coja
1e39e0cb4b
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.
83 lines
3.8 KiB
YAML
83 lines
3.8 KiB
YAML
# 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
|