Fixes the CI failure introduced by #817: test_model_resolver::test_custom_endpoint_uses_model_config_api_key_for_model_discovery was failing with KeyError: 'auth' due to the 60s TTL cache in get_available_models() being populated by test_byok_model_dropdown.py tests that ran earlier. Added autouse _isolate_models_cache fixture to 5 test files. Full suite 1747/1747, QA harness green.
Closes#815.
Three root causes fixed:
1. Provider aliases (z.ai/x.ai/google/grok/claude/aws-bedrock/dashscope/~25 more) not
normalized before _PROVIDER_MODELS lookup — provider fell to empty else-branch while
TUI worked (it normalizes at startup). Fixed via _resolve_provider_alias() + inlined
_PROVIDER_ALIASES table in api/config.py.
2. Silent ImportError in original normalization: 'from hermes_cli.models import
_PROVIDER_ALIASES' inside try/except silently failed without hermes-agent on sys.path
(CI, minimal installs). The inlined table fixes this — normalization now works
regardless of whether hermes-agent is installed.
3. /api/models/live?provider=custom now falls back to custom_providers entries from
config.yaml when provider_model_ids() returns empty.
Also: provider_id on every group in /api/models response for deterministic JS optgroup
matching (no substring false positives). 17 targeted tests, 1725/1725 full suite.
* fix: update banner conflict recovery + server self-restart after update (#813#814)
* fix(update): restart must wait for in-flight update + reset force button on retry
Two defects in the update banner flow found during review of PR #816:
1. Two-target race (webui + agent sequential)
The client posts targets sequentially: webui succeeds and schedules
a restart timer (2 s delay); client then posts agent; server begins
agent fetch+pull; at T=2 s the restart timer fires os.execv mid-pull,
killing the agent update and closing the client connection. User
sees "Update failed (agent): Failed to fetch" even though webui did
update, and the agent repo is in an unknown partial state.
Fix: _schedule_restart() now blocks on _apply_lock before calling
os.execv. If a second update is in flight when the timer fires, the
restart thread waits until it completes. If nothing is in flight the
lock acquire is instant, so no-op updates still restart immediately.
2. Stale force-update button across retries
_showUpdateError sets btnForceUpdate to display:inline-block when
res.conflict / res.diverged. Nothing resets it on the next retry,
so a subsequent non-conflict error (e.g. network) leaves the stale
force button visible pointing at the previous target.
Fix: applyUpdates() now hides the force button and clears its
data-target at the start of each attempt.
Tests:
- test_schedule_restart_waits_for_apply_lock: holds _apply_lock from a
helper thread, verifies execv is delayed until the lock is released.
- test_schedule_restart_still_fires_when_no_update_in_flight: sanity
check that the common path still works with no contention.
- test_apply_updates_resets_force_button_at_start: regression guard
that the reset appears before the update loop begins.
Full suite: 1683 passed, 0 failures.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(update): hold _apply_lock through execv + fix banner error layout
Two fixes from Opus review:
1. TOCTOU gap in _schedule_restart (api/updates.py): the original pattern
acquired _apply_lock, released it, then called os.execv — leaving a brief
window where a new update could start between release and execv. Fixed by
moving os.execv inside the 'with _apply_lock:' block so the process is
replaced while still holding the lock; no new update can acquire it.
2. Banner CSS layout (static/index.html): #updateError was a direct flex child
of .update-banner (display:flex row), so long error messages sat inline
between #updateMsg and the buttons instead of below the message.
Wrapped #updateMsg + #updateError in a flex-column container so errors
stack vertically under the status line.
