One sniffer owns magic-byte container detection (Teknium's one-concept-
one-owner rule): the new tools/audio_container.py is used by
- gateway/platforms/base.py _sniff_audio_ext (inbound cache — PR #36166's
central sniffer, now covering AAC/ADTS, MP4-brand disambiguation, webm)
- gateway/platforms/signal.py _guess_extension (audio/AV branches
delegated; RIFF/WAVE fix from PR #50690 and M4A-brand fix from
PR #72490 now live centrally)
- tools/tts_tool.py _sniff_audio_container (outbound repair, PR #73072)
cache_audio_from_url inherits the sniff via cache_audio_from_bytes.
Adds tests/tools/test_audio_container.py covering every magic-byte type,
wrong-extension repair on the inbound cache, unknown passthrough, the
URL path, and Signal's delegation.
iOS Signal delivers voice notes as MP4-container AAC carrying an audio
ftyp brand ("M4A "). `_guess_extension()` returned ".mp4" for every
`ftyp` file regardless of brand, so those attachments were cached as
documents instead of audio and STT rejected the upload:
API error: Error code: 400 - Invalid file format.
Supported formats: ['flac','m4a','mp3','mp4','mpeg','mpga','oga',
'ogg','wav','webm']
Read the 4-byte brand at offset 8 and return ".m4a" for audio brands
("M4A ", "M4B ") so they satisfy `_is_audio_ext()` and route to
`cache_audio_from_bytes()`. Video brands (isom/mp42/avc1/qt) still
return ".mp4".
This mirrors the existing brand/form-type disambiguation already used
in this function for RIFF (WEBP vs WAVE) and for ADTS AAC vs MP3.
Verified against a real iOS Signal voice note: pre-fix the upload was
rejected with the 400 above; post-fix the same bytes transcribe
successfully.
Telegram sends voice notes as .oga (OGG/Opus). SUPPORTED_FORMATS listed
.ogg but not .oga, so transcribe_audio rejected every Telegram voice note
with "Unsupported format: .oga" before reaching any STT backend. Add .oga
and .opus to the allowlist, with a regression test.
The grant_spent notice fired for every subscription user with top-up
funds the moment their cap was reached and camped in the CLI/TUI status
bar and desktop toasts with no action to take — the account keeps
working off top-up. Remove it everywhere:
- agent/credits_tracker.py: drop grant_cond + the emit/clear block;
the dev fixture state now (correctly) produces no notice
- TUI: keep the turn-start clear of credits.grant_spent as back-compat
for older backends that still emit the key
- Desktop: drop the demo step and stale comment references
- Docs/config comments: remove grant-spent from credits_notices text
- Tests updated: policy/cold-start now assert the key never fires
Usage bands, depleted, and restored notices are unchanged; /usage still
reports the full balance breakdown.
Responds to hermes-sweeper review on #54717: existing tests only mocked
platform_registry.get(). Add a hermetic fake deferred-loader test that
runs real PlatformRegistry resolution → PluginContext.register_cli_command
→ argparse subparser/choices visibility, without Photon SDK imports.
Closes#54678
`hermes photon ...` could fail with argparse `invalid choice: 'photon'`
even when the bundled Photon platform plugin is present. Photon registers
its top-level CLI command from the platform adapter module via
`ctx.register_cli_command(name="photon", ...)`, but bundled platform
plugins are cheap-registered as *deferred* entries to avoid importing every
gateway SDK during normal startup.
On the unknown-top-level-command slow path, `discover_plugins()` records the
deferred loader but never imports the matching platform module, so the CLI
registration side effect doesn't run and `photon` stays absent from
`_cli_commands` — argparse then rejects it.
Fix: after `discover_plugins()` on that slow path, resolve only the deferred
platform whose name matches the first positional token (via
`platform_registry.get(name)`) before reading `_cli_commands`. This imports
exactly the targeted platform, leaving normal startup cheap (a bare `hermes`
or flags-only invocation has no positional token and touches nothing). The
resolution is best-effort: registry/import failures are logged at debug and
never crash startup.
Added 3 tests in tests/hermes_cli/test_startup_plugin_gating.py: resolves the
matching platform, ignores empty/None command, and swallows registry errors.
Fails without the fix (symbol absent).
