delegation.max_iterations is the per-subagent tool-call budget. The old
default of 50 truncated substantial delegated work: leaf agents spend
~15-20 turns on reconnaissance before producing output, then ran out of
budget mid-task and returned 'completed but unfinished' summaries. 250
gives real delegated work room to finish.
Changes:
- config_defaults.py: delegation.max_iterations 50 -> 250; _config_version 35 -> 36
- tools/delegate_tool.py: DEFAULT_MAX_ITERATIONS fallback 50 -> 250 (kept in
sync with the shipped default to prevent drift)
- config_migrations.py: _migrate_to_36 lifts configs still pinned at exactly
the OLD default 50 -> 250 on update, so existing installs inherit the new
headroom. Any other explicit value (deliberate override) is preserved;
unset inherits 250 at read time.
- cli-config.yaml.example: doc the new default
The cap is per-child and children run concurrently (max_concurrent_children
default 3), so this raises worst-case fan-out cost; delegation.child_timeout_seconds
(default 0 = off) remains available as a wall-clock guardrail, and users can
still pin a lower max_iterations explicitly.
Verified: migration lifts 50->250, preserves a deliberate 120, leaves unset
untouched (3/3); DEFAULT_CONFIG reads version=36, max_iterations=250, fallback=250.
Per-provider ssl_ca_cert / ssl_verify reached the httpx chat client and the
auxiliary clients (#56681), but the endpoint discovery and pricing probes did
not. Both probe families resolved TLS from process-wide env vars only:
- the requests-based metadata/pricing probe
(agent/model_metadata.py::_resolve_requests_verify)
- the urllib-based /models catalog probe
(hermes_cli/models.py::probe_api_models)
A custom endpoint whose chain verifies against the provider's configured
bundle, but not the process SSL_CERT_FILE, then logged a spurious
CERTIFICATE_VERIFY_FAILED on every probe even though the chat path worked.
Pointing a global CA env var at the bundle fixes it but changes verification
for every provider, defeating the point of a per-provider setting.
This threads the selected provider's TLS settings into both probe paths,
reusing get_custom_provider_tls_settings so there is no second precedence
chain:
- _resolve_requests_verify(base_url) looks up the provider's ssl_verify /
ssl_ca_cert before falling back to the env vars. Callers with no base_url
keep the exact env-only behavior.
- probe_api_models builds an ssl.SSLContext from the provider settings and
passes it through open_credentialed_url, which gains an ssl_context seam on
the cloned secure opener. Unmatched or public endpoints pass None and keep
urllib's default policy.
Tests: tests/agent/test_custom_provider_ca_probes.py covers both probe
families (provider CA, ssl_verify:false, unmatched, missing file, config
lookup failure) plus end-to-end assertions that the resolved verify value and
SSLContext actually reach the request seam. Verified against the neighboring
metadata, pricing, TLS, and urllib-security suites (266 tests) with no
regressions.
Earlier releases accepted api_mode: openai on custom provider entries.
The canonical transport set is now {chat_completions, codex_responses,
anthropic_messages, bedrock_converse, codex_app_server}, and an
unrecognized value was silently ignored at both consumption sites
(_normalize_custom_provider_entry passes the raw string through and
agent_init's accepted-set check drops it; _parse_api_mode returns None),
falling through to hostname-based detection.
For hosts with a detection rule the provider silently switches
transports after an update. Observed live: a custom entry for
api.actual.inc with api_mode: openai (valid when written) flipped to
codex_responses via the hostname rule, and every reasoning-bearing
request to the relay's /v1/responses failed with a wrapped non-JSON
error while /v1/chat/completions worked throughout.
Fix: one shared alias map (_canonical_api_mode) consulted by both
sites. openai/openai_chat -> chat_completions, responses ->
codex_responses, anthropic/messages -> anthropic_messages, bedrock ->
bedrock_converse. Canonical names and unknown values pass through
unchanged, so invalid-config behavior is untouched.
Tests: alias map contract (every alias lands in _VALID_API_MODES),
normalizer canonicalization incl. the transport: key alias, and the
runtime gate accepting legacy spellings while still rejecting unknowns.
* feat(dashboard-auth): extend RFC 8252 native sign-in to password providers
The desktop app runs password sign-in for gated gateways in an embedded
Electron BrowserWindow, where OS password managers (macOS Passwords /
iCloud Keychain autofill) cannot reach the form — Chromium-in-Electron
has no bridge to them, so users retype credentials by hand even though
the /login form already carries the right autocomplete attributes.
The existing RFC 8252 native flow (system browser + loopback + PKCE)
solves exactly this for OAuth providers, but was explicitly disabled for
password providers on the grounds that they have "no IDP round trip to
broker". The brokering is still worth having: it moves the credential
form into the system browser, where password-manager autofill just works.
Gateway-only change; the desktop needs no changes (runNativeLogin is
already page-agnostic), and older desktop builds pick the capability up
automatically once the gateway advertises it:
* /auth/native/authorize now accepts a supports_password provider:
register the pending broker authorization as usual, then 302 the
system browser to the interactive /login form with the opaque
broker_state in the gateway's PKCE cookie (the same server-controlled
channel the OAuth branch uses) instead of an IDP redirect.
