores, and big allocations die (`ErrorOutOfDeviceMemory`, SIGSEGV
+  in llama-cli) — it broke the 30B load outright. Only re-enable offload if a bench proves the
+  iGPU wins (it won't).
+
+## Benchmarking
+
+Run **on the laptop** (against its own localhost server):
+
+```
+./bench.py                     # full sweep of everything the server lists
+./bench.py -m id1,id2          # just the changed presets
+./bench.py -n 512 --ctx 8000   # longer gen + a long-context decode column
+```
+
+For a clean per-model RAM footprint, restart the server first (with `--models-max 1` only a
+re-bench of the *same* still-resident model reads a `~0` RAM Δ, but a restart keeps runs comparable). Results land in
+`bench-results.md` (latest) + `bench-history.md` (append-only — diff runs there).
+
+⚠ Absolute numbers swing **±20–30%** with desktop load and thermals (E2B has measured 6.1 idle
+vs 4.9 warm/busy). Only rows from the **same run** are directly comparable — A/B via extra
+preset ids in one `./bench.py -m a,b` command (see the rig sections in config.ini), never by
+comparing across runs. For spec/MTP rows the draft **acceptance %** appears in the spec column.
+
+## Model roster (reference numbers = clean sweep 2026-07-25, n-gpu-layers=0; Granite/30B since
+promoted to threads=3 and all presets to threads-batch=4, 2026-07-26)
+
+| model id | decode t/s | prefill t/s | ctx | extras | role |
+|---|---|---|---|---|---|
+| `gemma-4-E2B-it-UD-Q4_K_XL` | **6.1** | 8 | 4k | vision-capable (mmproj) | fastest — quick Q&A, askllama-class |
+| `Qwen3-1.7B` | 4.8 | 11 | 8k | thinking | snappy small tasks |
+| `Granite-4.0-H-Tiny` | 4.3–4.8 | 11–14 | 8k | 7B/~1B-active hybrid MoE, threads 3 | ⭐ speed AND brains |
+| `Qwen3-4B-Instruct-2507` | 3.9 | 7 | 4k | non-thinking | daily driver |
+| `Qwen2.5-Coder-3B-Instruct` | 3.3 | 6 | 8k | temp 0.2 | small coder |
+| `Qwen3-30B-A3B-Instruct-2507-UD-IQ3_XXS` | 1.9–2.7 | 4 | 4k | ~12 GB, TTFT ~6 s, threads 3 | ⭐ quality when you can wait |
+| `gemma-4-E4B-it-UD-Q4_K_XL` | 2.4–2.5 | 4 | 4k | vision (mmproj) | quality vision; else the 30B is smarter at the same speed |
+| `Jan-v3-4b` | 2.1¹ | 4 | 4k | agentic tune | agentic variant of the 4B (~0.9× its speed) |
+
+¹ variance-depressed in the sweep; same-run A/B vs the base (2026-07-26) shows ~0.9×.
+
+## Client wiring
+
+Clients run **on the laptop itself** (the server is localhost-only, see Endpoint above); same
+rule as fl otherwise: client model **id = config.ini section name**, client context ≤ server
+`ctx-size`. Point it at `http://127.0.0.1:11343/v1` (no key). If LAN access is ever needed,
+switch the abbr back to `--host 0.0.0.0` **and add `--api-key`** — never open it keyless.
+
+**Wired (2026-07-26), lw-only via stow host overlays** — these are full-file `--override`
+copies of the common configs with an added local provider; when the common client configs
+change, mirror the change here:
+
+| tool | overlay file | provider / client id |
+|---|---|---|
+| aichat | `lw/.config/aichat/config.yaml` | `lw:` (e.g. `lw:gemma-4-E2B-it-UD-Q4_K_XL`) |
+| opencode | `lw/.config/opencode/opencode.json` | `lwcpp/` |
+| pi | `lw/.pi/agent/models.json` + `settings.json` | `lwcpp` (⚠ pi hides providers not in settings.json `enabledModels`) |
+
+All 8 presets are listed in each, with client context capped to the server `ctx-size` and the
+benched t/s in the display names. Defaults still point at the fl LAN models — switch to a local
+model in-tool when roaming. Re-run `./install.sh` on lw once so stow links the new files.
@@ -0,0 +1,262 @@
+#!/usr/bin/env python3
+"""
+llama.cpp router benchmark — LAPTOP (lw) edition → writes a results ledger for tuning.
+
+This is the CPU analog of the fl (RX 7600 XT) bench.py. lw has NO discrete GPU, so all
+the ROCm / VRAM / GTT / freeze-guard / force-unload machinery from the fl version is GONE
+— on a CPU box there is no display-driving GPU to starve. The only resource ceiling is
+system RAM, so instead of VRAM this records per-model **RAM footprint** (MemAvailable
+delta, from /proc/meminfo) alongside decode/prefill/TTFT.
+
+Per model it records decode/prefill t/s, TTFT, RAM used (Δ) / RAM free, and the preset's
+knobs (ctx, threads, spec-type) parsed from config.ini. For spec/MTP presets the draft
+acceptance % is appended to the spec column, read from the response's `timings` object
+(the child server also prints a `draft acceptance` log line; the table is easier to compare).
+
+Output: bench-results.md (latest run, overwritten) + bench-history.md (every run, appended)
+— both next to this script. Run ON THE LAPTOP against its own server. Stdlib only.
+
+  ./bench.py                            # all models the server lists
+  ./bench.py -m id1,id2                 # only these
+  ./bench.py -x some-id                 # skip some (e.g. not downloaded yet)
+  ./bench.py -n 512 --ctx 8000          # longer gen + a long-context decode column
+
+Note: the laptop runs `--models-max 1` (LRU eviction), so mid-sweep the RAM Δ is NET —
+each load evicts the previous model, and a smaller model after a bigger one reads negative.