* docs: add v0.50.134 CHANGELOG entry
---------
Co-authored-by: nesquena-hermes <nesquena-hermes@users.noreply.github.com>
Co-authored-by: Nathan Esquenazi <nesquena@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes#461
Adds full /reasoning CLI parity to the WebUI slash command system:
- /reasoning show|on → window._showThinking = true; writes display.show_reasoning to config.yaml (same key as CLI); mirrors to settings.json for boot.js
- /reasoning hide|off → same in reverse; re-renders immediately
- /reasoning none|minimal|low|medium|high|xhigh → POST /api/reasoning → writes agent.reasoning_effort to config.yaml; takes effect next turn (matching CLI semantics)
- /reasoning (no args) → GET /api/reasoning → live status toast from config.yaml
- Autocomplete shows all 8 options: show|hide|none|minimal|low|medium|high|xhigh
- Profile-isolated: _get_config_path() is thread-local so per-profile settings never bleed across
- Boot hydration: window._showThinking initialised from settings.json show_thinking on page load
- Inspect.signature guard in streaming.py so older hermes-agent builds don't TypeError
28 new tests, 1708/1708 total passing. Full browser QA on port 8789 with isolated state. CLI/config.yaml sync verified with hermes_constants.parse_reasoning_effort().
Fixes the multi-client profile isolation bug (#798).
- get_hermes_home_for_profile(): pure path resolver, validates name against
_PROFILE_ID_RE (rejects path traversal), never mutates os.environ or globals
- new_session() accepts explicit profile= param from POST body (S.activeProfile),
short-circuits the process-level _active_profile global
- streaming handler resolves HERMES_HOME from s.profile instead of the global
- sessions.js sends profile: S.activeProfile in every new-session POST
10 tests in tests/test_issue798.py including concurrency and traversal coverage.
Co-authored-by: nesquena <nesquena@users.noreply.github.com>
fix(onboarding): recognize credential_pool OAuth auth for openai-codex (#797)
The onboarding readiness check in `api/onboarding.py` only looked at the legacy
`providers[provider]` key in `auth.json`. Hermes runtime resolves OAuth tokens from
`credential_pool[provider]` (device-code / OAuth flows), so WebUI could report "not ready"
while the runtime chatted successfully. The check now covers both storage locations with
a fail-closed helper. Adds three regression tests.
Reported in #796, fixed by @davidsben.
Co-authored-by: davidsben <davidsben@users.noreply.github.com>
The v0.50.124 Docker build failed with:
cannot create /apptoo/api/_version.py: Permission denied
Root cause: 'USER hermeswebuitoo' is set before 'COPY . /apptoo', but
COPY without --chown creates files owned by root. The subsequent RUN
step (which writes api/_version.py) runs as hermeswebuitoo and has no
write permission to the root-owned api/ directory.
Fix: COPY --chown=hermeswebuitoo:hermeswebuitoo so the unprivileged user
owns the app files and can write _version.py at build time.
Regression from #790.
Co-authored-by: nesquena-hermes <hermes@nesquena.com>
* fix: bootstrap.py loads REPO_ROOT/.env so direct invocation matches start.sh
When users run 'python3 bootstrap.py' directly (the primary documented
entry point in README), HERMES_WEBUI_HOST, HERMES_WEBUI_PORT and other
.env settings were silently ignored because the shell-level 'source .env'
in start.sh was never executed.
Add _load_repo_dotenv() in bootstrap.py that reads REPO_ROOT/.env into
os.environ before DEFAULT_HOST / DEFAULT_PORT are evaluated at module
level. Uses unconditional assignment matching 'set -a; source .env'
shell semantics. Only loads the repo .env (bootstrap config) — not
~/.hermes/.env, which the server still loads independently at startup
for provider credentials.
Reported in #730 by @leap233 who had HERMES_WEBUI_HOST=0.0.0.0 and
HERMES_WEBUI_PORT=18787 in the webui .env; running bootstrap.py directly
caused the server to ignore both settings.
Tests: 15 new tests in tests/test_bootstrap_dotenv.py covering the
full loader (key=value, comments, blank lines, quoted values, no-file,
unreadable-file, overwrite semantics, values with = signs) and structural
assertions that _load_repo_dotenv() is called before DEFAULT_HOST/PORT.
1613 tests total.
