add_directory cleared exclusions with a raw path.startswith(directory)
test, which also matched sibling directories sharing a name prefix —
adding /docs would silently un-exclude files under /docs2. Match the
directory itself or paths under it (directory + os.sep) instead.
_process_pdf prepends "\n\n[PDF content]:" to extracted text, and two
call sites in document_routes.py stripped it with .lstrip("\n[PDF content]:").
str.lstrip(chars) treats its argument as a *set of characters*, so it keeps
eating into the page text that follows the marker — e.g. a body starting
with "to the board" loses its leading "to" because 't'/'o' are in the
marker's character set. Replace both sites with a shared
strip_pdf_content_marker() helper that uses str.removeprefix.
_pip_install_fallback_chain silently discarded pip stderr via
2>/dev/null on every attempt. When pip failed (network error, venv
mismatch, disk full), the wrapper exited 0 and the Cookbook UI showed
the download as running — the silent-failure mode from #354.
Extract _pip_install_attempt() which wraps each pip invocation in a
bash -c subshell that captures output to a temp file, prints tail -5
on failure, cleans up, and exits with pip's real exit code. This
avoids the | tail pipefail masking (the first blocker on #363) while
surfacing the last 5 lines of pip output in the tmux log so users
can see what went wrong.
Both local wrapper and remote SSH runner use the same helper through
_pip_install_fallback_chain, so the fix is symmetric.
In Docker, a model-endpoint URL pointing at loopback (e.g. the LM Studio
default http://localhost:1234/v1) targets the Odysseus container itself, not
the host running the server, so the probe gets a connection error and the
endpoint is rejected with a misleading 'No models found for that provider/key'.
Rewrite loopback to host.docker.internal (which compose already maps to
host-gateway) for the probe and the saved URL, mirroring the existing Ollama
handling. Gated on actually being in a container with the gateway reachable, so
native installs and gateway-less deploys are untouched.
Fixes#25
Co-authored-by: Claude <noreply@anthropic.com>
Deep Research surfaced 'Error: unknown error' whenever every search
provider returned an empty result set without raising (e.g. SearXNG is
reachable but all its engines fail internally). _last_search_error was
only set on exceptions, so the empty-but-no-exception path left it unset
and the caller fell back to 'unknown error'.
Record an actionable reason on that path naming the providers that were
tried, so users can tell it's a search-backend problem rather than a
model problem. The provider-raised path is unchanged.
Re: #344.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The image-edit endpoint lookup compared stored vs incoming base URLs with
`.rstrip("/v1")`. `str.rstrip(chars)` treats its argument as a character
set, not a suffix, so any URL ending in '/', 'v', or '1' is over-stripped
(e.g. `http://host1/v1` -> `http://host`). Two endpoints that are not the
same can then compare equal, or the real endpoint fails to match its own
stored record, leaving `api_key` unset and sending the upstream image call
unauthenticated.
Use `.removesuffix("/v1")` (exact-suffix removal) with surrounding
`.rstrip("/")` on both sides so only a genuine trailing `/v1` is dropped.
Adds a focused test that parses the actual comparison expression out of
gallery_routes.py via AST and evaluates it — it fails if the fix is
reverted and uses no mocking.
`add_hwfit_models.py` infers `parameter_count` and `parameters_raw` by
regexing the HF repo name for a `<num>B` token, optionally with an
`-A<num>B` MoE active-param suffix. Repos that don't encode a size in
their name at all (e.g. `zai-org/GLM-4.5`, where the "4.5" is a version
not a parameter count) fall through to the safetensors element-count
path. That path works for unquantized FP16 / BF16 repos but is brittle
in two cases the catalog hits often:
1. Author-bulk runs (`AUTHORS = ["cyankiwi"]`) pull pre-quantized AWQ /
GPTQ / MLX repos. The safetensors metadata stores the packed I32
tensors and a per-dtype `parameters` map, which the script unpacks
via a per-quant pack factor. When the upload doesn't populate that
map (older repos, custom shards), `st.total` is used raw and the
parameter count is off by 4-8x.
2. Repos where the safetensors block is absent from `model_info()`
entirely. The current code returns `None` and silently drops the
model, which then has to be added to `EXTRA_REPOS` by hand with a
literal `parameter_count` string.