+Only the first model after a server restart shows a true footprint.
+Compare a change: edit config.ini → restart server → re-run → diff runs in bench-history.md.
+"""
+import argparse, json, os, re, subprocess, sys, time, urllib.request, urllib.error
+from datetime import datetime
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+
+
+def mem_available():
+    """(available_bytes, total_bytes) from /proc/meminfo; (None, None) if unavailable."""
+    try:
+        info = {}
+        for ln in open("/proc/meminfo", encoding="utf-8"):
+            k, _, rest = ln.partition(":")
+            m = re.search(r"(\d+)\s*kB", rest)
+            if m:
+                info[k.strip()] = int(m.group(1)) * 1024
+        return info.get("MemAvailable"), info.get("MemTotal")
+    except OSError:
+        return None, None
+
+
+def with_retries(fn, retries, wait, label=""):
+    """Retry on 5xx/connection errors (model still mounting). A deterministic 'failed to
+    load' from the router is NOT retried — that's a broken preset, not a slow mount."""
+    last = None
+    for i in range(retries + 1):
+        try:
+            return fn()
+        except urllib.error.HTTPError as e:
+            if e.code < 500:              # 4xx = real client error
+                raise
+            body = ""
+            try:
+                body = e.read().decode("utf-8", "ignore")
+            except Exception:
+                pass
+            if "failed to load" in body:  # deterministic → don't burn retries
+                raise RuntimeError("model failed to load — check the llama-server log") from None
+            last = e
+        except (urllib.error.URLError, ConnectionError, TimeoutError) as e:
+            last = e
+        if i < retries:
+            print(f"    … {label} not ready ({last}); waiting {wait}s (try {i + 1}/{retries})", file=sys.stderr)
+            time.sleep(wait)
+    raise last
+
+
+def _open(url, key, payload=None, timeout=600):
+    data = json.dumps(payload).encode() if payload is not None else None
+    hdrs = {"Authorization": f"Bearer {key}"}
+    if data:
+        hdrs["Content-Type"] = "application/json"
+    return urllib.request.urlopen(urllib.request.Request(url, data=data, headers=hdrs), timeout=timeout)
+
+
+def list_models(base, key):
+    d = json.load(_open(base.rstrip("/") + "/models", key))
+    return [m["id"] for m in d.get("data", [])]
+
+
+def parse_config(path):
+    """{preset: {ctx, threads, spec}} from the router config.ini."""
+    sections, sec = {}, None
+    try:
+        lines = open(path, encoding="utf-8").read().splitlines()
+    except OSError:
+        return {}
+    for ln in lines:
+        s = ln.strip()
+        if s.startswith("[") and s.endswith("]"):
+            sec = s[1:-1]; sections[sec] = {}
+        elif sec and "=" in s and not s.startswith("#"):
+            k, v = s.split("=", 1)
+            sections[sec][k.strip()] = v.split("#", 1)[0].strip()
+    return {sec: {"ctx": kv.get("ctx-size", "-"),
+                  "threads": kv.get("threads", "-"),
+                  "spec": kv.get("spec-type", "-") or "-"}
+            for sec, kv in sections.items()}
+
+
+def run(base, key, model, prompt, n, temp, timeout):
+    payload = {"model": model, "messages": [{"role": "user", "content": prompt}],
+               "max_tokens": n, "temperature": temp, "stream": True,
+               "stream_options": {"include_usage": True},
+               # llama.cpp extension: makes the server embed its own timings in the stream →
+               # the table shows the SAME predicted_per_second as the llama.cpp web UI
+               # (without it we fall back to wall-clock estimates, which read lower).
+               "timings_per_token": True}
+    t0 = time.perf_counter(); ttft = None; ntok = 0; usage = None; timings = None
+    for raw in _open(base.rstrip("/") + "/chat/completions", key, payload, timeout):
+        line = raw.decode("utf-8", "ignore").strip()
+        if not line.startswith("data:"):
+            continue
+        body = line[5:].strip()
+        if body == "[DONE]":
+            break
+        try:
+            c = json.loads(body)
+        except ValueError:
+            continue
+        ch = c.get("choices") or [{}]
+        d = ch[0].get("delta", {}) if ch else {}
+        # reasoning models stream `reasoning_content` first and may never emit `content`
+        if d.get("content") or d.get("reasoning_content"):
+            if ttft is None:
+                ttft = time.perf_counter() - t0
+            ntok += 1
+        if c.get("usage"):
+            usage = c["usage"]
+        if c.get("timings"):
+            timings = c["timings"]
+    total = time.perf_counter() - t0
+    comp = (usage or {}).get("completion_tokens") or ntok
+    ptok = (usage or {}).get("prompt_tokens")
+    gen_s = total - (ttft or total)
+    tg = comp / gen_s if gen_s > 0 else 0.0
+    pp = (ptok / ttft) if (ptok and ttft) else None
+    acc = None
+    if timings:  # llama.cpp's own numbers are authoritative when the router forwards them
+        tg = timings.get("predicted_per_second", tg)
+        pp = timings.get("prompt_per_second", pp)
+        ptok = timings.get("prompt_n", ptok)
+        if timings.get("prompt_ms"):
+            ttft = timings["prompt_ms"] / 1000
+        if timings.get("draft_n"):  # spec decode active — mirror the child's acceptance stat
+            acc = timings.get("draft_n_accepted", 0) / timings["draft_n"]
+    return {"tg": tg, "pp": pp, "ttft": ttft or 0.0, "ptok": ptok or 0, "acc": acc}
+
+
+def gb(b):
+    return f"{b / 1e9:.1f} GB" if b else "-"
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("-u", "--url", default="http://127.0.0.1:11343/v1",
+                    help="run this ON the laptop against its own server (localhost)")
+    ap.add_argument("-k", "--key", default="no-key-required")
+    ap.add_argument("-m", "--models", default="all", help="comma-separated ids, or 'all'")
