Three converging fixes so the chat agent + external Codex/Claude skills can actually debug a crashed serve instead of staring at a post-crash neofetch banner:
* Serves now `tee` to /tmp/odysseus-tmux/SESSION.log on the host running them. Runner saves fds 3/4 before the tee and restores them right before `exec ${SHELL}`, so the post-crash interactive zsh banner does NOT pollute the log file.
* `tail_serve_output` (chat agent) and `/api/codex/cookbook/output/{sid}` (Codex+Claude skills) both prefer the persistent log file over the tmux pane. Pane is fallback for sessions predating the tee runner. Default tail bumped 150 -> 400.
* `list_served_models` "recent log" snippet seeks to the Traceback line instead of showing the last 6 lines (which was always the bash prompt).
Cookbook auto-adoption sweep on `/api/cookbook/tasks/status`: every 20s (rate-limited) the cookbook SSHes each configured server, finds `serve-*` / `cookbook-*` tmux sessions running an actual model process (vllm/python/llama-server/etc., filtered via `pane_current_command`), and writes them into state.tasks. So when the agent falls back to raw ssh+tmux, the session appears in the Cookbook UI on the next poll.
`serve_model` error path now reads `data["detail"]` in addition to `data["error"]` so the FastAPI HTTPException message ("Invalid characters in cmd") actually reaches the agent instead of being swallowed as a generic "Serve failed". Tool description updated to warn against `cd …`/`source …`/`&&` prefixes.
Intent-without-action supervisor in agent_loop: when the model writes "Let me tail the output" / "I'll check the logs" / "Let me investigate" and ends the turn without emitting a tool call, the loop injects a sharp system nudge ("You said you would X — DO IT NOW") and continues. Capped at 2 nudges per chat so a model that genuinely cannot use the tool does not pin the loop.
Codex/Claude skill parity: adds `/cookbook/cached`, `/cookbook/presets`, `/cookbook/preset/{name}`, `/cookbook/adopt` so external agents have the same surface as the chat agent. SKILL.md docs + odysseus_api.py wrapper updated for both bundles.
`adopt_served_model` promoted to the always-on tool set so the agent has a documented fallback when serve_model rejects a cmd.
Also various cookbook UI tweaks accumulated alongside the above (cookbook.js, cookbookRunning.js, cookbookServe.js, cookbook-diagnosis.js, settings.js, style.css).
Backend (services/hwfit + routes):
- VRAM column sort now shows global highest first (was special-cased to
ascending then truncated top-N, which made "highest VRAM" mathematically
unreachable). Every column path uses reverse=True for the truncation.
- Hardware probe cache TTL 30min -> 24h so changing filters doesn't keep
re-probing the rig during a session; Rescan button still forces fresh.
- Multi-GPU rigs filter GGUF Q*/IQ quants (vLLM/SGLang can't serve them);
default non-prequantized to BF16 on 2+ GPUs.
- AWQ / AWQ-8bit / GPTQ-8bit get a -1.0 quality penalty so FP8 wins ties.
- Version-aware tiebreaker (parse Mn.n / Vn) — MiniMax-M2.7 ranks above M2.5.
- hf_models.json: zai-org/GLM-5.1 added; zai-org/GLM-5 quantization flipped
Q4_K_M -> BF16. DeepSeek-V4-Flash / -Pro + their -Base variants registered
with new FP4-MoE-Mixed / FP8-Mixed quant keys (calibrated BPP from the
actual 156 GB / 284 GB disk footprints).
- New FP4-MoE-Mixed + FP8-Mixed entries in QUANT_BPP / QUANT_SPEED_MULT /
QUANT_QUALITY_PENALTY / QUANT_BYTES_PER_PARAM / PREQUANTIZED_PREFIXES.
Frontend — Scan/Download:
- Engine + Quant swapped in the toolbar; Quant defaults to "All".
- Ctx (range slider) ported from origin/main: 8k/16k/32k/50k/128k/Max. Drag
re-sorts by vram ascending (smallest fitting first); back to Max → score.
- Ctx slider rail now visible — was background:transparent in a duplicate
later-cascade rule. Hardcoded grey + !important.
- Search input moved to the far right of the toolbar.
- Type/Standard default; "Context" not uppercased; Search placeholder dimmed.