- Extract the cmdk filter into exported rankSearchOption and cover it,
SearchableSelect selection/clear/placeholder, and ConfigField
searchable-schema routing with 12 vitest cases (sabotage-verified:
the backend schema test fails on unfixed main).
- Add searchPlaceholder/noResults/systemDefault strings to ja/ar/zh-hant
(zh + en came with the salvaged commits; defineLocale would have
fallen back to English otherwise).
- Add a backend invariant test: timezone ships as a searchable,
clearable select of sorted IANA ids with a UTC fallback.
- Add 'System default' clear option to SearchableSelect via clearLabel prop
- Add clearable flag to ConfigFieldSchema (schema-driven, not hardcoded)
- Add clearable: true to timezone schema override in web_server.py
- Fix CommandItem value for clear item: use clearLabel instead of '' so
cmdk can match it during search
- Fix backend: or ['UTC'] fallback for hosts without tzdata where
available_timezones() returns an empty set (not an exception)
- Add systemDefault i18n key (en, types, zh)
- Fix focus: use autoFocus on CommandInput instead of e.preventDefault()
- Fix filter: prioritize city segment match (return 2 for slash match)
- Fix handleSelect: always select, don't toggle-deselect
- Add aria-haspopup to trigger button
- Use defensive placeholder logic
The timezone field was a free-text input with no guidance on format.
Users had to know the exact IANA identifier (e.g. America/New_York)
to configure it. Replace it with a searchable combobox built on
Popover + cmdk Command — the same stack as Shadcn's Combobox.
Backend:
- Add `_timezone_options()` to web_server.py (cached at import time,
returns sorted zoneinfo.available_timezones() — ~598 identifiers)
- Add `"timezone"` to `_SCHEMA_OVERRIDES` with `type: "select"`,
`options`, and `searchable: true`
Frontend:
- New `SearchableSelect` component (Popover + cmdk Command)
— closed-world filterable dropdown for large option lists
- `ConfigField` routes to `SearchableSelect` when
`schema.searchable === true` (explicit opt-in, no threshold)
- Add `searchable?: boolean` to `ConfigFieldSchema` type
- Add i18n keys: `searchPlaceholder`, `noResults`
The `searchable` flag is deterministic — no existing field is
affected unless explicitly opted in. Future large-list fields
can adopt the same pattern by adding `searchable: true` to their
schema override.
The cap is shared across CLI, desktop/TUI and the messaging gateway, so the
surface that gets rejected is rarely the one holding the slots. The rejection
read "Hermes is at the active session limit (5/5). Try again when another
session finishes." while every slot was an idle desktop tab, which took
filesystem access to work out.
Name the holders in the message, and show slot usage plus each holder in
`hermes status`. Both are inert when max_concurrent_sessions is unset, which
is the default. The gateway's duplicate copy of the message now reuses the
shared helper.
An open chat window took a session-cap slot at session.create/resume time.
Every desktop tile paint and every background reconnect-resume opens one, so
on a websocket-flappy host they accumulated: five parked desktop tabs filled a
5-slot cap and locked the messaging gateway (which shares the cap) out for
fourteen minutes while running no agents at all.
A slot held that way is invisible everywhere. An unprompted draft has no DB row
and the sidebar filters it out with min_messages=1, so the only way to diagnose
it was reading runtime/active_sessions.json by hand.
Claim on the first turn instead, mirroring the lazy contract
_ensure_session_db_row already uses for the row itself. Capacity now means an
agent can run rather than that a window exists, and anything holding a slot is
something the user can see.
Also reclaim leases whose session skipped teardown. _prune_dead only fires when
the owning pid dies, and a dashboard/serve backend runs for days, so a leaked
lease was held until restart. The owning process reconciles against the leases
it still holds, which is exact and needs no heartbeat write on the turn path.
- openrouter + nous curated lists: replace google/gemini-3-pro-preview with
google/gemini-3.1-pro-preview as the sole Pro entry, and
google/gemini-3.5-flash with google/gemini-3.6-flash
- remove qwen/qwen3.7-plus and qwen/qwen3.6-35b-a3b from both lists
- regenerate website/static/api/model-catalog.json
- test fixture: swap qwen3.7-plus catalog-label fixture to qwen3.7-max
(must be a model present in the nous curated list)
Both new Gemini ids verified live on OpenRouter /api/v1/models and the
Nous portal /v1/models (1,048,576 ctx — covered by the existing 'gemini'
prefix in DEFAULT_CONTEXT_LENGTHS).