Both are exactly what the issue calls out — the regex / safetensors
combo can't size GLM-4.5 by itself because the name has no `<num>B`
and the upstream repo's safetensors block doesn't carry a usable param
total either.
Add a config.json fallback in front of the safetensors path:
- `_fetch_config_json(repo_id)` downloads `config.json` via
`hf_hub_download` (so the standard HF on-disk cache handles
deduplication across runs, no extra cache layer needed). Network /
404 / gated-repo errors return `None` and the caller proceeds to the
safetensors fallback. An in-process `_CONFIG_CACHE` dedupes the
base-model vs. source-repo lookups within a single run.
- `_params_from_config(cfg)` first honours explicit `num_parameters` /
`n_params` / `total_params` fields when present. Otherwise it sums
embeddings + attention (GQA-aware via `num_key_value_heads` and
`head_dim`) + dense MLP (`3 * hidden_size * intermediate_size`,
covering SwiGLU / GeGLU). For MoE configs it picks up both naming
conventions in the wild — `num_experts` / `num_experts_per_tok`
(Qwen3-MoE) and `n_routed_experts` / `n_shared_experts` (GLM-4-MoE,
DeepSeek-V3) — uses `moe_intermediate_size`, and respects
`first_k_dense_replace` so the first N layers stay dense. Active
parameters come out as `num_experts_per_tok + n_shared_experts` of
the routed experts, which matches how each architecture reports its
active count.
- In `_entry_from_modelinfo`, try config.json on the source repo first
(works for unquantized models) and then on the `base_model:` parent
(covers AWQ / GPTQ children whose own config is just a quantization
manifest). Both lookups run only when regex + override + base_model
tag all failed, so the normal author-bulk run still resolves sizes
from names without touching the Hub.
Spot-checks against the three architecture families this script
actually pulls — within ~5% of the documented param counts, which is
well inside the `parameter_count` rounding (one decimal of "B") and
the `min_vram_gb` downstream bucket:
Qwen2.5-7B-Instruct 7.62B (HF card: 7.6B)
Qwen3-30B-A3B 30.5B / 3.34B active (card: 30.5B / 3.3B)
GLM-4.5 352.7B / 33.6B active (card: 355B / 32B)
The safetensors path is unchanged and remains the last resort, so
repos with neither a parsable name nor a fetchable config.json behave
exactly as before.
Closes#955.
KNOWN_CONTEXT_WINDOWS lists 'o1' (200k) before 'o1-mini' (128k), and
_lookup_known returned on the first substring hit — so "o1-mini" matched
'o1' and reported 200000 instead of 128000. Track the longest matching
key instead, so the most specific entry wins regardless of table order.
`_ORIG_RE` (and its JS mirror `_TALON_ORIG_RE`) already recognised the
Japanese forward marker `転送` alongside the "Original Message" delimiters,
but not the English "Forwarded message" one. So Gmail-style forwards —
including the ones Odysseus itself emits (`---------- Forwarded message
----------`, static/js/emailInbox.js) — were not treated as a quote
boundary:
- with a following Outlook From:/Date: header block, the divider line
leaked into the level-0 reply bubble as noise;
- with only the divider marking the forward (no header block), the body
was not split into turns at all.
Add `Forwarded\s+message` to the same `[-_=]{3,}`-delimited alternation in
both the server-side parser and the JS mirror, so forward dividers are
consumed as an attribution boundary like "----- Original Message -----".
Locale variants of "Forwarded message" can follow the existing pattern.
Tests cover both manifestations plus a negative control (the bare words
"forwarded message" without `[-_=]{3,}` delimiters must not split).
Checks: python -m pytest tests/test_forwarded_message_divider.py (3 passed),
python -m py_compile src/email_thread_parser.py, node --check
static/js/emailLibrary/utils.js, git diff --check.
`get_tools_for_query` force-includes whole tool families when the query
mentions an intent keyword, but matched with a raw substring test
(`kw in ql`). Short hints therefore fired inside unrelated words, bloating
the tool set with irrelevant tools:
- "fix" matched "prefix" -> document tools
- "line" matched "deadline"/"online" -> document tools
- "serve" matched "observe"/"reserve" -> cookbook serve tools
- "reply" matched "replying" -> all email tools
- "unread" matched "unreadable" -> all email tools
Match each keyword on word boundaries instead
(`re.search(rf"\b{re.escape(kw)}\b", ql)`), the same fix already applied to
the keyword matcher in topic_analyzer.py. Genuine intent keywords
("reply to this email", "edit the document", "serve the model") still match.