- Engine "?" + Quant "?" inline help chips inside their dropdown boxes.
- Fit-column dot toggles fit-only filter; un-toggling re-sorts by VRAM desc.
- Quant column truncates to 9 chars + ellipsis ("FP4-MoE-M..."), full in
tooltip. Smart title-suffix strips the parts already in the repo name
(QuantTrio/MiniMax-M2-AWQ + quant AWQ-4bit -> just "(4bit)").
- Conditional warning for safetensors models on non-GPU rigs only.
- Dependency Install / Installed / Installed▾ / N/A all 75.85px wide.
- Rebuild llama.cpp moved into the llama_cpp dep row, styled as a tag.
- Foldable Download admin-card (h2 chevron); line under h2 only when folded.
- HF token save gets a green ✓ + "Saved" flash.
- Cached scan no longer counts stalled rows as downloaded.
- Footer: "Request it →" link with GitHub mark to the public discussion
(#1962) for model-add requests.
Frontend — Running tab:
- Strict download-finish check (DOWNLOAD_OK or /snapshots/, not bare
"Download complete"). True overall % for multi-shard downloads:
((N-1)+frac)/total instead of hf_transfer's per-shard aggregate.
- ETA in the uptime ticker: "downloading: 12m 34s · ETA 1h 23m".
- Clear button kills the tmux session too; if the output still shows a
live shard line, the pill is hidden + relabels as "reconnect" + revives
on click.
- Self-heal: on cookbook open AND every bg-monitor cycle (10s, throttled
to 8s), scan persisted done/error/crashed downloads and probe their
tmux session — if alive, flip status back to running and reattach.
- Per-launch zombie probe: clicking Download on a model whose persisted
state is done but tmux is still alive revives the existing task and
refuses to start a duplicate.
- Pre-launch GPU probe: vllm / sglang / diffusers serve check
/api/cookbook/gpus first; warns + confirms if no GPU is visible.
- Server-side state guard: rejects "done" POSTs for downloads lacking
DOWNLOAD_OK / DOWNLOAD_FAILED / /snapshots/ when the last-mentioned
shard is N<total — stale tabs can't poison persisted state any more.
- Running count includes tasks whose output looks active even if persisted
status got stuck. Dir text on the running row, font matched to uptime.
Serve panel:
- Ctx text input always resets to model max on open (default 20000 when
metadata is missing).
- Max Seqs default 8 -> 4. KV Cache dtype select 32px tall.
- Lightning icon on Launch (same as Action toggle).
- Diagnosis card simplified (no fold/copy/dismiss), suggestion font
matches body; action buttons get icons on the left (Retry/Copy/Edit/
Install/Kill/Switch/etc.).
- Incomplete-download serve warning when model status is
downloading / stalled / has_incomplete.
- MTP "?" tooltip ("supported on a few model families … up to ~3× faster").
Backend (services/hwfit + routes):
- rank_models picks visible set by REQUESTED column, not always score —
sorting by Param now shows highest-param models PERIOD (incl. too_tight).
- New fit_only param. Multi-GPU rigs filter GGUF Q*/IQ quants (vLLM/SGLang
cannot serve them); default non-prequantized to BF16 on 2+ GPUs.
- AWQ / GPTQ-8bit get a -1.0 quality penalty (was 0.0, tied with FP8), so
FP8 wins when both fit.
- Version-aware tiebreaker (parse Mn.n / Vn) — MiniMax-M2.7 ranks above
M2.5 on equal composite score; >=100B integers not misread as versions.
- /api/cookbook/hf-latest no longer drops models without an "NB" pattern in
the repo id (MiniMax-M2.7, DeepSeek-V4-Pro etc. were silently filtered).
- Cached-model scan: atexit flushes models JSON even if the script is
killed mid-walk; each scan_dir wrapped in try/except; timeout 60s -> 180s.
- KB granularity for sub-MB sizes (was "0 MB" for 12 KB shells). New
"stalled" status for shells <1 MB with no .incomplete files.
- /api/cookbook/state POST guard: rejects "done" download tasks lacking
DOWNLOAD_OK / DOWNLOAD_FAILED / /snapshots/ when the last-mentioned
shard is N<total — stops stale tabs from poisoning persisted state.