- Route _get_read_conn through _connect_tracked_db so per-thread
read-only connections are registered with the POSIX lock-safety
guard (connect_tracked), matching the writer and existing read-only
paths. Without this, byte-level probes of state.db could close() an
fd that cancels locks held by an untracked read connection.
- Convert _search_unindexed_gap, _run_trigram_search, CJK-bigram FTS
search, and get_meta to _read_ctx — these are pure SELECT queries
called from search_messages that were still taking self._lock,
defeating the PR's contention fix for those paths.
- Add @pytest.mark.requires_wal to the 5 tests that assume WAL is
active. Hermes disables WAL on SQLite < 3.51.3 (WAL-reset bug),
so these tests fail on the venv's SQLite 3.46.0 without the marker.
- Remove unused 'time' import.
The gateway shares ONE SessionDB across every agent, so every recall/browse
read (session_search discover/scroll/browse, memory prefetch, title resolve)
queued behind every writer flush on self._lock — one Python lock in front of
a WAL database that natively supports concurrent readers. Measured convoy:
a 0.23s FTS query stretched to 112s and a browse flush to 137s while 6-8
concurrent turns flushed hundreds of tool results.
Fix: under WAL, read-only methods (get_session, resolve_session_by_title,
list_sessions_rich, get_messages, get_messages_around, get_anchored_view,
search_messages) run on a per-thread mode=ro connection via _read_ctx(),
taking no lock at all. Fresh read transactions begin per statement, so
read-your-committed-writes holds for flush-then-search patterns. Non-WAL
(NFS DELETE fallback) or read-conn open failure keeps the legacy locked
single-connection path, remembered per thread to avoid per-query retries.
CI shard 4 caught 4 failures the local baseline diff missed (local env
pollution made them look pre-existing): _make_compressor() constructs
under a get_model_context_length mock but the lazy deferral (#32221)
pushed resolution past the with block, so the 96K window / 75% floor the
tests rely on never materialized. Resolve inside the mock.
Review-pass findings on the lazy-init deferral:
- No-op guard: the codex app-server usage callback assigns
compressor.context_length on EVERY response (same window each time).
The setter unconditionally invalidated the derived budgets, wiping
runtime corrections applied directly to threshold_tokens /
tail_token_budget (aux-context threshold sync) — those persisted on
main's eager init. Same-value assignment is now a no-op.
- Re-floor on genuinely new window: the setter invalidates budgets but
previously kept the stale threshold_percent, so a codex window switch
recomputed threshold_tokens from the new window with the old model's
floored percent. Re-apply the raise-only small-context floor from
_base_threshold_percent so percent and tokens derive from the same
window (guarded with getattr for object.__new__ test instances).
- Init log extracted to _emit_init_summary_once() and also fired from
the setter path, so a consumer assigning context_length before any
read no longer strands the startup line forever.
- threshold_tokens getter resolves the window into a local before
reading threshold_percent — correctness no longer depends on
left-to-right argument evaluation order.
- reasoning_timeouts: fix inaccurate 'tuples are immutable' comment
(the container is a list; safety comes from build-once-at-import),
document why the slug stays in the tuple.
Adds TestContextLengthSetterCoherence (3 tests): same-value assignment
preserves overrides; new-window assignment re-floors both directions.
With the selective prompt-cache copy (#57046), un-marked messages on the
decorated api_messages list share their nested content parts with the
persistent conversation history — the per-message copy in
conversation_loop is shallow and decoration now deep-copies only the
marked messages. try_shrink_image_parts_in_messages previously wrote the
re-encoded image INTO the aliased part/source dicts, so an
image-too-large retry on an Anthropic route would silently replace the
original image bytes in agent.messages (and persist the degraded copy).
Replace the in-place writes with copy-on-write: rebuild the content list
with fresh part/source/image_url dicts and reassign msg['content'] — a
top-level write on the per-call copy that never reaches history.
Adds two regression tests simulating the aliasing; both fail against the
old in-place implementation (mutation-verified).