This only removes substring-inside-a-word matches; it does not change whole
-word matches (so e.g. an unrelated whole word like "tell" is a separate
keyword-choice question, left untouched here).
Checks: python -m pytest tests/test_tool_index_keyword_boundaries.py (4 passed;
3 of them fail on the pre-fix substring code), python -m py_compile
src/tool_index.py, git diff --check.
PresetManager.load already heals a forward-incompatible presets.json: the
block just above repairs the legacy `custom` shape and re-saves the file.
But if the file exists and is missing a whole built-in preset (e.g. an older
install written before `reason` existed), load returned it as-is, so that
built-in stayed permanently absent — silently missing from the picker that
GET /api/presets feeds, with no way for the user to get it back.
Extend the same self-heal: after the legacy migration, fill in any built-in
presets the loaded file is missing, defaults-first so user edits win, and
persist the result. This never clobbers an intentional removal — there is no
delete path for the built-in keys (only user_templates entries can be
deleted), and presets are hidden via an `enabled: False` flag, not removal.
Checks: python -m pytest tests/test_preset_fill_missing_defaults.py (3 passed;
2 fail on the pre-fix code), the existing preset cases in
tests/test_review_regressions.py still pass, python -m py_compile
src/preset_manager.py, git diff --check.
The sync-chat endpoint's Case 3 fallback selected a ModelEndpoint with an
unscoped `query(ModelEndpoint).filter(is_enabled == True).first()` and then
used that row's decrypted `api_key` for the LLM call. ModelEndpoint is a
per-user resource (owner non-null = private to that user), so a chat-scoped
API token for user A that sent no session and no api_key could fall back onto
user B's PRIVATE endpoint — spending B's API key/quota and reaching whatever
internal base_url B configured. This is the same multi-tenant owner-scoping
class already fixed for the session gate on this very endpoint
(_caller_owns_session) and for companion/models.
Scope the fallback to the token owner's own rows plus legacy null-owner
(shared) rows via the existing owner_filter helper, matching
routes/model_routes.py and companion/routes.py. A null/empty owner stays a
no-op, preserving single-user/legacy behaviour.
Add regression tests pinning the scoped fallback (cross-owner, shared-only,
no-visible-row, disabled-owned, and the legacy null-owner no-op).
The list keyboard handler (_onSessionListKeydown) treats Backspace and
Delete as "delete the focused session". When the user double-clicks a
chat to rename it, an <input class="session-rename-input"> is mounted
inside the .list-item row. Backspace on the input bubbles up to the list
container, the handler walks closest('.list-item[data-session-id]') from
e.target, finds the parent row and DELETEs the session via the API —
so a single typo correction nukes the whole conversation.
Bail out at the top of the handler when e.target is an INPUT, TEXTAREA,
or contentEditable element. Arrow / Enter / Delete navigation still
works for rows themselves (the row is the focused element then, not the
input). Mirrors the guard pattern already used in ui.js, notes.js,
tasks.js, calendar.js, emailLibrary.js and galleryEditor.js.
Closes#1007.
The DELETE /api/skills/{skill_id} handler resolves the caller, loads the
skill with skills_manager.load(owner=user), and verifies ownership with
_verify_owner(match, user) — but then calls
skills_manager.delete_skill(match.get("name")) without the owner.
SkillsManager.delete_skill filters candidates with
`(sk.owner or "") != (owner or "")`, so when owner is None an owner-scoped
skill is skipped and the method returns False. The route then raises a
spurious 404 "Skill not found" — meaning a logged-in user can never delete
their own skills through the API.
Pass the resolved owner through to delete_skill so the skill is matched and
removed.
tests/test_skills_delete_owner.py drops a real owner-scoped SKILL.md on disk
and (1) checks the manager directly: delete_skill without owner returns
False (regression lock) while delete_skill(owner="alice") returns True and
removes the dir; (2) drives the real DELETE route handler and asserts it
returns {"ok": True} and deletes the file. The route test fails before this
change (404). Real SkillsManager + real filesystem, no mocking.