- hf_models.json: add zai-org/GLM-5.1; flip zai-org/GLM-5 quantization
Q4_K_M -> BF16 (it is the native base, not a quant).
Frontend (static/js):
- Scan/Download toolbar: quant defaults to All; ctx slider (8k/16k/32k/
50k/128k/Max) ported from origin/main with sort=fit on drag, sort=score
on Max. GPU toggle commits _activeCount to maxGpu on initial render. Fit
column header tagged with active budget (RAM / GPU / N GPU).
- Foldable Download admin-card: the Download h2 is the chevron trigger;
state persists in localStorage.
- Download card surfaces destination dir (Dir: <path>). Same dir on running
task row, font/color matched to uptime (9px Fira Code muted, opacity .4).
- Serve panel ctx text input always resets to model max on open. Sub-MB
cached models show with red "download stalled" badge.
- Bulk-select Cancel + Delete reset the Select button label on exit.
- Cookbook running: false-finished bug fixed — DOWNLOAD_OK or /snapshots/
required; bare "Download complete" no longer marks the task done after
the first config file. Clear button now sends tmux kill-session too.
True overall % for multi-shard downloads: ((N-1)+frac)/total instead of
hf_transfer per-shard aggregate.
- Diagnosis card simplified: removed fold toggle, copy button, dismiss X.
Suggestion font matches message body (12px).
- HF token field flashes green check + "Saved" on save.
- Cached scan no longer counts stalled rows as downloaded in Scan/Download.
CSS:
- dep Install button width pinned to 76px to match Installed split.
- task-sub row +1px; task-status badge gets margin-right 8px.
- Ctx slider styled like gallery editor sliders (thin pill rail, red thumb).
- Bulk-select cancel button top -3px -> -5px.
The serve bootstrap builds llama-server from source only when it is missing
from PATH, so a host that first compiled CPU-only (no nvcc present at build
time) reuses that CPU-only binary on every later serve and never gets a GPU
build, even after a CUDA/ROCm toolkit is installed. There was no UI lever to
force a rebuild.
Adds a 'Rebuild llama.cpp' button to the Cookbook Dependencies tab. It clears
the cached ~/bin/llama-server symlink and ~/llama.cpp/build directory (locally
or on the selected remote server) so the next serve recompiles and picks up
CUDA/HIP if a toolchain is now present. It installs and downloads nothing.
- routes/cookbook_helpers.py: _llama_cpp_rebuild_cmd() (single source of truth)
- routes/shell_routes.py: POST /api/cookbook/rebuild-engine (admin-only, reuses
the existing SSH plumbing for remote hosts)
- static/js/cookbook.js: header button + handler honoring the deps server selector
- tests: cover the command shape and a clean run on a fresh HOME
Motivated by #831 (RTX 4070 user stuck on a CPU-only build with no way to
re-trigger the build).
Co-authored-by: ghreprimand <203024559+ghreprimand@users.noreply.github.com>
A CPU-only llama.cpp serve config still emitted --flash-attn on and exported
GGML_CUDA_ENABLE_UNIFIED_MEMORY=1 (independent toggles, often left on by an Auto
profile), so the command mixed "zero GPU layers" with CUDA/flash-attn and failed
to start (issue #1291). Gate both on a _cpuOnly check (ngl == 0). GPU serving is
unchanged — the gate only affects the ngl=0 path.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Fix Cookbook dependency install completion state
Mark Cookbook dependency installs as complete when the background runner
exits successfully, even when HuggingFace-specific download markers are
absent.
* Add focused regression coverage for cookbook dependency completion.
Keep the fix narrowly scoped while carrying env_path through dependency tasks and locking the completion reconciliation behavior with targeted tests.
* Cookbook: Engine filter + intelligent hardware-computed serve profiles
Two related Cookbook serving improvements for accurate, hardware-aware model
serving (especially on consumer GPUs that can only run GGUF/llama.cpp).
Engine filter
- New "Engine" dropdown (All / llama.cpp / vLLM / SGLang) beside the quant
picker. Pure client-side view filter over the fetched list via the same
_detectBackend() the serve commands use, so what you filter to is exactly what
would launch. Re-renders from cache (no refetch). Empty-state message + the
instant-cache-paint path account for it too.