* /auth/password-login: when the server-set PKCE cookie carries a
broker handle, a successful credential check completes the pending
authorization exactly like the /auth/callback native branch — mint
the one-time loopback code, return the loopback redirect (validated
loopback-only at authorize time) as `next`, clear the PKCE cookie,
and set NO session cookies. A lapsed broker is a clean 400 telling
the user to restart sign-in; a failed credential attempt leaves the
pending entry intact so the user can retype.
* /api/status now advertises "native_pkce" whenever any interactive
session provider is registered (previously only for non-password
providers), so the desktop selects the system-browser strategy for
password-only gateways.
Security posture is unchanged from the existing flow: loopback-literal
redirect_uri enforcement, PKCE S256 binding, single-use short-TTL codes,
constant-time comparison, and the same rate limiter on password attempts.
Tests: full authorize → /login → password-login → loopback → token →
bearer round trip, wrong-password keeps the pending entry, lapsed broker
→ 400, no-broker browser login keeps minting cookies, and the /api/status
advertisement for password-only gateways.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(dashboard-auth): bind native password completion to the authorize-time provider
Review follow-ups for #75808:
* /auth/password-login now enforces that body.provider matches the
provider recorded in the server-set PKCE cookie by
/auth/native/authorize before completing a pending native
authorization. /login renders a form for every session provider, so
without this a native flow started for provider A could be completed
with provider B's credentials, binding B's session into A's pending
entry. The mismatch is rejected BEFORE credential verification (no
session minted, no oracle) and preserves both the pending entry and
the cookie, so the user can still submit the correct provider's form.
Covered by a two-password-provider E2E regression test.
* Update the two docs spots that still said password-only providers do
not advertise native_pkce (website desktop-native-signin guide and the
auth_flows type comment in web/src/lib/api.ts).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore: map contributor email for #75808 (buffpesos)
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Brooklyn Nicholson <brooklyn.bb.nicholson@gmail.com>
Everything Browser Use is now managed by Hermes: the canonical binary
is the one install_cli() provisions into HERMES_HOME/bin, and every
resolution and provisioning site prefers it.
- _find_cli(): probe order flipped to managed bin -> PATH ->
user-level tool dir (then uvx across the same order). A user's own
uv tool install can no longer shadow the Hermes-managed copy with a
drifted version; side installs only matter when we have nothing.
- install_cli(): a browser-use on PATH no longer short-circuits the
install — only the managed copy does, so selecting any backend
provisions the copy Hermes controls and updates.
- _ensure_browser_use_cli() (hermes tools): drops its own PATH check
and always delegates to install_cli(), the single owner of the
managed-copy policy.
- install.sh / install.ps1: same short-circuit fix — only
HERMES_HOME/bin/browser-use counts as installed.
Follow-up to #86240 and #86320: with every non-Camofox backend
selection installing the CLI, managed-first closes the remaining
version-drift/shadowing class instead of guarding single sites.
Tests updated to pin the new contract: managed beats PATH and
user-local; PATH install does not satisfy install_cli; helper always
delegates. E2E-verified precedence chain with real files and a real
degraded-PATH install attempt.
Fixes#84851
The SSH fast path in _check_via_local_git compared only the exact tip SHA
of local HEAD against upstream main. When a local carried commit makes the
SHAs differ, it returned 1 ('behind') without checking ancestry, so an
ahead-of-origin checkout was misreported as '1 commit behind' — nudging the
user to run 'hermes update' and wipe their carried work. Fall back to
git rev-list --count HEAD..origin/main (mirroring the full-clone path) when
the SHAs differ.
Ports Claude Code's /loop (and its /proactive alias) across every Hermes
surface. /loop [interval] <prompt> re-runs a prompt or slash command on a
recurring cadence inside the live session; omitting the interval enables
self-paced mode (starts at the floor, backs off exponentially while the
agent's replies stop changing, snaps back on change — local digest
comparison, zero extra LLM cost).
Stop conditions: agent-emitted LOOP_COMPLETE marker, --times N,
--until <condition> (judged by the existing goal_judge aux task,
fail-open), /loop stop, and a loops.max_ticks backstop budget.
Core: hermes_cli/loops.py (LoopState + LoopManager + shared
dispatch_loop_command), persisted per session in SessionDB state_meta
(loop:<sid>) so /resume picks it up; migrates across compression
boundaries like /goal. New SessionDB.list_meta_prefix() powers the
gateway's cross-session scan.
Surfaces:
- CLI: /loop handler + idle-fire and post-turn-complete hooks in
process_loop (mirrors the /goal hook shape; Ctrl+C pauses the loop)
- Gateway: /loop handler with route capture, mid-run control-verb guard,
post-turn tick completion, and a supervised loop_wakeup_watcher that
injects due wakeups into idle chats via the synthetic-message path
- TUI/dashboard/desktop: command.dispatch handler + per-session
notification-poller wakeup driver + post-turn completion in the turn
dispatcher; /loop added to the desktop slash palette
- /goal mixing: an active non-parked goal owns the idle boundary — loop
ticks defer until it finishes, pauses, or parks; real user input always
wins over both
Config: loops.{min_interval_seconds,max_ticks,self_paced_floor_seconds,
self_paced_ceiling_seconds}. Docs page + sidebar entry. 77 new tests.