APIKeyManager.load() decrypts every stored key with a dict comprehension
and no error handling. If the .key file no longer matches the ciphertext in
api_keys.json — key rotated, a partial/!mismatched data restore, or a
corrupted .key — Fernet.decrypt raises cryptography.fernet.InvalidToken.
app_initializer.py calls api_key_manager.load() during startup, so a single
undecryptable entry takes down the whole app at boot, and the user can't
reach the UI to fix it.
Decrypt each key in a loop and, on InvalidToken/ValueError, log a warning
and skip that one entry while still returning every key that decrypts
cleanly. One bad/stale key no longer blocks startup.
tests/test_api_key_manager_resilience.py saves a valid key, then injects an
entry encrypted under a different Fernet key (InvalidToken) and a malformed
token (ValueError), and asserts load() returns the good key and skips the
bad ones without raising. Fails before this change.
The webhook URL guard's _ip_is_private() only checks a hardcoded
_PRIVATE_NETWORKS list, which misses several addresses that route
internally. validate_webhook_url() therefore ALLOWED:
- http://[::]/ (IPv6 unspecified, reaches localhost)
- http://[::ffff:127.0.0.1]/ (IPv4-mapped IPv6 loopback = 127.0.0.1)
- http://[::ffff:169.254.169.254]/ (IPv4-mapped cloud metadata endpoint)
The last one is the dangerous case: a webhook pointed at the mapped
169.254.169.254 can pull cloud instance credentials (SSRF -> credential
theft).
Harden _ip_is_private(): first unwrap IPv4-mapped IPv6 to its embedded IPv4
(addr.ipv4_mapped), then reject via the stdlib address properties
(is_private, is_loopback, is_link_local, is_reserved, is_multicast,
is_unspecified) in addition to the existing network list. Public addresses
still pass.
tests/test_webhook_ssrf_resilience.py asserts validate_webhook_url raises
for the three IPv6 bypasses plus 127.0.0.1 and 0.0.0.0, and still accepts a
public IP literal. The IPv6 cases fail before this change.
_ping_endpoint() is the reachability fallback the model-endpoint POST
handler invokes when _probe_endpoint() returns no model ids. It GETs
base + "/models" and, on any sub-500 response, returns immediately with
`reachable = (status < 400)`. That early return runs before the
Ollama-native /api/version / /api/tags fallback below it.
For an Ollama URL without /v1 (the quickstart accepts both
http://localhost:11434 and http://127.0.0.1:11434, and the reporter
on #1025 explicitly tried both), the OpenAI-style probe target is
http://127.0.0.1:11434/models. Ollama returns 404 there because /models
only lives under /v1. _ping_endpoint then returned reachable=False and
the picker showed "Added (offline — will retry on next load)" on an
install that was running fine. /api/version was never tried.
Same shape for http://127.0.0.1:11434/api (the native Ollama root):
/api/models is also 404, same premature offline verdict.
_probe_endpoint() does fall through to /api/tags on a 4xx (the response
raises via raise_for_status), so the endpoint quietly recovers once
cached_models becomes non-empty on the next background refresh —
matching the second commenter's "had to disconnect manually then
reconnect for it to be detected" note. The bug is most visible while
no models are pulled yet (cached_models stays empty, _ping_endpoint
keeps voting offline).
Fix:
- Hoist the Ollama-shaped-URL test (port == 11434 or "ollama" in
hostname — the same condition _probe_endpoint already uses) to the
top of the function so both code paths share it.
- Stop short-circuiting on 4xx when the URL looks like Ollama: fall
through to the existing /api/version + /api/tags reachability loop
so an alive Ollama gets recognised even when its OpenAI surface has
the wrong prefix for the user's input.
- Fix the `root` computation in that loop to strip a trailing /api as
well as /v1, so http://127.0.0.1:11434/api no longer gets probed at
/api/api/version.
- 4xx on non-Ollama hosts keeps the current semantics: a 401 from
api.openai.com/v1/models is still a definitive offline verdict, not
a reason to GET /api/version on OpenAI.
Closes#1025.
_build_ollama_payload sends options.temperature and options.num_predict
to /api/chat, but never options.num_ctx. Ollama defaults num_ctx to 2048
when the option is omitted, so prompts going to any Ollama backend are
silently truncated there regardless of the model's actual capability.