Intelligent serve profiles (Quality / Balanced / Speed)
- services/hwfit/profiles.py: compute_serve_profiles() turns detected VRAM +
model size into concrete llama.cpp flags (n_gpu_layers, n_cpu_moe, cache-type,
context). Encodes the by-hand tuning: a too-big MoE offloads experts to CPU
instead of failing; a model that fits stays fully on GPU; quant tracks profile
intent; vision models keep image-encoder headroom. Reuses models.py VRAM math
so filtering and serving agree on what fits. Pure/deterministic (no t/s claims
— partial-offload speed isn't reliably predictable; fit is what's computed).
- /api/hwfit/profiles endpoint returns the profiles + the model's trained
context limit, with loose name matching (strips org/ prefix, -GGUF suffix,
quant tag) so a local GGUF folder name resolves to its catalog entry.
- _buildServeCmd (llama.cpp) now emits --n-cpu-moe / --flash-attn /
--cache-type-k/v when set, with llama-cpp-python fallback equivalents. It
previously only set -ngl/-c, which is why it OOM'd or ran slow.
- Serve panel: profile chips that fill the fields on click, plus CPU-MoE / KV
Cache / Flash Attn fields. Context is clamped to the model's trained limit
(and an absolute 1M sanity ceiling) on type/blur/profile-load and at launch —
fixes a crash where a stale 256k/16M preset + quantized KV cache caused an
amdgpu ErrorDeviceLost.
Tests: tests/test_serve_profiles.py (7) — offload vs full-GPU fit, never exceed
VRAM, context cap, launchable flags, vision headroom, no-GPU empty.
Checks: py_compile + node --check pass; pytest test_serve_profiles + test_hwfit_amd
green; verified live on an RDNA4 box (gfx1200) — Balanced lands ~ncm18 q4 128k,
matching hand-tuning.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Cookbook: make column-header sorting discoverable (incl. Newest)
Sorting in Cookbook is via clickable column headers (pewds' design), but the
headers had no visual cue that they're interactive — so sorting in general, and
the Newest sort on the Model header specifically, was undiscoverable.
- Style sortable headers as interactive: pointer cursor, hover underline, and
the active sort column bolded/highlighted. There was no CSS for
.hwfit-sortable / .hwfit-sort-active at all; this helps every existing sort,
not just Newest.
- The Model column header sorts by release_date (newest first), reusing the
existing header-click sort wiring and the "newest" SORT_KEY.
No new sort control — uses the existing column-header paradigm.
Checks: node --check passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Cookbook serve profiles: keep the on-disk file's quant fixed (don't propose Q6/Q2)
In the Serve tab the model is a specific GGUF file already on disk, so its quant
can't change — but the profiles were suggesting "Quality · Q6_K" / "Speed · Q2_K"
as if you could re-quantize it. That's meaningless when serving a fixed file.
- compute_serve_profiles gains serve_weights_gb / serve_quant. When set (SERVE
mode), the quant is locked to the file's and profiles differ only in the real
serving knobs — n_cpu_moe, KV-cache type, context. _weights_gb / _cpu_moe_for_budget
use the file's actual size instead of a quant-derived estimate. DOWNLOAD mode
(no override) still varies the quant to show download options.
- /api/hwfit/profiles accepts serve_weights_gb & serve_quant.
- The Serve panel parses the file's size (from m.size "20.6 GB") and quant (from
the repo/file name) and passes them, so profiles match what's actually served.
Result for a 20.6 GB Q4_K_M file: all three profiles stay Q4_K_M and differ by
KV/ctx/offload (Quality q8 KV 128k ncm21, Balanced q4 128k ncm17, Speed q4 32k
ncm15) — no nonsensical quant changes.
Tests: test_serve_mode_keeps_fixed_quant. Full serve-profile suite green (9).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Cookbook serve: Vision toggle (auto-find mmproj) + live VRAM/RAM-spillover monitor
Two serve-panel additions:
1. **Vision toggle.** A "Vision" checkbox that serves the model with its
multimodal projector so it can read images. The mmproj path is resolved at
runtime (find mmproj-*.gguf next to the model), so dropping an mmproj file in
the model folder makes the toggle just work; `--mmproj … --image-max-tokens
1024` (native) / `--clip_model_path` (llama-cpp-python) only when on + found.