+    ap.add_argument("-x", "--skip", default="", help="comma-separated ids to skip")
+    ap.add_argument("-n", "--tokens", type=int, default=256)
+    ap.add_argument("-t", "--temp", type=float, default=0.3)
+    ap.add_argument("--ctx", type=int, default=0, help="also record decode at ~this many prompt tokens")
+    ap.add_argument("--timeout", type=int, default=600, help="per-request stall timeout (s)")
+    ap.add_argument("--config", default=os.path.join(HERE, "config.ini"))
+    ap.add_argument("--out", default=os.path.join(HERE, "bench-results.md"))
+    ap.add_argument("--history", default=os.path.join(HERE, "bench-history.md"))
+    ap.add_argument("--retries", type=int, default=4, help="retries while a model is still mounting")
+    ap.add_argument("--retry-wait", type=int, default=15, help="seconds between retries")
+    ap.add_argument("--settle", type=float, default=1.0,
+                    help="seconds to wait after warmup before reading RAM (let allocation settle)")
+    a = ap.parse_args()
+
+    models = list_models(a.url, a.key) if a.models == "all" else [m.strip() for m in a.models.split(",")]
+    skip = {s.strip() for s in a.skip.split(",") if s.strip()}
+    models = [m for m in models if m not in skip]
+    cfg = parse_config(a.config)
+
+    avail0, total_ram = mem_available()
+
+    task = "Write a Python function that merges two sorted lists, with a short docstring and one example."
+    filler = ("The quick brown fox jumps over the lazy dog. " * max(1, a.ctx // 9)) if a.ctx else None
+
+    cols = ["model", "cfg ctx", "threads", "spec", "decode t/s", "prefill t/s", "TTFT s", "RAM Δ", "RAM free"]
+    if a.ctx:
+        cols.append(f"decode@{a.ctx // 1000}k")
+
+    latest = open(a.out, "w", encoding="utf-8")
+    hist = open(a.history, "a", encoding="utf-8")
+    meta = (f"_Run {datetime.now():%Y-%m-%d %H:%M} · server `{a.url}` · RAM total {gb(total_ram)} · "
+            f"baseline avail {gb(avail0)} · gen {a.tokens} tok (CPU inference, no GPU)_")
+
+    def emit(line, both=True):
+        print(line)
+        latest.write(line + "\n"); latest.flush()
+        if both:
+            hist.write(line + "\n"); hist.flush()
+
+    latest.write("# llama.cpp benchmark results — laptop (lw), latest run\n\n")
+    hist.write(f"\n## run {datetime.now():%Y-%m-%d %H:%M}\n\n")
+    emit(meta + "\n")
+    emit("| " + " | ".join(cols) + " |")
+    emit("|" + "|".join(["---"] * len(cols)) + "|")
+
+    for m in models:
+        c = cfg.get(m, {})
+        base_cells = [m, c.get("ctx", "-"), c.get("threads", "-"), c.get("spec", "-")]
+        pre_avail, _ = mem_available()   # per-model baseline (a model may already be resident)
+        try:
+            # warmup — retries wait out the mount; 'failed to load' aborts immediately
+            with_retries(lambda: run(a.url, a.key, m, "hi", 8, a.temp, a.timeout), a.retries, a.retry_wait, m)
+            if a.settle:
+                time.sleep(a.settle)
+            post_avail, _ = mem_available()
+            free = post_avail
+            delta = (pre_avail - post_avail) if (pre_avail is not None and post_avail is not None) else None
+            # with --models-max 1 the Δ is net-of-eviction: negative = replaced a bigger model
+            if delta is None:
+                delta_cell = "-"
+            elif delta > 0.2e9:
+                delta_cell = gb(delta)
+            elif delta < -0.2e9:
+                delta_cell = f"-{gb(-delta)} (net swap)"
+            else:
+                delta_cell = "~0 (resident/net)"
+            r = with_retries(lambda: run(a.url, a.key, m, task, a.tokens, a.temp, a.timeout),
+                             a.retries, a.retry_wait, m)
+            cells = base_cells.copy()
+            if r.get("acc") is not None:
+                cells[3] = f"{cells[3]} ({r['acc'] * 100:.0f}% acc)"
+            cells += [f"{r['tg']:.1f}", (f"{r['pp']:.0f}" if r['pp'] else "-"),
+                      f"{r['ttft']:.2f}", delta_cell, gb(free)]
+            if a.ctx:
+                rc = with_retries(lambda: run(a.url, a.key, m, filler + "\n\n" + task, a.tokens, a.temp, a.timeout),
+                                  a.retries, a.retry_wait, m)
+                cells.append(f"{rc['tg']:.1f}")
+        except Exception as e:
+            err = f"HTTP {e.code} (check server log)" if isinstance(e, urllib.error.HTTPError) else f"ERROR {e}"
+            cells = base_cells + [err] + [""] * (len(cols) - len(base_cells) - 1)
+        emit("| " + " | ".join(str(x) for x in cells) + " |")
+
+    latest.write("\n_Tuning hints (CPU box): decode reads the model's active bytes every token — smaller "
+                 "quants and small-active MoE are directly faster. **RAM free** is the real ceiling. "
+                 "Sanity-check: decode t/s × model GB = effective GB/s; dense models measure ~8-10 GB/s "
+                 "here (the practical wall), a model far below that is compute-bound (MoE/hybrid) → more "
+                 "`threads` may help it; `threads-batch = 4` may help prefill either way. With "
+                 "`--models-max 1`, mid-sweep RAM Δ is net-of-eviction. For spec/MTP models the draft "
+                 "acceptance % is shown in the spec column. ⚠ Absolute numbers swing ±20-30% with desktop "
+                 "load/thermals — only same-run rows are directly comparable._\n")
+    latest.close(); hist.close()
+    print(f"\nwrote {a.out} (+ appended {a.history})")
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,214 @@
+# llama.cpp model config — LAPTOP (lw) · CPU inference, NO discrete GPU
+# ═════════════════════════════════════════════════════════════════════════════
+# HARDWARE: Intel Core i7-5600U (Broadwell 2015, 2 cores / 4 threads, AVX2, no AVX-512)
+#           · iGPU HD Graphics 5500 (i915, not used for compute) · 15.49 GiB RAM
+#           · DDR3L-1600 dual-channel ≈ ~20 GB/s effective → THE bottleneck.