* fix: address review feedback on PR #791
- bootstrap.py: document overwrite semantics and 'export' note in docstring
- bootstrap.py: handle 'export FOO=bar' prefix (strip before splitting on =)
- bootstrap.py: print warning to stderr on .env parse failure (not silent swallow)
- bootstrap.py: add side-effect comment at _load_repo_dotenv() call site
- CHANGELOG.md: restore v0.50.124 and v0.50.123 headers (were merged into
v0.50.125 section, making three consecutive ### Fixed blocks with no ## header
between them)
- tests: fix test_noop_when_dotenv_unreadable to assert warning is emitted
- tests: tighten test_does_not_set_empty_values with concrete assertion
- tests: add test_export_prefix_stripped
- tests: remove dead _import_bootstrap_with_env() helper (never called)
1614 tests total
---------
Co-authored-by: nesquena-hermes <hermes@nesquena.com>
* fix: dynamic version badge — read from git tag, never hardcoded
The settings panel showed v0.50.87 and the HTTP Server: header said
HermesWebUI/0.50.38 — both hardcoded strings that drift further behind
with every release because there was no mechanism to keep them in sync.
Changes:
- api/updates.py: add _run_git() (moved before _detect_webui_version),
_detect_webui_version(), and WEBUI_VERSION module constant resolved
once at import time via 'git describe --tags --always --dirty'.
Fallback chain: git → api/_version.py → 'unknown'.
- api/routes.py: inject webui_version into GET /api/settings response
so the frontend can read it without a separate API call.
- static/panels.js: loadSettingsPanel() populates .settings-version-badge
from settings.webui_version — one line after the existing api() call.
- static/index.html: replace stale hardcoded 'v0.50.87' with '—'
placeholder; JS overwrites it as soon as the settings panel opens.
- server.py: replace hardcoded 'HermesWebUI/0.50.38' server_version with
'HermesWebUI/' + WEBUI_VERSION.lstrip('v') — stays in sync automatically.
- Dockerfile: add ARG HERMES_VERSION=unknown and write api/_version.py
so Docker images (where .git is excluded) still show the correct tag.
- .github/workflows/release.yml: pass build-args: HERMES_VERSION=${{ github.ref_name }}
to the Docker build step on tag pushes.
- .gitignore: exclude api/_version.py (generated by Docker/CI, never committed).
No manual 'update the version badge' step is required going forward.
Tagging is sufficient — the badge and HTTP header update automatically.
Tests: 18 new tests in tests/test_version_badge.py covering the full
resolution chain, /api/settings injection, HTML placeholder, JS wiring,
and server.py import. 1596 tests pass total.
* fix: address review feedback on PR #790
- api/updates.py: replace exec() with regex parse for api/_version.py
(no supply-chain risk from build artifact; exec unnecessary for one assignment)
- api/updates.py: cap git describe timeout at 3s (was 10s — import-time
stall on NFS/.git would block server startup unnecessarily)
- server.py: lstrip('v') → removeprefix('v') (lstrip strips chars not prefix)
- server.py: emit bare 'HermesWebUI' when version is 'unknown' rather than
'HermesWebUI/unknown' (log aggregators expect semver-ish suffix or none)
- CHANGELOG.md: add v0.50.124 entry for this user-visible change
- tests: rename exec-error test to reflect regex behaviour; add tests for
removeprefix usage and unknown-version header guard (1598 tests total)
---------
Co-authored-by: nesquena-hermes <hermes@nesquena.com>
The test was restoring original_model from /api/models, but after prior runs
the config.yaml model.default field could be stale, causing the restore to
bake in the wrong value. Fix: always restore to TEST_DEFAULT_MODEL (the
conftest-injected env value) for deterministic ordering-independent cleanup.
Also exposes TEST_DEFAULT_MODEL from _pytest_port.py so other tests that
mutate the default model can use it for clean teardown.
TESTING.md: update automated test count from 1353 to 1578.
At the @media(max-width:900px) breakpoint both .close-preview and .mobile-close-btn were visible simultaneously. Since boot.js wires both to handleWorkspaceClose(), only the mobile-close-btn needs to show at that width. Adds .close-preview{display:none} to the 900px media block.
Fixes#781
cancel_stream() now pops STREAMS/CANCEL_FLAGS/AGENT_INSTANCES and clears session.active_stream_id immediately after signalling cancel. Fixes sessions permanently stuck at 409 when the agent thread is blocked in a bad tool call. Session cleanup runs outside STREAMS_LOCK to preserve lock ordering.