Slack's 50-slash cap: /version moves to /hermes version to free the
native slot for /loop.
The hermes update APPLY path still ran an unconditional
rev-list --count HEAD..origin/<branch> — on a depth-1 installer checkout
that walks the truncated graph and reports the entire remote ancestry
(#53479's 'Found 9980 new commit(s)' on Windows 11). The zero/nonzero gate
stays (a 0 count is trustworthy on any graph); when the count is positive
on a shallow repo, recover the real number via the GitHub compare API
(added in PR #86257) and print count-free wording when that fails.
ahead_by==0 (local-ahead) falls through to the up-to-date path.
Completes the class fix from PR #86257 on its last remaining site.
The honesty half (no fabricated counts) leaves shallow installs permanently
count-less. The compare API knows the full graph regardless of local clone
depth: GET /repos/<o>/<r>/compare/<current>...<target> returns ahead_by —
exactly the behind count the shallow boundary lost.
- hermes_cli/banner.py: _github_compare_behind() (bounded, unauthenticated,
best-effort); wired into _check_via_rev and the shallow branch of
_check_via_local_git. ahead_by==0 with differing tips = local-ahead => 0.
- hermes_cli/update_cmd.py: hermes update --check shallow path prints the
exact count when recoverable, presence-only wording otherwise.
- apps/desktop/electron/update-count.ts: compareApiUrl() +
parseCompareBehindCount() pure helpers; main.ts fetches the count when
resolveBehindCount() returns null, and the SSH-official passive path stops
fabricating behind:1 (uses compare API + updateAvailable flag).
- apps/desktop/src/lib/version-status.ts: updateAvailable now applies to the
client target too, so a shallow desktop install shows '(update)' instead of
nothing (or the old frozen '(+1)').
Fixes#84591; CLI siblings of #78253 / #53479 behavior.
E2E: live compare API returned 61/62 for real 61/62-commit gaps and 0 for the
reversed (local-ahead) pair; real shallow-clone fixture (depth-1 clone +
depth-1 fetch, merge-base broken) recovers the exact count with the API and
falls back to the honest sentinel offline.
The SSH-official-remote path in _check_via_local_git was hard-coded to
return 1 when _check_via_rev reported UPDATE_AVAILABLE_NO_COUNT, so
'hermes --version' and the CLI banner surfaced a stable but false
'Update available: 1 commit behind — run hermes update' message. The
count never grew: whether upstream was 1 commit or 100 commits ahead,
the banner always said '1 commit behind'.
Root cause: an ls-remote probe against the upstream URL can only tell
us tip SHAs, not a real commit count. Returning the sentinel
UPDATE_AVAILABLE_NO_COUNT (-1) already means 'update exists, count
unknown' — the exact right shape for this path.
The dashboard/desktop UI does not depend on the fabricated 1:
- The REST /api/hermes/update/check endpoint
(hermes_cli/web_server.py::check_hermes_update) treats any nonzero
behind as update_available=true, and its docstring explicitly
documents -1 as a legitimate value.
- The desktop store (apps/desktop/src/store/updates.ts::mapBackendCheck)
clamps behind<=0 to 0 and reads updateAvailable as a separate boolean
field.
So restoring the sentinel is a strict improvement: CLI banner and
hermes --version now say 'Update available' honestly instead of
inventing a count, and every REST/desktop consumer keeps working.
Changes:
- hermes_cli/banner.py: drop the 'return 1' override in the SSH branch;
propagate the sentinel unchanged.
- hermes_cli/main.py::_print_version_info: render the sentinel as
'Update available — run <cmd>' (without a count).
- tests/hermes_cli/test_update_check.py: update the SSH-official test
to assert on the sentinel; add 3 new tests covering the CLI
renderer's -1 / >0 / 0 branches.
The Browser Use CLI 3.0 is the primary driver engine for all browser
backends except Camofox, but only the explicit 'Browser Use' picker row
ran the install hook. Local Browser, Browserbase, Firecrawl, and the
Nous-managed cloud rows left the CLI uninstalled, so those selections
depended on the uvx zero-install fallback (first-use PyPI download
inside the tool-call timeout) or silently downgraded to the built-in
browser tools where uvx was unavailable.
- Extract the install logic into _ensure_browser_use_cli() and run it
from the agent_browser/browserbase post_setup branch too (Firecrawl
and the Nous cloud row both declare post_setup: browserbase).
- Camofox is untouched: Firefox-based, no CDP surface, cannot be driven
by the CDP-only browser-use harness.
- Failure stays non-fatal: uvx fallback, then built-in tools.
Tests pin the contract: every browser post_setup key except camofox
attempts the install; camofox never does; install failure never raises.
The 21600s (6h) default shipped in #77184 makes an interactive
'hermes gateway restart' block for up to six hours when a turn wedges
(hung tool call, wedged event loop, stuck provider stream) — the exact
scenario the cap exists for. The intent (don't force-kill an agent
mid-turn) is sound, but the default must be a safety valve for hung
agents, not a target latency.
Lower to 1800s (30 min): still protects the overwhelming majority of
long autonomous turns (tool calls have their own timeouts well below
that), keeps worst-case interactive restart latency human-tolerable, and
users running very long unattended turns can raise it in config.yaml.