+#
+# This is the laptop analog of fl/.config/llamacpp/config.ini (the RX 7600 XT box), but on a
+# CPU-only machine almost every fl rule inverts:
+#
+#   • NO VRAM / GTT / freeze risk — no display-driving GPU to starve. The only ceiling is SYSTEM
+#     RAM. The `llamaserver` abbr uses `--models-max 1` (one model resident; a new request LRU-
+#     evicts the previous — safest on this 15 GB laptop, no co-residency swap). Raise to 2-3 only
+#     to keep small models hot together (E2B+4B+1.7B ≈ 6 GB fits).
+#   • Decode speed ≈ effective bandwidth ÷ active-bytes, so SMALL QUANTS are directly faster and
+#     MoE (few active params) is the single biggest trick. Clean bench 2026-07-25 (n-gpu-layers=0):
+#     dense models reach ~8-10 GB/s effective (decode t/s × model GB) ≈ the PRACTICAL DDR3L wall,
+#     i.e. dense decode IS bandwidth-bound as designed. (The earlier "core-starved ~5-7 GB/s"
+#     reading was the Vulkan-offload taint.) Granite (~2.6 GB/s eff) and the 30B (~3.5) sit far
+#     BELOW the wall → COMPUTE-bound: threads 3/4 might lift those two (dense won't gain).
+#   • `threads = 2` = PHYSICAL cores for the dense presets (at the wall — swept 2026-07-26, no
+#     gain). `threads = 3` on the compute-bound Granite/30B: +~5-7% same-run (t4 LOSES to HT
+#     contention on both — never promote it). `threads-batch = 4` EVERYWHERE: prefill is
+#     compute-bound on all models — benched +8% prefill / −9% TTFT on Granite, no decode cost
+#     (HT only bursts during prefill, so no sustained desktop pressure).
+#   • ⚠ RESPONSIVENESS: on 2 cores, generation pins the whole CPU → the desktop can freeze. The
+#     `llamaserver` abbr runs the server under `nice -n 19`, so inference YIELDS to interactive apps
+#     (still full speed when the desktop is idle). Still laggy? set `threads = 1` (≈half speed, always
+#     responsive).
+#   • `flash-attn = on` EVERYWHERE EXCEPT the two gemma-4 presets: benched ~+11% here (gen 3.7→4.1,
+#     prompt 1.7→1.9 t/s) so it's worth keeping — BUT this build (b10068) FAILS TO LOAD the gemma3n
+#     arch (E2B/E4B) with flash-attn, so those two stay OFF. (If another model also won't load with
+#     it, drop it there too.) No cache-type-k/v (quantized KV) on CPU: dequant overhead, no VRAM to
+#     reclaim → slower. KV stays f16.
+#   • ⚠ `n-gpu-layers = 0` REQUIRED in every preset — the pacman build has a VULKAN backend and
+#     by default AUTO-OFFLOADS layers to the HD 5500 iGPU (common_fit_params). Seen live
+#     2026-07-25: Granite/30B loads died with "ggml_vulkan ... ErrorOutOfDeviceMemory" (llama-cli
+#     even SIGSEGVs). The iGPU shares the same DDR3 and is slower than the cores — offload also
+#     silently poisons benchmarks ("omit it for CPU-only" was wrong on this build).
+#
+# Clean baseline 2026-07-25 17:41 (n-gpu-layers=0, threads=2, gen 256 — bench-history.md);
+#   decode noted per preset below, roster table in the README. Prefill is slow on 2 cores, so
+#   huge ctx is slow to FILL (not RAM-limited) — 4096 is a sane default; raise only if needed.
+#
+# Router:  llama-server --host 127.0.0.1 --port 11343 --cors-origins localhost --models-max 1 \
+#          (localhost-only: the laptop roams onto untrusted networks and there's no API key —
+#          unlike fl, nothing on the LAN targets this server. --cors-origins localhost blocks
+#          browser pages from calling the keyless API; non-browser clients unaffected — PR 25655)
+#          --models-preset ~/.config/llamacpp/config.ini   (abbr `llamaserver`; llama.cpp is
+#          pacman-installed → `llama-server` is on PATH)
+# This file stows to ~/.config/llamacpp/ (like fl); the GGUFs live in ~/software/models/.
+# Section names ARE the API model ids.
+# ═════════════════════════════════════════════════════════════════════════════
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Fast / small — the "instant" tier (quick Q&A, askllama, autocomplete)
+# ─────────────────────────────────────────────────────────────────────────────
+
+[gemma-4-E2B-it-UD-Q4_K_XL]
+# Gemma 3n E2B (~2B effective/active) — FASTEST here: benched 6.1 t/s (clean 2026-07-25). Add an mmproj for vision.