Fixes#653
Co-authored-by: bergeouss <bergeouss@users.noreply.github.com>
Queued follow-up messages now survive page refresh. Persisted atomically in queueSessionMessage/shiftQueuedSessionMessage. On reload: if agent still active, queue is silently hydrated (done handler drains it); if idle, first entry is restored as a composer draft with a toast. Stale entries discarded.
Fixes#660
- quota_exhausted error type: distinguishes credit exhaustion from rate limits
- Streaming errors persisted to session file so they survive page reload
- _error flag excludes persisted errors from subsequent LLM API calls
- stream_end and title SSE events use original session_id (not s.session_id which rotates during context compaction)
Fixes#739, #652, #653
Removes the bubble_layout toggle from Settings, all persistence, CSS, i18n strings, and the UI docs demo. The CSS was already effectively dead. Users with a saved bubble_layout value in settings.json get a clean migration via _SETTINGS_LEGACY_DROP_KEYS.
Credit: @aronprins (PR #760 / #777)
Co-authored-by: aronprins <aronprins@users.noreply.github.com>
Removes split-brain where WebUI Settings persisted default_model separately from Hermes runtime config.yaml. New POST /api/default-model endpoint writes to config.yaml. Existing saved values migrated on first load.
Fixes#761
Co-authored-by: aronprins <aronprins@users.noreply.github.com>
Adds sub-argument suggestions for /model, /personality, /reasoning slash commands. /reasoning is now discoverable from the first slash. Keyboard navigation pre-selects the first item. Fixes bug where no-arg commands (/clear, /new, /stop, etc.) would loop the dropdown on selection.
Fixes#632
Co-authored-by: franksong2702 <franksong2702@users.noreply.github.com>
Adds compact/detailed toggle for the session list sidebar. Compact is the default (no behavior change for existing users). Detailed mode shows message count and model; profile names only appear when mixing sessions across profiles.
Fixes#673
Co-authored-by: franksong2702 <franksong2702@users.noreply.github.com>
## Summary
Rebased on behalf of @aronprins from fork branch `codex/dark-user-bubbles`. Two asset-only commits (PR screenshot add/remove) were dropped; the two code commits are applied cleanly on top of current master (v0.50.110).
### What changed
**Dark-mode user bubbles** (`static/style.css`):
- `:root.dark` now overrides `--user-bubble-bg`/`--user-bubble-border` to `var(--accent-bg-strong)` (a 15% opacity tint) — keeps the bubble visually subdued in dark skins instead of a glaring bright accent fill
- Removes 6 per-skin `--user-bubble-text` hacks (ares, mono, slate, poseidon, sisyphus, charizard); text falls back to `var(--text)` which is already correct in dark mode
- Adds `--user-bubble-placeholder` token; edit-area box-shadow now uses `--focus-ring` instead of hardcoded `rgba(255,255,255,.15)`
**Thinking card collapsibility** (`static/ui.js` + `static/style.css`):
- `_thinkingMarkup()` now includes `onclick` toggle and chevron affordance, matching the compression reference card pattern
- `.thinking-card-header` gets `display:flex; gap:8px` for proper icon/label/chevron alignment
**Tests**: 2 new in `test_bugbatch_apr2026.py` (dark bubble token contract + no-per-skin-hack assertion), 2 updated in `test_ui_card_animation.py` (flex header layout + onclick pattern).
1520 passed. QA 20/20. Browser verified: dark mode bubble uses subtle tint, thinking card toggles correctly.
(credit: @aronprins)
Squash merge of PR #717 — rebased on behalf of @franksong2702.
## What it does
Fixes#680. Footer chrome (timestamps, copy, edit, regenerate) is now hover-only for both user and assistant message rows, consistent throughout the conversation. The last assistant turn keeps cumulative usage visible at rest; timestamp and actions are revealed inline on hover in the same row.