RED: new contract test fails on old 21600 default. GREEN: 4/4.
* fix(dashboard): suppress Ctrl+C shutdown traceback
* fix(dashboard): extend clean Ctrl+C exit to the Windows serve branch
The Windows loop-factory branch (and its pre-0.36 asyncio.run fallback)
runs under the same uvicorn capture_signals() re-raise as the POSIX
path, so console Ctrl+C leaked the identical KeyboardInterrupt
traceback there. Guard both serve calls with the same clean-exit
contract, keeping the import-resolution try/except comment accurate
(genuine serve-time errors still propagate).
Also ports the reworded POSIX-test docstring (the serve path is no
longer 'byte-for-byte unchanged'), wraps the POSIX KI test in
pytest.fail so a regression reports red instead of aborting the pytest
session, and adds the windows_only sibling test.
Extends #52970 to the whole bug class.
* chore: map contributor email for @wangs1203
* test(dashboard): actually exercise the pre-0.36 Windows fallback KI contract
Copilot review caught that patching uvicorn._compat.asyncio_run with
raising=False makes the import succeed, so _runner is non-None and the
extra asyncio.run patch never covered the fallback. Split it out: a
dedicated windows_only test halts the _compat import (None in
sys.modules) so the fallback branch is genuinely selected, then asserts
the same clean-KI contract on bare asyncio.run.
---------
Co-authored-by: Emiya·Leon <wangs.coder@gmail.com>
Follow-ups on top of the salvaged #70324:
- _confirm_startup_expensive_model_override evaluates the unified
registry (combined_selection_warning) so id-keyed guards like the
data-training-tier warning fire at startup too, not just the cost guard.
- The Termux-adjacent light oneshot fast-path (added after the PR
branched) ran _run_and_exit_oneshot without the guard — same bug
class, third sibling site now covered.
Run the expensive-model warning for explicit startup `-m` / `--provider`
overrides before the chat loop starts, and fail closed for non-interactive
invocations that select an expensive or known-confusing model.
Also classify Nous paid-model 404s that say credits are required as billing
exhaustion so they fail fast with billing guidance.
Tested:
- scripts/run_tests.sh tests/hermes_cli/test_cli_startup_model_cost_guard.py tests/hermes_cli/test_model_cost_guard.py tests/agent/test_error_classifier.py -- --tb=short -q
Custom providers (custom:xxx) serve their own pricing; models.dev stores
OpenRouter prices for the same model ids. The cost guard fired on that
foreign pricing and blocked composer/CLI model switches on custom
providers with a wildly wrong warning (#54348).
expensive_model_warning now only trusts model_info/models.dev pricing
when the provider maps to a models.dev provider and the info's
provider_id matches, and only consults the pricing-entry lookup when
the billing route is known. Salvaged from #54422; the PR's desktop-hook
half predates the use-model-controls rewrite and is superseded by the
hook's existing rollback handling.
Adds hermes_cli/model_selection_guards.py: a single evaluation point that
runs every selection guard (cost + the new data-policy guard) and returns
the warnings that fired. All seven model-selection surfaces (CLI picker,
cli.py TUI modal, gateway typed /model, dashboard web_server, TUI gateway,
Telegram and Discord pickers) now call the registry instead of importing
model_cost_guard directly — so the data-training-tier warning from
PR #81416 fires everywhere at once, and future guards need zero surface
wiring.
Guard modules keep their public APIs; existing mock patch points
(hermes_cli.model_cost_guard.expensive_model_warning) remain valid.
muse-spark-1.2-contributor is heavily discounted BECAUSE Meta uses your
prompts and completions to train future models. Selecting it for the price
without realising the data trade-off is a footgun.
Add hermes_cli/model_data_policy_guard.py (mirrors model_cost_guard):
data_training_warning(model_id, provider, base_url) -> DataTrainingWarning|None,
driven by a vendor-agnostic rule table. The status is not machine-readable on
/v1/models or models.dev, so the v1 rule keys on the documented '-contributor'
model id (fires regardless of provider, so it also covers custom/gateway
routes). Message mirrors Meta's pricing-doc language and figures
(https://dev.meta.ai/docs/pricing-rate-limits/).
Wire it into the CLI model picker's confirm flow (auth.py) as a [y/N]
disclosure, chained after the expensive-model cost guard. Fires only on the
contributor tier; silent on muse-spark-1.1/1.2 and all other models.
`_reap_unsupervised_gateway_orphans()` short-circuits on Linux hosts
with systemd via `supports_systemd_services()`, but returns `False` on
macOS — there is no systemd. This means the orphan reaper runs
unconditionally on macOS and treats the launchd-managed gateway as an
unsupervised orphan, SIGTERM-ing it.
When Hermes Desktop opens, `hermes serve` calls this function during
startup (web_server.py line ~251, gated on `HERMES_DESKTOP == "1"`).
The launchd gateway is killed, launchd restarts it via `KeepAlive: true`,
and the user sees a spurious gateway restart every time they reopen the
Desktop app.
Fix: exclude PIDs managed by launchd (`_get_service_pids()` already
returns launchd-managed PIDs on macOS) from the orphan scan, the same
way systemd PIDs are excluded on Linux.