Thread the discovered context length through the three call sites
(llm_call, llm_call_async, stream_llm) and emit options.num_ctx when it
is known and positive. The builder filters out the DEFAULT_CONTEXT
fallback (128000) so we don't lie to Ollama about models whose window
we couldn't actually discover. The issue's literal 'when > 2048'
heuristic is dropped: a model with a real context smaller than 2048
would OOM if Ollama used its default, so we pass the real value
regardless of size. Matches how src/context_compactor.py uses the
same helper.
Sister fix to PR #753 — that PR teaches the compactor the right budget,
this one tells Ollama to actually use that budget on the way in.
* feat(setup): prompt for admin credentials interactively on first run
When setup.py runs in a terminal (TTY) without env vars set, it now
asks the user to choose a username and password instead of generating
a random one that scrolls off-screen. Includes confirmation prompt
to catch typos.
Existing behavior is preserved:
- ODYSSEUS_ADMIN_USER + ODYSSEUS_ADMIN_PASSWORD env vars take priority
- Non-interactive contexts (Docker, CI) still get a random password
- ODYSSEUS_SKIP_ADMIN_PROMPT=1 opts out of the interactive prompt
- Re-runs still skip if auth.json already exists
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(macos): use venv Python for pip install and uvicorn launch
On PEP 668 systems (newer Homebrew Python), pip install outside a venv
is rejected. The script creates a venv but then called the system $PY
for pip and uvicorn. Switch to ./venv/bin/python for both.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Revert "fix(macos): use venv Python for pip install and uvicorn launch"
This reverts commit 7a1be956659d86183da2edcde2114eb363efd3e4.
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Rebased on current main. Integrates with the new Recent/Favorites
system — provider groups appear below Recent and Favorites in browse
mode for large catalogs (>12 models).
Changes:
- Models grouped by canonical provider with collapsible sections
- Chevron animation consistent with sidebar sections
- Domino cascade on expand (only on just-opened group)
- Provider display names (deepseek-ai -> DeepSeek, meta -> Llama, etc.)
- Alias merging (meta + meta-llama -> one Llama group)
- Search includes provider display names for filtering
- Collapsed state persists in localStorage
- No screenshot binary committed
Co-authored-by: danielxb <5981902+danielxb@users.noreply.github.com>
tool_execution.py returns web search results as {"output": ..., "exit_code": 0}.
The sources-extraction block in stream_agent_loop only checked result.get("results")
and result.get("stdout"), so _src_text was always "" for every tool-call-mode web
search. Two consequences:
1. The SOURCES marker was never parsed and the web_sources SSE event was never
emitted -- the sources panel never appeared after agent-mode searches.
2. The marker (a large JSON blob) was left in result["output"] and forwarded
verbatim to the LLM in round 2 via format_tool_result, confusing some local
models into producing no tokens.
Fix: prepend result.get("output") to the lookup chain, and update the cleanup
assignment so result["output"] is overwritten with the stripped text.
Adds six regression tests in tests/test_agent_loop.py documenting the before/after
behaviour and verifying backward compat with the legacy results/stdout paths.
Co-authored-by: MohammadYusif <MohammadYusif@users.noreply.github.com>
The Fit column shared the Score column's sort key, so clicking the Fit
header sorted by Score instead of by hardware fit. There was also no
fit option in the hidden sort <select> and no fit branch in the
client-side comparator.
- Give the Fit column its own sort key (fit).
- Add a fit option to the sort select (kept Score as the default so
first-load ordering is unchanged).
- Sort by the categorical fit_level rank
(perfect > good > marginal > too_tight), tie-broken by score, honoring
the ascending/descending toggle.
Fixes#842
Co-authored-by: SabixMaru <285860855+SabixMaru@users.noreply.github.com>
Document safer defaults and deployment guidance for network-accessible
Odysseus installs. The guidance emphasizes keeping auth enabled,
disabling localhost bypass outside development, using secure cookies for
HTTPS/reverse-proxy deployments, and exposing only the authenticated
Odysseus entrypoint through a trusted proxy or private access layer.
Also clarify that bundled services, databases, vector stores,
notification services, and raw model/provider APIs should remain
internal-only.