Key changes:
- `static/ui.js`: new `_formatMessageFooterTimestamp()` (local timezone, cross-day fuller format); `timeHtml` no longer gated to user-only; last assistant usage moved from separate `.msg-usage` div to inline `.msg-usage-inline` span in the footer
- `static/style.css`: `.msg-foot-with-usage` class + rules; assistant footer opacity changed from 0.45 to 0 (hover-only); `:focus-within` alongside `:hover` for keyboard users
- `api/streaming.py`: `_restore_reasoning_metadata()` now preserves `_ts`/`timestamp` for unchanged historical messages
- `tests/test_sprint49.py`: 8 new tests covering rendering contract, hover CSS, timestamp preservation
Tests: 1518 passed. QA: 20/20. Browser verified. Reviewed and approved by @nesquena and @aronprins.
## Summary
Fixes the ordering-dependent test failure in `test_custom_provider_display_name.py` (issue #754).
**Root cause:** `_models_with_cfg()` patches `config.cfg` then calls `get_available_models()`. That function checks `config.yaml`'s mtime on every call — if it has changed since the last `reload_config()`, it calls `reload_config()` again, which reads from disk and silently overwrites the patch. Any test that writes `config.yaml` (e.g. via `save_settings()`) before this test runs changes the mtime and triggers the reload.
**Fix:** Pin `config._cfg_mtime` to the current `config.yaml` mtime before calling `get_available_models()`, then restore it in the `finally` block. This is the same pattern already used in `test_model_resolver.py` (lines 249, 393).
**Also restores `_cfg_mtime`** in the `finally` block so the patch leaves no side effects on subsequent tests.
## Tests
1510 passed — the previously-flaky test now passes regardless of which tests ran before it.
Closes#754
## Summary
Follow-up to #751/#752. Code review identified a case where `_normalize_session_model_in_place` could call `session.save()` (which triggers a full session index rebuild) for sessions with `model: null` or missing model field.
Root cause: `_resolve_compatible_session_model(None)` returns `(default_model, True)` when a default exists — which was interpreted as "changed, needs save." But there's nothing to correct for a session with no model; the default is just a fallback for display purposes, not a cross-provider correction worth persisting.
Fix: capture `original_model` before calling `_resolve_compatible_session_model`. Only call `session.save()` if `original_model` was non-empty and actually changed.
Adds a test asserting `save_calls == []` when `session.model is None`.
No behavior change for sessions with a real model (the primary use case of #751 is unaffected).
## Summary
Regression fix for #751.
Models with custom or unrecognized prefixes (e.g. `custom-provider/my-model`, `test/import-model`) were being incorrectly replaced with the active provider default. Root cause: `_normalize_provider_id("custom-provider")` matched the `"custom"` prefix and returned `"custom"`, which ≠ `active_provider` → normalization fired.
Two-part fix:
1. Add `"custom"` and `"openrouter"` to the `model_provider` exclusion set in `_resolve_compatible_session_model` (parallel to the existing `active_provider` guard)
2. Return `""` for unknown prefixes in `_normalize_provider_id` so the `if model_provider` truthiness check safely short-circuits
Adds a regression test covering `custom-provider/`, `test/`, `my-local-llm/`, and `lmstudio-community/` prefixes.
## Tests
1499 passed, 0 failures (was 2 failures before this fix)
## Summary
Rebased-on-behalf of @likawa3b (originally PR #748 — stale base).
Sessions can outlive provider changes. When an old session still points to a model from a previous provider (e.g. `gemini-3.1-pro-preview` after switching the agent to OpenAI Codex), starting a chat hits the wrong backend and fails silently.
This PR adds a lightweight normalization pass:
- `_normalize_provider_id()` maps common prefixes to canonical provider IDs
- `_resolve_compatible_session_model()` checks the session model's provider against `active_provider` and returns the default model if they differ
- `_normalize_session_model_in_place()` is called at GET `/api/session` — corrects and persists stale models once
- Chat start also normalizes via `_resolve_compatible_session_model()` and returns `effective_model` in the response
- `messages.js` applies `effective_model` back to the UI/localStorage/dropdown if set
Closes#748
## Tests
1498 passed (2 pre-existing ordering failures unrelated to this PR; 5 new tests added in `test_provider_mismatch.py`).
**Original author:** @likawa3b
## Summary
Rebased-on-behalf of @franksong2702 (originally PR #728 — had CHANGELOG conflict after #747 merged).