Baseline diff vs clean upstream/main (182 pre-existing environmental
failures on both sides) showed exactly 6 PR-caused failures, all the same
mechanism: tests construct ContextCompressor under a
get_model_context_length mock and read threshold_percent/threshold_tokens
after the with block, or build via object.__new__ and assign
context_length AFTER the derived budgets (the setter now resets the
lazily-cached budgets).
- test_compression_small_ctx_threshold_floor: resolve inside _make()
- test_cjk_token_estimation: resolve inside mock
- test_per_model_compression_threshold: resolve inside mock (2 tests)
- test_pre_compress_memory_context: assign context_length before
threshold/tail/summary budgets; add summary_target_ratio
TestThresholdTokensCap and TestLazyContextResolution landed on main after
the #38991 lazy-init base; they construct ContextCompressor under a
get_model_context_length patch and read threshold_tokens after the with
block. With deferred resolution the probe now fires lazily, so resolve
inside the mock (same pattern as the rest of the suite) and give the
lazy-resolution mock a real return_value.
Self-review found the deepcopy->shallow-copy change in
apply_anthropic_cache_control (#57046) had no test pinning the
"prompt caching is sacred / never mutate the caller's list" invariant.
The old test_returns_deep_copy only exercised the single marked message,
never an un-marked shared reference, so a regression that deep-copied too
little would pass the whole suite.
Add two tests: (1) caller list + every element left byte-identical after
the call, un-marked middle messages returned as shared references, marked
messages fresh copies, and mutating a returned marked message does not leak
upstream; (2) structural byte-equivalence vs a reference full-deepcopy
implementation across both native_anthropic modes and two TTLs.
Mutation-verified: neutering the per-message deepcopy makes test (1) fail.
The lazy-init change in #32221 deferred get_model_context_length() out of
ContextCompressor.__init__, but the "Context compressor initialized" log
(emitted only when quiet_mode=False) reads self.context_length,
self.threshold_tokens and self.tail_token_budget. Those property reads
resolve the deferred value, so the synchronous model-metadata probe still
ran inside __init__ on the interactive-CLI path — the exact blocking the
PR removed, just narrowed to the non-quiet path.
Emit the informative line once, on first context-length resolution, so
construction stays non-blocking on every path. Add a regression test
asserting no probe fires in __init__ with quiet_mode=False and that the
init log is emitted exactly once on first access.
_match_any() was re-sorting _REASONING_STALE_TIMEOUT_FLOORS (21 elements)
on every call. This function runs per API turn via error_classifier,
chat_completion_helpers, and thinking_timeout_guidance.
Also fixes thread-safety: the old _PATTERN_CACHE was a mutable dict
accessed from multiple threads without locking. Pre-compiling all
patterns at module load time eliminates both the per-call sort and
the TOCTOU race condition. The resulting list is effectively
immutable after import, safe for free-threaded Python 3.13+.
(cherry picked from commit e8b006b853)
Replaces full deepcopy with selective shallow copy in apply_anthropic_cache_control.
Only the 4 messages that receive cache_control markers are deep-copied; the rest
stay as references.
Measured on a 100-message conversation (typical long session):
- Before: ~15ms per call
- After: ~2ms per call
- 7.5x faster, scales with conversation length
Memory impact is also significant — no need to duplicate dozens of unchanged
messages on every turn.
The contract is unchanged: callers still get an independent message list
they can mutate. Only messages we modify (by injecting cache markers) are
copied. The rest are shared references to immutable history entries, which
is safe since the agent never mutates past turns.
(cherry picked from commit 17892df6cc)
The structural AST walker in test_compression_session_id_persistence.py
only recursed into control-flow children found via iter_child_nodes(stmt).
When a session_entry.session_id assignment lives inside an `else` block
whose statements are all assigns (no nested If/Try/etc to trigger
_walk_node), the assignment was invisible to the walker and the test
florped with 'No assignments found'.
Walk the stmt itself when it is a control-flow node so its
body/orelse/finalbody (and Try handlers) are always expanded, regardless
of whether iter_child_nodes yields an inner control-flow child.
Manual /compress already persists the rotated child transcript first and
only then repoints the live session_entry; a False rewrite_transcript
return keeps the entry on the original session_id so the conversation
stays reachable. Session hygiene auto-compress did the opposite: it
rebound session_id (and lease/topic) first, then called rewrite_transcript
without checking the return value. On a failed write the live entry
already pointed at an empty child SID and the turn continued — permanent
silent conversation loss. Persist first; rebind only after success.