+model              = /home/anon/software/models/gemma-4-E2B-it-UD-Q4_K_XL.gguf
+# flash-attn OFF: gemma3n fails to load with it on this build (b10068).
+ctx-size           = 4096
+threads            = 2
+threads-batch      = 4
+n-gpu-layers       = 0
+jinja              = on
+temp               = 1.0         # Gemma default (1.0 / top-p 0.95 / top-k 64); drop to 0.7 for determinism
+top-p              = 0.95
+top-k              = 64
+sleep-idle-seconds = 30
+
+[Qwen3-1.7B]
+# Qwen3-1.7B, Q8_0 (~1.8 GB — Q8 is fine at this size). Benched 4.8 t/s (clean 2026-07-25) —
+# ~8.8 GB/s effective ≈ at the bandwidth wall; more threads won't lift this one.
+model              = /home/anon/software/models/Qwen3-1.7B-Q8_0.gguf
+ctx-size           = 8192        # tiny KV — can afford more ctx
+threads            = 2
+threads-batch      = 4
+n-gpu-layers       = 0
+flash-attn         = on
+jinja              = on
+temp               = 0.7
+top-p              = 0.8
+top-k              = 20
+min-p              = 0
+repeat-penalty     = 1.05
+sleep-idle-seconds = 30
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Daily drivers — 4B-class (best quality/speed balance; benched 2.1-3.9 t/s, clean 2026-07-25)
+# ─────────────────────────────────────────────────────────────────────────────
+
+[Qwen3-4B-Instruct-2507]
+# Qwen3-4B Instruct 2507, Q4_K_M (~2.5 GB). Benched 3.9 t/s (clean 2026-07-25). Best small
+# all-rounder; non-thinking, compact answers. The natural DAILY DRIVER — askllama points here.
+model              = /home/anon/software/models/Qwen3-4B-Instruct-2507.Q4_K_M.gguf
+ctx-size           = 4096
+threads            = 2
+threads-batch      = 4
+n-gpu-layers       = 0
+flash-attn         = on
+jinja              = on
+temp               = 0.7
+top-p              = 0.8
+top-k              = 20
+min-p              = 0
+repeat-penalty     = 1.05
+sleep-idle-seconds = 30
+
+[gemma-4-E4B-it-UD-Q4_K_XL]
+# Gemma 3n E4B (~4B active), QAT Q4_K_XL (~4-7 GB). Quality step up from E2B; add an mmproj for vision.
+# (File is the -qat- build — QAT keeps 4-bit quality; the id drops "qat" to match fl's convention.)
+model              = /home/anon/software/models/gemma-4-E4B-it-qat-UD-Q4_K_XL.gguf
+# flash-attn OFF: gemma3n fails to load with it on this build (b10068).
+# Benched 2.4 t/s (nodraft, clean 2026-07-26) — same speed as the far-smarter 30B MoE, so E4B's
+# niche is vision (mmproj). MTP draft TESTED AND REMOVED 2026-07-26: lost the final CPU-pinned
+# same-run A/Bs on both gemmas (E2B 4.0 vs 4.9 at 47% acc; E4B 2.3 vs 2.4 at 62% acc) — batched
+# verify is compute-bound on 2 cores; fl's +50% is a GPU result. The mtp-*.gguf drafts in
+# ~/software/models/ can be deleted.
+ctx-size           = 4096
+threads            = 2
+threads-batch      = 4
+n-gpu-layers       = 0
+jinja              = on
+temp               = 1.0
+top-p              = 0.95
+top-k              = 64
+sleep-idle-seconds = 30
+
+[Jan-v3-4b]
+# Jan v3 4B (Qwen3-4B-based, tuned for agentic/reasoning use). Same-run A/B vs its base
+# (2026-07-26): 2.6 vs 2.9 t/s — NOT anomalous, just ~10% slower (Q4_K_XL reads a bit more per
+# token than the base's Q4_K_M). The earlier 2.1-vs-3.9 reading was cross-run variance.
+model              = /home/anon/software/models/Jan-v3-4b-base-instruct-Q4_K_XL.gguf
+ctx-size           = 4096        # init had 256 — unusably small for reasoning; raised
+threads            = 2
+threads-batch      = 4
+n-gpu-layers       = 0
+flash-attn         = on
+jinja              = on
+temp               = 0.7
+top-p              = 0.95
+top-k              = 20
+sleep-idle-seconds = 30
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Coding
+# ─────────────────────────────────────────────────────────────────────────────
+
+[Qwen2.5-Coder-3B-Instruct]
+# Qwen2.5-Coder-3B Instruct, Q5_K_M (~2.3 GB, official Qwen GGUF). Small dedicated coder — benched 3.3 t/s (clean 2026-07-25).
+model              = /home/anon/software/models/qwen2.5-coder-3b-instruct-q5_k_m.gguf
+ctx-size           = 8192
+threads            = 2
+threads-batch      = 4
+n-gpu-layers       = 0
+flash-attn         = on
+jinja              = on
+temp               = 0.2         # low temp for code
+top-p              = 0.9
+sleep-idle-seconds = 30
+
+# ─────────────────────────────────────────────────────────────────────────────
+# ⭐ Quality when you can wait — 30B MoE, only 3B ACTIVE → ~30B quality at ~4B speed
+# ─────────────────────────────────────────────────────────────────────────────
+
+[Qwen3-30B-A3B-Instruct-2507-UD-IQ3_XXS]
+# Qwen3-30B-A3B Instruct 2507, UD-IQ3_XXS (~12 GB). Benched 1.9-2.7 t/s decode (cross-run
+# variance) / 4 t/s prefill / TTFT ~6.5 s — ~30B quality at E4B speed: THE quality pick when you
+# can wait. Compute-bound (~3.5 GB/s eff) → thread sweep 2026-07-26: t3 2.0 > t2 1.9 > t4 1.8
+# same-run → threads=3 promoted. (Earlier load failures were the Vulkan auto-offload OOM.)