Tested on macOS 26.5 (Tahoe) with Hermes Desktop 0.16+ and a
launchd-managed gateway (`ai.hermes.gateway` plist with `KeepAlive`).
Before the fix, quitting and reopening Hermes Desktop restarted the
gateway every time. After the fix, the gateway stays running across
Desktop quit/reopen cycles.
Background process completions on messaging platforms now default to a
one-line status message (✅/❌ + command + duration; failures append a
short output tail) instead of dumping the raw output buffer into the
chat. New display.background_process_notifications mode 'concise' is
the default; 'all' keeps the old raw-dump behavior for anyone who wants
it. Config migration v35 moves users still on the old implicit default
'all' to 'concise' on their next update; explicit result/error/off
choices are preserved.
* fix(desktop): load the intact renderer bundle when an update tears one copy
index.html and the hashed chunks it names are one generation. A packaged app
ships that bundle twice (inside app.asar and, via asarUnpack, beside it in
app.asar.unpacked), so an update that replaces the app while its files are
locked can leave the two copies from different generations. resolveRendererIndex
took the first index.html that existed, so it could pick the torn one and the
window died on its first lazy import with "Failed to fetch dynamically imported
module" -- with no way out, because every relaunch reloaded the same copy.
Check each candidate's declared modules and prefer a complete generation; when
both are torn, log which files are missing and how to repair instead of leaving
the crash unexplained.
* fix(cli): rebuild the desktop app when its renderer bundle is half-replaced
The content stamp hashes the SOURCE tree, which an interrupted update leaves
intact, so `hermes desktop` reported "up to date" and skipped the rebuild that
would repair a torn bundle -- the app relaunched into the same crash and
reinstalling looked like the only option.
Treat a bundle whose index.html names missing chunks as stale regardless of the
stamp, and say so on the way into the rebuild.
* fix(cli): --in accepts Git Bash / MSYS paths on Windows
Under Git Bash, 'hermes chat --in ~' reaches the CLI as /c/Users/<user>
(the shell expands ~ to an MSYS POSIX path; MSYS2 argument conversion
is disabled for native executables), and the isdir check failed with
'--in directory not found: /c/Users/...'. Route the value through the
existing _msys_to_windows_path translator (MSYS + Cygwin + WSL drive
spellings; no-op elsewhere) before expanduser/abspath.
Hit live: Bot Mode's agent-messaging protocol delivers with --in ~, so
every bot-to-bot send from a Git Bash-driven agent failed on Windows.
Tests pin both the translation cases and (source-level) the call site
actually using it.
* test: assert the MSYS translation, not platform abspath
The prior assertion ran os.path.abspath on the translated Windows path,
which on the Linux CI runner (posixpath) treats 'C:\Users\alice' as
relative and prepends the runner cwd. Pin the translation output and
ntpath absoluteness instead — same contract, platform-independent.
Stop freezing the xAI/xAI-OAuth catalog at import so /model and setup
pick up new Grok IDs after the models.dev cache refreshes. Put xai and
xai-oauth on the shared picker-time models.dev merge path and pin
grok-4.6 as the default headline model.
Extends the NS-656 memory-pressure surface to cover disk exhaustion
(OOF-2 / OOF-107 lineage: agents fill their data volume — SQLite writes
fail, sessions stop persisting — while every dashboard looks healthy).
- gateway/disk_status.py (new): collect_disk_status() samples
shutil.disk_usage(HERMES_HOME) and classifies pressure
(critical: <256 MB free or >=95% used; elevated: <512 MB free).
Never raises — degrades to pressure="unknown" with null telemetry,
same contract as collect_memory_status().
- /api/status: sibling `disk` block next to `memory`, advisory only —
not folded into component/overall health.
- web: DiskPressureStatus type; MemoryPressureBanner generalized to a
resource banner with worst-first triggers (disk critical > memory
critical > OOM restart > disk elevated > memory elevated) and
cascading dismissals — hiding the top trigger surfaces the next one
instead of silencing everything. All dismissals stay boot_id-scoped.
- i18n: diskCriticalBanner / diskElevatedBanner (en, optional fields
with English fallback per existing pattern).
Tests: gateway/test_disk_status.py (14), web_server disk-block
presence/degradation, banner disk trigger/priority/dismissal-cascade
suite (21 total).
Hosted agents can be OOM-killed hourly while the dashboard and the NAS
agent card both look perfectly healthy — every memory signal the gateway
already produces (heartbeat mem samples, lifecycle-ledger unclean-exit
verdicts, cache-pressure evictions) dies in server-side log files. The
BlueAtlas incident (NS-608) ran for three days like this.
This is the read-side fix:
* New gateway/memory_status.py distills the existing 30s loop heartbeat
(gateway RSS + system MemAvailable/MemTotal + swap) and the lifecycle
sentinel into a compact `memory` block: pressure ok/elevated/critical/
unknown, coarse MB numbers, and last-boot unclean/suspected-OOM flags.
Pure file reads, no new sampling, no gateway IPC. Stale (>150s) or
future-dated heartbeats degrade pressure to "unknown" so a dead
gateway's final gasp can't render a live "critical" banner forever.
Critical thresholds mirror the ledger's OOM-suspicion heuristics: if a
level would make a later unclean death "suspected OOM", warn at that
level while the process is still alive.