This is documentation and config-example only. It does not change
runtime behavior.
* 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.
Models (notably Gemini) emit a native 'google_search' function call, but the
agent loop had no mapping for it, so the call failed to convert, the round
produced 0 chars and 0 tool blocks, and generation died silently — the web
client hung on 'waiting for first token' with no error (also #443).
- Map google_search / google_search_retrieval / google_search_grounding to the
web_search tool, and read Gemini's 'queries' array (falling back to 'query').
- In stream_agent_loop, when a round yields no response text and no tool
events, emit a visible fallback message instead of leaving the user hanging.
- Give the unknown-tool execution branch an explicit exit_code=1 so the failure
is logged as an error rather than 'n/a'.
Unknown/unconvertible tool names still return None (unchanged) so they are
dropped safely rather than executed. Added tests covering the google_search
mapping, the queries array, and unknown/invalid-JSON returning None.
* fixed confusing credentials prompt
* fix(setup): return status from create_default_admin function
* fix(setup): initialize admin creation status in main function
* fix(setup): enhance admin creation feedback and status handling
* Enhance admin user login messages with conditional feedback based on creation status
* Refine admin user creation feedback messages for clarity and actionability and formatted code
* Add fallback error message for admin creation failure in setup script
* Add run script for Uvicorn with dotenv integration
* Refactor server runner to use argparse for host and port configuration
* Remove captured output print statement from server runner
* Fix server runner to ensure cross-platform compatibility and improve log handling
* Remove run.py script to match main repo
* feat: add custom option for search result count in settings
* fix: enforce minimum and maximum values for custom search result count
The existing test_endpoint_resolver.py copies the pure functions to avoid
import side effects, so its assertions can silently drift from the shipped
src/endpoint_resolver.py (the copies already lag: no OpenRouter headers, no
anthropic.com host matching). This adds a sibling module that imports the
REAL resolver and locks in behavior for every provider named in ROADMAP.md's
"Provider setup/probing audit" — Anthropic, Gemini, Groq, xAI, OpenRouter,
OpenAI, DeepSeek — plus Ollama (local + cloud) and the Tailscale self-host
fallback in resolve_url.
Covers build_chat_url, build_models_url, build_headers, normalize_base,
_first_chat_model, _anthropic_api_root, _ollama_api_root, and resolve_url.
conftest.py already stubs the heavy deps, so the import is side-effect free.
Test-only; no behavior change. 55 new tests, all passing.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Split 3/4 of the companion bridge (#863, #871 landed 1/4 and 2/4). Adds admin-only
device pairing to the companion router.
- GET /api/companion/pair -- renders a form; never mints (a GET must not mint a
credential: SameSite=Lax session cookies ride top-level GET navigations, so
GET-minting would be CSRF-triggerable via a link/<img>)
- POST /api/companion/pair -- mints a one-time chat-scoped token. Admin-cookie
only; CSRF-safe because a SameSite=Lax cookie is not sent on a cross-site POST,
the same protection POST /api/tokens relies on. ?format=json returns the
pairing payload for an in-app screen.
Minting invalidates the auth middleware's token cache so the code works on the
next request with no restart. companion/pairing.py holds the mint/LAN/QR helpers;
the token is shown once and stored only as a bcrypt hash + prefix
(mirrors routes/api_token_routes.py).
Tests (tests/test_companion_pairing.py):
- a bearer/'api' caller and a non-admin user are rejected by require_admin (403);
an admin passes
- the token is returned once and persisted only as a hash
- minting invalidates the cache (works without restart)
- minting is exposed on POST, never GET (CSRF)
Issue #234: the "Character" tab and its "Style of response" label made it
unclear that this is where a system prompt is set. Rename the user-facing
labels for clarity:
- "Character" tab + section heading -> "Persona"
- "Style of response" -> "System prompt"
- supporting strings: select placeholder, name placeholder, button/title
text, toasts, confirm/notice text, the chat-bar indicator tooltip, the
settings visibility toggle, and the assistant personality picker
("Characters" optgroup -> "Personas").
Used "Persona" rather than the issue's suggested "Preset" because the app
already has a distinct, user-facing "Presets" concept (built-in presets
like Code Analyze/Brainstorm/Reason, shown as their own group in the
assistant picker). "Persona" matches what this tab actually creates -- a
named persona with its own memories -- without colliding with that term.