2. **Live GPU-memory monitor.** A readout that polls /api/cookbook/gpus every 4s
while the panel is open and shows VRAM used/total/%, free, and — crucially on
a discrete card — **RAM spillover** (AMD gtt_used_mb), with a plain-language
health hint: green/healthy, amber/tight, red/"spilled to RAM — slow (raise
CPU MoE or lower context)". Surfaces gtt_used_mb from the gpus endpoint
(previously read for total only and discarded for 'used').
Lets you see at a glance whether a config fits VRAM (fast) or is paging to system
RAM over PCIe (slow) instead of guessing.
Checks: node --check + py_compile pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The model-name detector treated every Qwen model as a Qwen3, falling
into the qwen3_xml parser:
if (n.includes('qwen3') && n.includes('coder')) return 'qwen3_coder';
if (n.includes('qwen')) return 'qwen3_xml'; // catches qwen2.5 too
qwen3_xml is the parser for Qwen3 reasoning/instruct models. Qwen2.5
(and Qwen2, Qwen1.5) ship with hermes-style tool calling, so the
qwen3_xml parser never recognises their tool calls — they leak through
as plain text in the assistant reply and the agent silently fails to
execute anything.
Reproduces with:
vllm serve Qwen/Qwen2.5-Coder-14B-Instruct-AWQ ... \
--enable-auto-tool-choice --tool-call-parser qwen3_xml
→ ask the agent to call any tool → JSON shows up in chat, no call runs.
Fix the ordering:
qwen3 + coder → qwen3_coder
qwen3 → qwen3_xml
qwen → hermes (Qwen2.5 / Qwen2 / Qwen1.5)
Verified against the model matrix:
Qwen2.5-Coder-14B-Instruct-AWQ → hermes
Qwen2.5-7B-Instruct → hermes
Qwen3-8B → qwen3_xml
Qwen3-32B → qwen3_xml
Qwen3-Coder-30B-A3B → qwen3_coder
Qwen2-72B-Instruct → hermes
Qwen1.5-7B-Chat → hermes
The Cookbook Scan/Download (hwfit) table gave the Fit column key:'score', so
clicking the Fit header sorted by score instead of by fit. Give the Fit column
its own 'fit' sort key, add a matching option to the #hwfit-sort select, and
rank fit_level (perfect > good > marginal > too_tight > no_fit) in the
client-side sort. Default puts the best fit first; clicking again reverses it.
Score still sorts by score.
Closes#842
* fix: show docker as N/A inside the container
* test: cover in-container docker detection
* fix: make the N/A dependency chip legible
* refactor: make remote docker applicability explicit and tested
* Add Apple Silicon (Metal) GPU detection and unified-memory fit tuning
hardware.py detects Apple Silicon locally and over SSH, reporting
backend=metal, the chip name, and a RAM-scaled fraction of unified
memory as the usable GPU budget. fit.py gains an M1-M4 memory-bandwidth
table for realistic tok/s and drops vLLM-only formats (AWQ/GPTQ/FP8)
that can't be served on Metal.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
(cherry picked from commit 32ac81dbc680361463a088dae867d555d5a79c3b)
* Generate macOS/Metal serve commands and surface the Metal GPU
cookbook_routes.py adds a macOS serve path (Ollama, Metal-aware
llama.cpp build using `sysctl hw.ncpu` instead of `nproc`, and a clear
error if vLLM is attempted). The frontend defaults Metal serving to
llama.cpp and offers llama.cpp/Ollama instead of vLLM/SGLang. The
odysseus-cookbook CLI's `gpus` command reports the Metal GPU via
sysctl/vm_stat.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
(cherry picked from commit 4ba01ce25d256ae032029898f361c824a34fcd4b)
* Add launchd LaunchAgent for macOS (systemd equivalent)
com.odysseus.ui.plist + install-service-macos.sh run Odysseus at login
and restart on crash, the macOS counterpart to odysseus-ui.service. The
installer auto-fills paths from the venv, so there's no hand-editing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
(cherry picked from commit 3d4b6b2c7b8b31af32201ed278115df9a559dea9)
* Document macOS install (brew, Ollama, AirPlay port, launchd)
README + setup.py cover the Homebrew / Apple Silicon path: brew install
python@3.11 tmux ollama, Metal serving via Ollama/llama.cpp, the launchd
service, and the macOS AirPlay Receiver conflict on ports 7000/5000.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
(cherry picked from commit 8dc9a3578a1726f070ed9f75c0958ae291a6d966)
* Add downloadable macOS launcher app builder
build-macos-app.sh generates dist/Odysseus.app and a drag-to-Applications
dist/Odysseus.dmg. The app starts the local server from this repo's venv and
opens the UI in a chrome-less app window (Chromium --app mode, falling back to
the default browser). It's a launcher wrapper — it drives the venv rather than
bundling Python — so the install path is baked in at build time.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
(cherry picked from commit 7927940c3810ee34640803b198d334a6ac93474d)
* Harden macOS Cookbook support: hide MLX, fix Metal build cache
Builds on the adopted PR #213 macOS/Metal work with two fixes and tests:
- fit.py: always drop MLX-quantized models. Odysseus only generates serve
commands for llama.cpp/Ollama (Metal) and vLLM/SGLang (CUDA); MLX needs the
mlx_lm runtime and the catalog's MLX repos ship no GGUF alternative, so they
were surfaced on Apple Silicon but could never be served.