* lifecycle_ledger.record_startup now carries prior_unclean_exit /
prior_suspected_oom onto the reclaimed sentinel — previously the
verdict survived only in append-only diag prose. Flags age out on the
next sentinel rewrite (scoped to the life after the crash).
* /api/status serves the block (profile-aware, executor-offloaded,
fail-safe to pressure=unknown). Deliberately NOT folded into
components/overall: memory pressure is advisory, and flipping overall
to "degraded" on it would page NAS's availability sweep for a
condition the eviction valve is already handling. Public-safety:
coarse numbers/enums/booleans only — same disclosure class as the
existing nous_session_valid field, added for the same NAS-sweep
audience.
* Dashboard: new MemoryPressureBanner (app-shell, next to
ProfileScopeBanner) with worst-first trigger precedence
(critical > suspected-OOM restart > elevated), per-trigger
session-scoped dismissal, and escalation re-opening past a dismissal.
i18n keys optional with English fallbacks, matching the
managingProfileBanner convention.
Tests: gateway/test_memory_status.py (classification bands, staleness,
clock skew, corrupt files, bool-is-not-int), lifecycle sentinel
carry-forward, /api/status contract (block always present, collector
crash degrades instead of 500), and 7 banner component tests.
NAS-side ingestion (agent-card notice + memory-tier upsell) ships
separately.
Refs NS-656; context: NS-608, NS-657, OOF-77.
- web_server CONFIG_SCHEMA: fold the one-field models_dev category
(models_dev.url) into the agent tab via _CATEGORY_MERGE, matching the
established pattern for single-field categories (slice 7,
test_no_single_field_categories).
- image_routing._lookup_supports_vision: pass allow_network=True to
get_model_capabilities. The vision-capability lookup runs when an
image actually needs routing (not per conversation turn), and the
#31179 text-only-main guard depends on catalog data — with the new
allow_network=False default a cold cache returned 'unknown', which
falls back to attempting the call and reintroduced the #31179
failure shape (slice 8, test_text_only_main_skipped_when_no_
aggregator). This preserves that path's historical
network-on-cold-cache behavior; the fetch stays 4h-TTL cached and
backoff-limited.
Harden the models.dev catalog refresh path (#35838) with three missing
pieces:
1. ETag conditional GET — every network request sends If-None-Match
with the last-known ETag (persisted alongside the cache file). A 304
Not Modified re-confirms the existing cache without re-downloading
the full ~2 MB registry. This makes the 4-hour TTL effectively free
to maintain.
2. No-network-on-hot-paths invariant — allow_network=False is now the
default for every query function called on the conversation hot path:
get_model_capabilities, get_model_info, lookup_models_dev_context,
_get_provider_models. These are called during vision routing, image
routing, cost-guard checks, and context-length resolution on every
turn — they must never block on the network. Interactive flows
(model picker, model switch) explicitly pass allow_network=True.
3. Mirror URL override — models_dev.url in config.yaml lets deployments
point at a self-hosted mirror without code changes. Follows the same
pattern as model_catalog.url.
Additional hardening:
- Cache TTL bumped from 1h to 4h (ETag makes refresh cheap)
- Corrupt/empty disk cache is rejected with a warning instead of being
served as {} and silently breaking provider/model resolution
- _validate_registry() guards against non-dict and empty-dict payloads
Fixes#35838
Two bugs caused resumed sessions to use the config default model instead
of the model the session was actually using:
1. CLI /model switch didn't persist the new model to the session DB row.
The gateway calls update_session_model() after a /model switch, but
the CLI path only updated in-memory state and the agent's runtime —
it never wrote the new model to the sessions.model column. So the DB
row always kept the original model from session creation.
2. Resume didn't restore model/provider from the session DB row.
_preload_resumed_session and _init_agent restored CWD and YOLO from
session_meta, but never read session_meta['model'] back into
self.model/self.provider. So even if the DB had the right model,
resume would use whatever was in config.yaml.
Fix:
- _handle_model_switch / _apply_model_switch_result: call
update_session_model() after a session-scoped /model switch (skipped
for --once and --global), mirroring the gateway's behavior.
- New _restore_session_model() method: restores model/provider from
session_meta on resume, with provider/base_url/api_mode from
model_config.gateway_runtime. Also swaps the running agent in-place
for mid-chat /resume.
- Call _restore_session_model() from all three resume paths:
_preload_resumed_session, _init_agent, and _handle_resume_command.
- Track _explicit_model_override flag so -m/--model on the CLI overrides
resume (user intent wins). Cleared on /new.
Review follow-ups on the model_overrides feature:
- ONE canonical override schema everywhere. get_model_info previously
merged the override dict raw into the models.dev catalog shape
({**raw, **override}), so the documented context_window/supports_*
keys silently did nothing on that path (cost guard, inventory) while
working in capabilities/context paths — same config key, two
incompatible schemas. Overrides are now translated into the catalog
shape at the get_model_info boundary (_override_to_catalog_shape),
and sub-dicts (limit, modalities) are MERGED, not clobbered — an
override setting only context_window no longer wipes the catalog's
limit.output.