Moves `#cmdDropdown` from outside `composer-box` to inside it, so the `position:absolute` anchor is scoped to the composer width rather than the full chat panel. CSS updated to use `bottom:calc(100% + 4px)` and `width:auto;max-width:100%` for clean upward positioning.
Closes#633
## Changes
- `static/index.html` — moved `cmd-dropdown` div inside `composer-box`
- `static/style.css` — updated `.cmd-dropdown` positioning (remove `margin-bottom`, use `bottom:calc(100% + 4px)`, add `width:auto;max-width:100%`)
- `tests/test_sprint50.py` — 2 new structural tests verifying DOM position and CSS rules
## Tests
1493 passed, 1 warning (2 new tests added)
**Original author:** @franksong2702
fix(ui): restrict edit to latest user message (#747)
Only the latest user turn shows the pencil/edit affordance. Older user
messages remain read-only (copy + timestamp still work). Avoids the
misleading implication that historical messages can be lightly edited
when the actual action truncates the session and restarts the
conversation from that point.
Closes#744
Co-authored-by: franksong2702 <franksong2702@users.noreply.github.com>
Strips <function_calls> XML from assistant messages before rendering, adds workspace file panel empty-state messages, and changes notification description from 'tab' to 'app'. 16 new tests. Fixes#702, #703, #704.
Fixes the /compress reference card showing only a short 3-line summary immediately after compression. Now prefers the persisted compaction message (full handoff) over the raw API summary, matching what is shown after page reload. Closes#695.
MiniMax M2.7/highspeed added to _FALLBACK_MODELS. MINIMAX_API_KEY and MINIMAX_CN_API_KEY added to env scan tuple so os.environ is checked. 11 tests. Independent review by @nesquena confirmed correct, needed rebase only.
Providers in config.yaml with explicit models: list were silently ignored. Fix extends the model-list builder to check cfg.providers[pid].models, covering both dict and list formats. Also includes providers only in config.yaml (not _PROVIDER_MODELS). 5 regression tests added. Independent review by @nesquena.
DEFAULT_MODEL now defaults to "" instead of "openai/gpt-5.4-mini". Guards added in model-list builder so empty default does not create blank model entries. Adds 3 tests in test_issue646.py. Independent review by @nesquena.
Fixes <|turn|>thinking delimiter (was wrong as <|turn>thinking) in api/streaming.py, static/messages.js, and static/ui.js. Adds 13 regression tests. Independent review by @nesquena.
Independent review by @nesquena confirmed all blockers resolved. Theme×skin two-axis system replaces old monolithic color schemes. Closes#627. Co-Authored-By: aronprins <aronprins@users.noreply.github.com>
Fixes the root cause of OPENROUTER_API_KEY being overwritten with test-key-fresh on every pytest run.
Three-layer fix:
1. Unit tests: mock _get_active_hermes_home in TestApplyOnboardingSetupGuard so .env writes land in /tmp, never ~/.hermes
2. Test server subprocess: add HERMES_BASE_HOME=TEST_STATE_DIR to hard-lock profile resolution inside the server process
3. Test server subprocess: strip real provider keys (OPENROUTER_API_KEY etc.) from the inherited env before server starts
Reviewed and approved by @nesquena. 1373 passed, 0 skipped.
Squash-merges PR #614. Fixes Docker 500-on-every-request crash from PermissionError in load_settings() (issue #570 follow-up).
Both SETTINGS_FILE.exists() call sites now catch OSError and fall back to defaults. Reviewer nits addressed: removed unused imports/var in tests, improved log message to say "inaccessible?" instead of "permission denied?". Rebased clean onto v0.50.73. 1373 tests passing, QA harness green.
Squash-merges PR #611 (@franksong2702). Fixes two edge cases in auto-generated session titles.
1. Strip Markdown labels (`**Session Title:**`, `Title:`) from sanitizer output — these were being persisted verbatim when the LLM emitted them.
2. Skip empty assistant tool-call placeholder messages when extracting the first exchange for title generation — previously the empty row could be latched onto instead of the first real answer.
Also tightens the title prompt to explicitly forbid Markdown, bullets, and label prefixes.
1371 tests passing, QA harness green.
Co-authored-by: Frank Song <franksong2702@gmail.com>