- cookbook_routes.py (macOS branch only): `rm -rf build` before configure so a
poisoned CMakeCache from a prior failed CUDA attempt can't make every later
build fail; explicit -DCMAKE_BUILD_TYPE=Release; a clear "brew install cmake"
hint if cmake is missing. Linux/CUDA path unchanged.
- tests/test_hwfit_macos.py: MLX hidden on metal, MLX still hidden on CUDA
(regression guard), Metal detection on Apple Silicon, and skipped on
Linux/Intel (proves non-macOS detection is untouched).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Propagate unified_memory flag and document macOS GPU/Docker caveat
- hardware.py: detect_system now carries the unified_memory flag from GPU
detection into the system dict (it was set by _detect_apple_silicon / AMD-APU
detection but dropped during result assembly, so the API always reported
null). Lets callers distinguish unified from discrete VRAM.
- README: prominent warning that Docker on Apple Silicon can't reach the Metal
GPU (runs a Linux VM) — Cookbook must run natively for GPU serving; fix stale
text that said Cookbook recommends MLX models (now hidden as unservable).
- test: detect_system propagates unified_memory.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Put Odysseus's venv bin on PATH for cookbook runners
Native (non-Docker) installs run from a virtualenv whose bin holds the `hf` CLI
and `python3` the cookbook download/serve tmux scripts shell out to. Those
scripts start in a fresh login shell with the venv NOT activated, so on a native
macOS install `hf download` failed with "hf: command not found" — and the
`pip --user` self-heal missed because macOS has no bare `pip` command.
- cookbook_helpers.py: _local_tooling_path_export() — pure helper returning a
PATH export for the running interpreter's bin dir (escaped for double quotes).
- cookbook_routes.py: download + serve runners prepend that dir on local runs
(gated off SSH/Windows); swap the `pip` install fallbacks to `python3 -m pip`.
- tests: helper output for normal and spaced paths.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Document macOS llama.cpp serving prerequisites
Clarify the two serving paths on Apple Silicon: the recommended zero-build
route (brew install llama.cpp ships a Metal llama-server Cookbook finds on PATH),
and the from-source fallback, which requires cmake + Xcode Command Line Tools.
Without those the build is skipped and serving silently degrades to a slow CPU
build, so new users now know to install them (or use the prebuilt) up front.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Recommend only GGUF-servable models on Metal
Apple Silicon's only serving engines are llama.cpp and Ollama, both GGUF-only
(vLLM/SGLang are CUDA/ROCm and don't run on macOS). The catalog tags raw
safetensors repos with a default Q4_K_M quant, so the fit-ranking was
recommending ~397/501 models that have no GGUF and fail to serve on Metal with
"No GGUF found" (e.g. microsoft/Phi-mini-MoE-instruct).