- _default is now a FILL-GAP default, not an override: it applies only
to models the catalog does not know (the #8731/#84482 self-unblock
path) and never displaces catalog data. A
_default: {context_window: 128000} can no longer clamp every model
of a provider. Explicit per-provider+model entries keep their
win-over-catalog semantics.
- Early-chain _override_context_window (model_metadata step 0b) is
explicit-only, so a _default can never preempt custom_providers
per-model settings or live probes; fill-gap defaults apply at the
lookup_models_dev_context catalog-miss boundary (step 5f) instead.
This fixes the precedence inversion where a provider/global _default
silently overrode an explicit per-endpoint per-model context_length.
- Provider keys accept BOTH id spaces (Hermes id and models.dev id:
copilot/github-copilot both work) and model ids match
case-insensitively, mirroring catalog lookup.
- Malformed override values (context_window: '512k') log a one-shot
warning instead of being silently swallowed.
- DEFAULT_CONFIG comment: removed the false family/dated-snapshot
inheritance claim, documented the recognized field list, fill-gap
semantics, and the id-space rule.
- Tests: rewritten for the new contracts (fill-gap invariants,
dual-id-space keys, sub-dict merge preservation, one-shot warning);
added a real-config-yaml e2e plumbing test (mutation-checked: fails
when the config key wiring is broken).
Add a unified model_overrides config section that lets users manually
declare context_window, max_output_tokens, capabilities, cost, and
family for any provider+model — winning over models.dev, OpenRouter, and
hardcoded defaults.
Resolution order (first hit wins):
1. model_overrides.<provider>.<model_id> (per-provider+model)
2. model_overrides.<provider>._default (per-provider default)
3. model_overrides._default (global default)
4. Normal catalog resolution
Key subtlety: an unknown model id (not in the
catalog) derives base metadata from sensible defaults before patching,
so overriding a model the catalog doesn't know yet is the supported
self-unblock path. This is exactly the #84482 scenario (Upstage
solar-pro4/syn-pro wrong context) and the #8731 scenario (custom/local
models with manual capability declaration).
Wired into:
- get_model_capabilities() — patches capability fields; unknown models
get safe defaults (tools on, vision/reasoning off) before patching
- lookup_models_dev_context() — context_window override, checked before
catalog lookup so it works even for providers not in PROVIDER_TO_MODELS_DEV
- get_model_info() — merges override dict onto catalog entry (shallow
merge); for unknown models, the override is the sole source of metadata
- get_model_context_length() — step 0b in the resolution pipeline,
before custom_providers (0c) and before any network probe
Config example:
model_overrides:
upstage:
solar-pro4:
context_window: 524288
syn-pro:
context_window: 65536
custom:my-local-vllm:
my-llava-model:
context_window: 8192
supports_vision: true
supports_reasoning: false
supports_tools: true
_default:
context_window: 128000
Fixes#8731Fixes#84482
Refs #47247
- Log (debug) instead of silently swallowing capability-lookup failures in
anthropic_prompt_cache_policy — a swallowed failure would otherwise
downgrade an explicit prompt_caching: true to (False, False) with zero
trace. Matches the sibling MoA branch's logger.debug style.
- Use load_config_readonly() for the None-fallback in
get_custom_provider_model_capability: the helper only reads, and the
fallback fires on the blank-stub paths (agent init before
_custom_providers is assigned, MoA/auxiliary destination planning), so
skip the ~135us defensive deepcopy per call.
- Add route-isolation regression tests at both levels (config helper +
agent policy): a prompt_caching declaration for one provider route must
never apply to another route with the same model name. Mutation-checked:
both tests fail when the URL match is disabled.
Now that subscriptions survive `done` (completion is reversible —
on every 5s notifier tick forever. Add
kanban_db.purge_stale_done_notify_subs(): one DELETE removing subs
whose task has been done with no new events past a retention window
(age = latest task event, falling back to completed_at/created_at, so
any activity exempts the task; a reopened task is exempt by status
alone). The notifier watcher runs it per board once at startup and at
most hourly, re-reading kanban.done_sub_retention_days (config.yaml,
default 30; 0 disables) at each sweep.
Builds on the three salvaged commits: adds the sources and integration points
they leave out, so a pip-installed memory provider is not a second-class
citizen next to a directory install.
Discovery
- Project-local providers (./.hermes/plugins/<name>/), gated on
HERMES_ENABLE_PROJECT_PLUGINS exactly as PluginManager gates its own project
scan. Completes the four sources CONTRIBUTING.md and AGENTS.md already
promised; memory was the only discovery system missing two of them.
- find_provider_dir() now resolves a package entry point to its directory.
This is load-bearing: config_schema.py (the dashboard panel) and cli.py (the
`hermes <provider>` subcommands) are read from disk rather than imported, so
without a directory a pip-installed provider silently lost both.
- list_memory_provider_names() includes entry-point providers, so they appear
in the dashboard's memory.provider dropdown.
Resolution stays import-free. hermes_cli.plugins.resolve_module_origin() is
extracted from _resolve_module_source() (added by the salvaged #76567) and
shared, so discovery walks a module's file layout instead of importing it.
find_provider_dir() is called from the dashboard and from argparse setup, long
before the operator has chosen a provider — importing every installed candidate
would execute third-party code on the strength of a package being present.
A test asserts the resolution leaves no side effects and no sys.modules entry.