+# ⚠ RAM: ~12 GB on 15.5 GB. With `--models-max 1` (the abbr default) it's always alone — no
+#   co-residency swap risk — but 12 GB + desktop is still tight, so keep ctx small and close the
+#   browser. If it swaps even alone, step the quant down to UD-IQ2_M (~10.5 GB).
+model              = /home/anon/software/models/Qwen3-30B-A3B-Instruct-2507-UD-IQ3_XXS.gguf
+ctx-size           = 4096
+threads            = 3
+threads-batch      = 4
+n-gpu-layers       = 0
+flash-attn         = on
+jinja              = on
+temp               = 0.7
+top-p              = 0.8
+top-k              = 20
+min-p              = 0
+sleep-idle-seconds = 30
+
+# ─────────────────────────────────────────────────────────────────────────────
+# ⭐ Faster "speed AND brains" — small-active MoE
+# ─────────────────────────────────────────────────────────────────────────────
+
+[Granite-4.0-H-Tiny]
+# IBM Granite 4.0 H Tiny — 7B total / ~1B ACTIVE (hybrid Mamba-2 + MoE). Benched 4.3-4.8 t/s
+# decode / 11-14 t/s prefill: faster than every 4B AND ~7B-class quality, tiny RAM Δ (mmap).
+# COMPUTE-bound (~2.6 GB/s eff) → thread sweep 2026-07-26: t3 4.8 > t2 4.5 > t4 4.3 same-run →
+# threads=3 promoted (t4 loses to HT contention). Revert to 2 if the desktop drags.
+# QUANT: Q4_K_M (~4.2 GB) — goal here is speed, and Q4_K_M is the sweet spot (Q5_K_M/Q6_K fine
+# too, RAM allows, small speed cost since decode reads only the ~1B active).
+# Needs a current llama.cpp (hybrid arch) — the pacman build should be fine.
+model              = /home/anon/software/models/granite-4.0-h-tiny-Q4_K_M.gguf
+ctx-size           = 8192
+threads            = 3
+threads-batch      = 4
+n-gpu-layers       = 0
+flash-attn         = on   # confirmed loading fine on the hybrid arch (2026-07-25)
+jinja              = on
+temp               = 0.7
+top-p              = 0.95
+top-k              = 20
+sleep-idle-seconds = 30
+
+# (Bench-rig history: the MTP A/B rigs and the thread-sweep rigs that lived here were deleted
+# 2026-07-26 after their verdicts — MTP rejected (E4B comment), threads=3 promoted to
+# Granite/30B, threads-batch=4 everywhere. Raw legs in bench-history.md.)
@@ -0,0 +1,259 @@
+{
+  "$schema": "https://opencode.ai/config.json",
+  "model": "llama.cpp/gemma-4-26B-A4B-it-UD-IQ4_XS",
+  "small_model": "llama.cpp/gemma-4-E4B-it-UD-Q8_K_XL",
+  "instructions": ["CLAUDE.md", "AGENTS.md"],
+  "share": "disabled",
+  "autoupdate": "notify",
+  "disabled_providers": [
+    "gemini",
+    "anthropic",
+    "groq",
+    "openrouter",
+    "opencode"
+  ],
+  "permission": {
+    "edit": "deny",
+    "webfetch": "ask",
+    "websearch": "ask",
+    "external_directory": "ask",
+    "read": {
+      "*": "allow",
+      "**/.env*": "deny",
+      "**/*.env": "deny",
+      "**/*.pem": "deny",
+      "**/*.key": "deny",
+      "**/*.p12": "deny",
+      "**/*.pfx": "deny",
+      "**/*.keystore": "deny",
+      "**/id_rsa*": "deny",
+      "**/.ssh/**": "deny",
+      "**/.aws/**": "deny",
+      "**/.npmrc": "deny",
+      "**/.netrc": "deny"
+    },
+    "bash": {
+      "*": "ask",
+
+      "ls *": "allow",
+      "cat *": "allow",
+      "grep *": "allow",
+      "rg *": "allow",
+      "find *": "allow",
+      "fd *": "allow",
+      "which *": "allow",
+      "file *": "allow",
+      "stat *": "allow",
+      "wc *": "allow",
+      "tree *": "allow",
+      "diff *": "allow",
+      "sort *": "allow",
+      "uniq *": "allow",
+      "echo *": "ask",
+      "pwd": "allow",
+      "date": "allow",
+      "whoami": "allow",
+      "jq *": "allow",
+      "tsc *": "allow",
+      "mkdir *": "ask",
+      "touch *": "ask",
+
+      "git *": "ask",
+      "git status": "allow",
+      "git status *": "allow",
+      "git diff": "allow",
+      "git diff *": "allow",
+      "git log": "allow",
+      "git log *": "allow",
+      "git show *": "allow",
+      "git branch": "allow",
+      "git remote -v": "allow",
+
+      "npm *": "ask",
+      "pnpm *": "ask",
+      "npx *": "ask",
+      "npm list": "allow",
+      "npm help": "allow",
+      "npm version": "allow",
+      "pnpm list": "allow",
+      "pnpm help": "allow",
+      "npx --version": "allow",
+
+      "curl *": "ask",
+      "wget *": "ask",
+
+      "rm *": "deny",
+      "rmdir *": "deny",
+      "dd *": "deny",
+      "mkfs*": "deny",
+      "shred *": "deny",
+      "chmod -R *": "deny",
+      "sudo *": "deny",
+      "doas *": "deny",
+
+      "npm install*": "deny",
+      "npm i": "deny",
+      "npm i *": "deny",
+      "npm ci*": "deny",
+      "pnpm install*": "deny",