Drop any model without a real GGUF (is_gguf/gguf_sources) on Apple Silicon —
subsumes the previous AWQ/GPTQ/FP8 special-case into one rule. On CUDA these
stay visible since vLLM serves safetensors directly. Metal recommendations go
501 -> 104, all actually servable.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Remove macOS launchd LaunchAgent (cherry-picked extra)
Drop the launchd service from the PR #213 cherry-picks: the
install-service-macos.sh installer, the com.odysseus.ui.plist template, and the
README section documenting them. Tangential to the core Cookbook/Metal support
and not wanted. The build-macos-app.sh launcher is kept.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Add one-command macOS quick start (start-macos.sh)
Running Odysseus natively on a Mac previously meant ~7 manual terminal steps
(brew deps, venv, activate, pip, setup.py, uvicorn with the right port) — not
friendly for a generic macOS user, and the native run is required because Docker
on macOS can't reach the Metal GPU.
- start-macos.sh: installs Homebrew deps (python@3.11, tmux, prebuilt Metal
llama.cpp), creates the venv, installs requirements, runs setup, and launches
on a non-AirPlay port (7860). Idempotent; re-run to start again.
- README: the Apple Silicon section now leads with this one-command quick start
and the clickable .app, with engine/port/manual details folded into a
collapsible block. Added a pointer at the top of the manual-install section.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* macOS quick start: auto-open browser when ready
The "open this URL" line scrolled out of view as uvicorn kept logging after it,
so users missed it. Now start-macos.sh waits (in the background) until the
server accepts connections, prints a boxed "ready" banner at that point (i.e.
after the startup burst, not before), and opens the URL in the default browser
automatically. Skippable with ODYSSEUS_NO_OPEN=1 for headless/SSH use.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Don't assume/force a specific Python version on macOS
The README claimed "system Python is 3.9" — a machine-specific generalization
that's often wrong (macOS ships no recent Python by default; many users already
have 3.11+). Make it generic, and make start-macos.sh detect an existing
Python 3.11+ and use it, only installing python@3.11 when none is found instead
of forcing it on top of the user's Python.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Align start-macos.sh venv path with build-macos-app.sh
start-macos.sh created the environment in .venv/, but build-macos-app.sh and
the manual install steps use venv/ — so the clickable .app wouldn't reuse the
quick-start's environment and would rebuild a second one. Use venv/ everywhere.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* README: state clearly that MLX is unsupported on Apple Silicon
Odysseus has no mlx_lm runtime; it serves GGUF (llama.cpp/Ollama) and CUDA
(vLLM/SGLang) only. MLX-only models can't run on a Mac and are hidden from
Cookbook — make that explicit in both the quick start and the details.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* start-macos.sh: build the venv with an arm64 Python on Apple Silicon
A clean-room run surfaced this: with a universal2/x86 Python (e.g. the
python.org installer under /usr/local), the venv's compiled extensions install
as arm64 but get loaded as x86_64 when launched from the .app bundle, so it
crashes with "incompatible architecture (have arm64, need x86_64)". The terminal
run happened to work only because a universal binary defaults to arm64 there.
On Apple Silicon, look only under /opt/homebrew (arm64-only) for the build
Python, and install Homebrew's python@3.11 if none is present — so the venv is
arm64-only and launches correctly from both the terminal and the .app. Intel
and non-mac paths are unchanged. Verified end-to-end in a clean clone: .app now
boots on Metal with no arch error.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Address dev-exp review: macOS setup robustness + doc/UX fixes
From the voltagent dev-exp review of the branch:
- README: fix broken anchor links (the em-dash heading produced a slug the links
didn't match); simplify the heading to a stable slug.
- cookbook_routes.py: add /opt/homebrew/bin and /usr/local/bin to the serve PATH
so a brew-installed llama-server/ollama is found instead of falling back to a
slow source build.
- start-macos.sh: guard against an empty Python path; fail fast with a clear
message on port-in-use; ERR trap with a "safe to re-run" message; show pip
progress (drop --quiet on the slow requirements install); stop the background
browser-opener cleanly on exit/Ctrl+C (no orphaned poller).
- setup.py: bind hint to 127.0.0.1; suppress the manual run-hint when launched
by start-macos.sh (ODYSSEUS_SKIP_RUN_HINT) so the URL isn't contradictory.
- build-macos-app.sh: the .app only opens the browser once the server is
actually ready (not after the readiness timeout).
- cookbookServe.js: drop "Diffusers" from the Metal backend picker —
diffusion_server.py is CUDA-only, so it was an unservable option on macOS.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: yunggilja <yunggilja@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