Registration
- PluginContext gains register_memory_provider(). Memory was the only provider
category without one; context engine, image gen, video gen, web search,
browser, TTS, transcription, secret source, dashboard auth and platform all
have one.
- _ProviderCollector delegates unknown register_* calls to a real
PluginContext instead of carrying three hand-written no-ops. It silently
dropped register_tool/register_hook, and had no register_auxiliary_task at
all — despite PluginContext.register_auxiliary_task documenting a memory
provider (hindsight's pre-retain dedup) as its worked example. It can no
longer drift behind PluginContext.
- A raise after register_memory_provider() no longer costs the provider. The
loader caught it into a debug log, discarded the registered instance, and
fell through to "instantiate any MemoryProvider subclass" — returning a
different, unconfigured provider. A silent downgrade that looked like
success, and the exact outcome of calling register_auxiliary_task.
Activation is unchanged: still gated on memory.provider naming the plugin, and
covered by a test so the real PluginContext cannot start requiring
plugins.enabled — that would break every existing user-installed provider.
Verified end to end against a real third-party provider (kainappsinc/elephant)
installed by pip alone, with no directory copy: it appears in the dropdown,
resolves its directory, loads with its tools, and renders its dashboard panel.
Closes#40101.
Documents and tests the routing contract the sweeper review asked about:
classification records the manifest but does not activate anything.
- model-provider test now exercises providers.get_provider_profile() against
the pip-only name (None today — providers discovery is directory-based)
and asserts the module never leaks into sys.modules via that path.
- new test for the mnemosyne shape: a pip entry point duplicating a
same-name directory provider. The pip copy is classified exclusive and
never imported; the directory copy still activates through
plugins.memory discovery, exactly once.
- _classify_entrypoint_kind docstring now states the activation contract
explicitly: pip-only providers were equally unactivatable pre-change
(both destination systems are directory-only; the
hermes_agent.memory_providers entry-point group has no consumers), so
classification only removes the wasted import. Entry-point activation
is tracked upstream (#40644 for memory); this change is its
prerequisite, preventing double import once it lands.
find_spec() on a dotted module name imports the parent package first,
executing its __init__.py — which is exactly where a provider's heavy
imports typically live (fastembed -> onnxruntime and friends). The
previous classifier only preserved the no-import property for
top-level entry points.
_resolve_module_source() now resolves only the top-level name with
find_spec() (import-free for top-level names) and walks the remaining
dotted segments through submodule_search_locations by hand, mirroring
PathFinder's file conventions (part.py module / part/__init__.py
package). Namespace packages, zipped modules, extension modules, and
anything else unexpected fall back to standalone (the safe default).
.pyc origins map back to source via source_from_cache.
Regression: a dotted entry point whose parent __init__.py writes an
execution marker and imports the child — asserts the parent never
executed and neither module enters sys.modules during classification.
Fails against the previous implementation (marker written), passes now.
Entry-point (pip-installed) plugins exposing register_memory_provider()
or register_provider() + ProviderProfile were treated as plain
standalone plugins and eagerly imported by the general PluginManager,
even though memory and model providers have their own discovery
systems and the module has no register() for the general manager to
call. The import registered nothing and paid the module's full import
cost in every Hermes process (a pip memory provider pulls fastembed ->
onnxruntime, ~60 MB RSS).
Entry-point manifests now get the same source-scan classification as
directory plugins via a shared _detect_kind_from_source() helper: the
module is resolved with importlib.util.find_spec (no import) and its
first 8192 chars are scanned for provider markers. Memory providers ->
kind=exclusive, model providers -> kind=model-provider; both are
recorded for introspection and skipped by the general loader.
Unresolvable or non-Python modules stay standalone (default behavior
unchanged).
Tests: an enabled pip entry-point memory provider is never imported;
a pip entry-point model provider routes to providers/ discovery.
A failed npm install during `hermes update` prints "Fix npm and re-run
`hermes update`" -- but re-running on a current checkout hit the
"Already up to date!" early return before the Node refresh, so the
repair advice could never work and node_modules stayed stale forever
(#77211).
The commit_count == 0 path now runs the Node refresh through
_repair_node_deps_on_current_checkout. _update_node_dependencies
self-gates on the lockfile hash, which is only recorded after a
SUCCESSFUL npm install (and re-trips when node_modules is missing or
the web toolchain never landed), so healthy installs pay one hash
check and nothing else; a previously failed install actually repairs.
A clean refresh pairs with the web build like every other call site;
a failed one surfaces the fix-npm hint instead of "Already up to
date!".
Fixes#77211.
Co-authored-by: RelaxJonh <RelaxJonh@users.noreply.github.com>
Co-authored-by: JonthanaHanh <JonthanaHanh@users.noreply.github.com>
Swaps the Google flash entry in the OpenRouter and Nous Portal curated
lists to the newly released gemini-3.7-flash (half the price of
3.6-flash: $0.375/M in, $1.875/M out per OpenRouter live metadata;
served on both endpoints, verified live). Also updates the OpenRouter
plugin fallback_models mirror and regenerates model-catalog.json.
Scoped to the two named providers: vertex/gemini/gmi curated lists and
aux defaults still carry 3.6-flash.
Teknium