+      "pnpm i": "deny",
+      "pnpm i *": "deny",
+      "pnpm add*": "deny",
+      "yarn add*": "deny",
+      "yarn install*": "deny",
+      "pip install*": "deny",
+      "pip3 install*": "deny",
+      "pipx install*": "deny",
+      "cargo install*": "deny",
+      "go install*": "deny",
+      "gem install*": "deny",
+      "brew install*": "deny",
+
+      "pacman *": "deny",
+      "yay *": "deny",
+      "apt *": "deny",
+      "apt-get *": "deny",
+      "dnf *": "deny",
+
+      "cat .env*": "deny",
+      "cat *.env": "deny",
+      "cat *.pem": "deny",
+      "cat *.key": "deny",
+      "cat *id_rsa*": "deny",
+      "cat *secret*": "deny",
+      "cat *credential*": "deny",
+      "cat ~/.ssh/*": "deny",
+      "cat ~/.aws/*": "deny",
+      "grep .env*": "deny",
+      "grep * .env*": "deny"
+    }
+  },
+  "provider": {
+    "lwcpp": {
+      "npm": "@ai-sdk/openai-compatible",
+      "name": "llamacpp (lw local)",
+      "options": {
+        "baseURL": "http://127.0.0.1:11343/v1"
+      },
+      "models": {
+        "gemma-4-E2B-it-UD-Q4_K_XL": { "name": "Gemma 4 E2B · 4k — fastest on lw (~6 t/s)", "limit": { "context": 4096, "output": 1024 } },
+        "Granite-4.0-H-Tiny": { "name": "Granite 4.0 H Tiny · 8k — speed AND brains (~4.8 t/s)", "limit": { "context": 8192, "output": 2048 } },
+        "Qwen3-1.7B": { "name": "Qwen3 1.7B · 8k — thinking, snappy (~4.8 t/s)", "limit": { "context": 8192, "output": 2048 } },
+        "Qwen3-4B-Instruct-2507": { "name": "Qwen3 4B 2507 · 4k — daily driver (~3.9 t/s)", "limit": { "context": 4096, "output": 1024 } },
+        "Qwen2.5-Coder-3B-Instruct": { "name": "Qwen2.5 Coder 3B · 8k — small coder (~3.3 t/s)", "limit": { "context": 8192, "output": 2048 } },
+        "Qwen3-30B-A3B-Instruct-2507-UD-IQ3_XXS": { "name": "Qwen3 30B A3B · 4k — quality when you can wait (~2.7 t/s)", "limit": { "context": 4096, "output": 1024 } },
+        "Jan-v3-4b": { "name": "Jan v3 4B · 4k — agentic tune (~2.6 t/s)", "limit": { "context": 4096, "output": 1024 } },
+        "gemma-4-E4B-it-UD-Q4_K_XL": { "name": "Gemma 4 E4B · 4k — quality; vision once mmproj added (~2.5 t/s)", "limit": { "context": 4096, "output": 1024 } }
+      }
+    },
+    "duskadiy": {
+      "npm": "@ai-sdk/openai-compatible",
+      "name": "DuskaDIY",
+      "options": {
+        "baseURL": "https://llm.duskadiy.com/api/v1"
+      },
+      "models": {
+        "Qwen3-Coder-30B-Instruct-UD-Q3_K_XL": { "name": "Qwen3 Coder 30B · 32k — main agent coder (remote)", "limit": { "context": 32768, "output": 4096 } },
+        "Qwen3-Coder-Next-UD-IQ3_XXS": { "name": "Qwen3 Coder Next 80B · 128k — long sessions (remote)", "limit": { "context": 131072, "output": 16384 } },
+        "Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS": { "name": "Qwen3.6 35B · 24k · vision — daily driver (remote)", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
+        "Qwen3.6-35B-A3B-Thinking": { "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems (remote)", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
+        "Qwen3.5-9B-UD-Q6_K_XL": { "name": "Qwen3.5 9B · 32k · vision — quick tasks (remote)", "attachment": true, "limit": { "context": 32768, "output": 8192 } },
+        "gemma-4-26B-A4B-it-UD-IQ4_XS": { "name": "Gemma 4 26B · 24k · vision — quality generalist (remote)", "attachment": true, "limit": { "context": 24576, "output": 4096 } },
+        "gemma-4-E4B-it-UD-Q8_K_XL": { "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs (remote)", "attachment": true, "limit": { "context": 65536, "output": 16384 } },
+        "GLM-4.7-Flash-UD-Q4_K_XL": { "name": "GLM-4.7 Flash · 24k — quality coder (remote)", "limit": { "context": 24576, "output": 8192 } },
+        "gpt-oss-20b": { "name": "gpt-oss 20B · 64k — fast reasoning + tools (remote)", "limit": { "context": 65536, "output": 8192 } },
+        "gpt-oss-20b-low": { "name": "gpt-oss 20B low · 64k — snappy answers (remote)", "limit": { "context": 65536, "output": 8192 } }
+      }
+    },
+    "llama.cpp": {
+      "npm": "@ai-sdk/openai-compatible",
+      "name": "llamacpp",
+      "options": {
+        "baseURL": "http://192.168.0.204:11343/v1"
+      },
+      "models": {
+        "Qwen3-Coder-Next-UD-IQ3_XXS": { "name": "Qwen3 Coder Next 80B · 128k — long sessions", "limit": { "context": 131072, "output": 16384 } },
+        "Qwen3-Coder-30B-Instruct-UD-Q3_K_XL": { "name": "Qwen3 Coder 30B · 32k — main agent coder", "limit": { "context": 32768, "output": 4096 } },