Internal identifiers (element IDs, data-chartab attributes, function names)
and the character_name backend field are intentionally left unchanged so
existing saved presets and JS wiring keep working.
Screen readers got no signal that a dialog opened — not one modal carried
role="dialog" — and several close buttons had no accessible name.
- The 6 static tool windows (Brain, Theme, Prompt, Rename session, Cookbook,
Settings) now carry role="dialog" + an accessible name. They are dockable,
tiling windows, so they are non-modal dialogs (intentionally no aria-modal).
- The four unlabelled close buttons (theme, prompt, cookbook, settings) get an
aria-label so they no longer read as just "heavy multiplication x".
- styledConfirm / styledPrompt ARE blocking modals: they get role="dialog" +
aria-modal="true" + aria-labelledby/aria-describedby, and now manage focus —
restore focus to the triggering element on close and trap Tab within the
dialog (they already moved focus in on open).
tests/test_dialog_aria.py pins the roles, labels, and focus management.
* 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>
* Improve Docker GPU setup diagnostics
Add a Docker GPU preflight script for NVIDIA users. The script is
read-only by default, checks host NVIDIA drivers, Docker availability,
and container GPU passthrough, and prints actionable next steps.
Add explicit opt-in modes to print install commands, install NVIDIA
Container Toolkit on Ubuntu/Debian, and enable the NVIDIA Compose overlay
in .env after passthrough is verified.
Document common NVIDIA Docker failure modes, ignore generated .env
backups, and clarify that Cookbook can only detect GPUs exposed to the
Odysseus container.
* Clarify Docker GPU diagnostic limits
Two small polish items in the Cookbook Serve panel.
Saved-config badge
The little count badge next to the Save button ("3 ▾" etc.) had a
generic "Saved launch configs" tooltip, so the number reads like a
notification dot. Make it spell out what it is and what clicking does:
"3 saved launch configs for <model> — click ▾ to load or delete"
(and "No saved launch configs for <model> yet — click Save to add
one" when empty). Tooltip stays in sync via _updateSavedToggleLabel
so save/delete updates both the count and the hint.
GPU chip on mixed-GPU boxes (#711)
The chip label was `${gpuCount}x ${gpu_name}`, where gpu_name is
just gpus[0].name — so a 4090 + 3060 reads as "2x RTX 4090". The
backend already emits gpu_groups (identical cards grouped, used by
the serve flow to pin CUDA_VISIBLE_DEVICES) and a per-card gpus[]
array, so use them:
- Label renders each homogeneous pool: "1× RTX 4090 + 1× RTX 3060".
Homogeneous setups keep the existing "2× RTX 4090" form.
- Tooltip lists each GPU with its index + VRAM, useful for picking
the right device when launching.
Refs #711.
Adds a diagnosis pattern for the 'Failed to infer device type' error
vLLM raises when no CUDA or ROCm GPU is found (e.g. systems with only
integrated or Intel Xe graphics). The existing pattern only caught
'No CUDA GPUs are available' which fires later in startup; this new
entry catches the earlier device-probe failure and the NVML/amdsmi
library-not-found messages that precede it.
Surfaces in the Cookbook serve card as: "vLLM could not find a supported
GPU — switch to llama.cpp or Ollama" instead of a raw Python traceback.
Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
Windows has 'App Execution Aliases' that can make shutil.which('python3')
and shutil.which('python') resolve to a Microsoft Store stub instead of
real Python -- even when Python is properly installed. The stub outputs:
'Python was not found; run without arguments to install from the
Microsoft Store, or disable this shortcut from Settings > Apps >
Advanced app settings > App execution aliases.'
and exits 9009, producing empty stdout. The JSON parse of the local
model cache scan then fails with 'Expecting value: line 1 column 1
(char 0)', and the Cookbook model list shows nothing.
Fix: prefer sys.executable as the interpreter for the local scan.
Odysseus already runs inside its own venv, so sys.executable always
points to the real venv Python and bypasses PATH / Store alias lookup
entirely. which_tool() is kept as a fallback.
Cross-platform: sys.executable works identically on Linux and macOS
(returns the real interpreter path), so this change is safe everywhere.