+        "Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS": { "name": "Qwen3.6 35B · 24k · vision — daily driver", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
+        "Qwen3.6-35B-A3B-Thinking": { "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
+        "Qwen3.5-9B-UD-Q6_K_XL": { "name": "Qwen3.5 9B · 32k · vision — quick tasks", "attachment": true, "limit": { "context": 32768, "output": 8192 } },
+        "gemma-4-26B-A4B-it-UD-IQ4_XS": { "name": "Gemma 4 26B · 24k · vision — quality generalist", "attachment": true, "limit": { "context": 24576, "output": 4096 } },
+        "gemma-4-E4B-it-UD-Q8_K_XL": { "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs", "attachment": true, "limit": { "context": 65536, "output": 16384 } },
+        "GLM-4.7-Flash-UD-Q4_K_XL": { "name": "GLM-4.7 Flash · 24k — quality coder", "limit": { "context": 24576, "output": 8192 } },
+        "gpt-oss-20b": { "name": "gpt-oss 20B · 64k — fast reasoning + tools", "limit": { "context": 65536, "output": 8192 } },
+        "gpt-oss-20b-low": { "name": "gpt-oss 20B low · 64k — snappy answers", "limit": { "context": 65536, "output": 8192 } }
+      }
+    }
+  },
+  "agent": {
+    "plan": {
+      "mode": "primary",
+      "model": "llama.cpp/gemma-4-26B-A4B-it-UD-IQ4_XS",
+      "permission": {
+        "edit": "deny"
+      }
+    },
+    "build": {
+      "mode": "primary",
+      "model": "llama.cpp/gemma-4-26B-A4B-it-UD-IQ4_XS",
+      "permission": {
+        "edit": {
+          "*": "allow",
+          "**/.env*": "deny",
+          "**/*.env": "deny",
+          "**/*.pem": "deny",
+          "**/*.key": "deny",
+          "**/id_rsa*": "deny",
+          "**/.ssh/**": "deny",
+          "**/.aws/**": "deny",
+          "**/.git/**": "deny",
+          "**/.npmrc": "deny",
+          "**/.netrc": "deny"
+        }
+      }
+    },
+    "code-reviewer": {
+      "description": "Reviews code for best practices and potential issues",
+      "prompt": "You are a code reviewer. Focus on security, performance, and maintainability.",
+      "mode": "subagent",
+      "model": "llama.cpp/GLM-4.7-Flash-UD-Q4_K_XL",
+      "permission": {
+        "edit": "deny"
+      }
+    },
+    "test-runner": {
+      "description": "Runs and interprets test suites",
+      "prompt": "You are a test runner. Focus on identifying failing tests and providing clear reproduction steps.",
+      "mode": "subagent",
+      "model": "llama.cpp/GLM-4.7-Flash-UD-Q4_K_XL",
+      "permission": {
+        "edit": "deny"
+      }
+    },
+    "doc-writer": {
+      "description": "Maintains project documentation and docstrings",
+      "prompt": "You are a technical writer. Focus on clarity, accuracy, and keeping documentation in sync with code changes.",
+      "mode": "subagent",
+      "model": "llama.cpp/gemma-4-26B-A4B-it-UD-IQ4_XS",
+      "permission": {
+        "edit": {
+          "*": "deny",
+          "**/*.md": "allow",
+          "**/*.mdx": "allow",
+          "**/*.markdown": "allow",
+          "**/*.txt": "allow",
+          "**/*.rst": "allow"
+        }
+      }
+    },
+    "security-auditor": {
+      "description": "Scans for security vulnerabilities and hardcoded secrets",
+      "prompt": "You are a security auditor. Focus on OWASP principles, secret detection, and dependency vulnerabilities.",
+      "mode": "subagent",
+      "model": "llama.cpp/GLM-4.7-Flash-UD-Q4_K_XL",
+      "permission": {
+        "edit": "deny"
+      }
+    }
+  }
+}
@@ -0,0 +1,40 @@
+# F13 is a real keycode that no app binds. We declare it a virtual modifier so it can
+# prefix the nav-layer chords below, then remap RightAlt -> F13 (see modmap). This mirrors
+# the wm/fl evremap setup: RightAlt only ever emits F13, so a bare tap/hold never triggers
+# app menu bars and no stray Alt leaks on unmapped chords.
+virtual_modifiers:
+  - F13
+
+modmap:
+  - name: CapsLock to LeftCtrl/Esc
+    remap:
+      CapsLock:
+        held: leftctrl
+        alone: esc
+        alone_timeout_millis: 150
+
+  - name: RightAlt to F13 (nav-layer modifier; Left Alt untouched)
+    remap:
+      Alt_R: F13
+
+keymap:
+  - name: RightAlt(F13) vim-nav layer — mirrors wm/fl evremap
+    remap:
+      F13-h: left
+      F13-j: down
+      F13-k: up
+      F13-l: right
+      F13-u: pageup
+      F13-d: pagedown
+      F13-b: C-left      # word left  (Ctrl+Left)
+      F13-w: C-right     # word right (Ctrl+Right)
+
+# keymap:
+#   - name: Freetube remap
+#     application:
+#       only: FreeTube
+#     remap:
+#       Alt-h: Left
+#       Alt-l: Right
+#       Alt-j: Down
+#       Alt-k: Up