Phase 2 review found two sibling sites with the same bug class:
- truncate_before_user_ordinal in prompt.submit counted display_kind
timeline rows as user turns, shifting the truncation target
- rollback.restore used the old pop-loop pattern that would pop a
display_kind marker instead of the last real exchange
Both now use the same predicate (role==user and not display_kind)
matching list_recent_user_messages, /undo, /retry, and CLI resume.
Added tests for both paths.
list_recent_user_messages and the in-memory /retry + session.undo walkers
treated every role=user row as a real user turn. Timeline bookkeeping
(model_switch, async_delegation_complete, auto_continue, hidden) is stored
that way, so /undo soft-deleted from a marker and /retry re-sent opaque
bookkeeping text. Exclude display_kind the same way CLI resume counting and
the prompt.submit ordinal path do.
Replaces the O(N) list_sessions_rich histogram in /api/sessions/stats
with a single GROUP BY query, reducing response time from ~575ms to
<1ms on large databases.
Original PR #48921 by @liuhao1024. Salvage fixes based on review
feedback from teknium1 and @wernerhp:
1. Preserve try/except guard — a DB error still degrades to empty
by_source instead of failing the whole stats response.
2. GROUP BY COALESCE(source, 'cli') — the original GROUP BY source
could emit duplicate 'cli' keys (NULL group + literal 'cli' group)
that the dict comprehension silently dropped.
3. Add exclude_children=True — list_sessions_rich excludes subagent
runs, delegates, and compression continuations by default; the
bare GROUP BY counted all rows, inflating source counts.
Aggregate shape (exclude_children/include_archived/limit params)
adapted from closed duplicate #61120 by @mijanx.
Closes#48914
Co-authored-by: mijanx <mijanx@users.noreply.github.com>
Every LLM call built a fresh openai.OpenAI wire client (new httpx pool,
TCP+TLS handshake, measured 19.2ms p50 / 35.5ms p95 per call at ~5 calls
per tool-loop turn) and closed it when the request finished. Cache ONE
reusable wire client on the agent, keyed by the effective client kwargs:
- _create_request_openai_client hands back the cached client when the
effective kwargs are identical; any change (credential rotation,
provider failover, vision default_headers) evicts and rebuilds.
- Only a request that produced a response reports a reuse close reason
(request_complete / stream_request_complete); error unwinds report
*_error_cleanup and really close, so a retry after a request error
always builds a fresh pool.
- Cross-thread aborts poison the slot: a pool whose sockets were
shutdown(SHUT_RDWR) from a stranger thread is never reused (#29507) —
the owner-thread close discards it and the next create rebuilds. The
holder read and the abort are atomic (under the holder lock) at all
three abort sites, so a late abort can never poison the NEXT request's
checked-out client.
- Worker-side interrupt breaks close the half-read SSE stream on the
owning thread before building the partial response; a failed close
poisons the slot (otherwise each interrupt leaked one checked-out
connection until the pool hit PoolTimeout). run_codex_stream gets the
same poison-on-close-failure handling.
- Single checked-out slot (in_use): a concurrent call gets an untracked
client with the old per-request lifecycle.
- release_clients() / close() really close the cached client when idle;
if a worker has it checked out they abort the sockets and detach the
slot, deferring the FD release to the worker's own close.
- MoA facade and Mock passthroughs never enter the cache; max_retries=0
is preserved on all request clients.
Follow-up to PR #64171. The _TOKEN_DELTA_* classification must exactly
cover update_token_counts' keyword surface: an unclassified kwarg is
silently kept only from the first delta of a merged run. Introspect the
live signature and fail with a pointed message when a future kwarg is
added without classification (or a classified field is removed).
Follow-up to PR #64171. The writer loop and flush_token_counts'
caller-drain both set _token_writer_busy BEFORE popping the queue —
that ordering is what makes flush's lock-free fast path (reads queue
then busy, no cond held) sound. _stop_token_writer's leftover drain did
it backwards (clear queue, then set busy), leaving a few-bytecode window
where a concurrent flush could observe 'empty and idle' and return True
with the popped batch still unapplied. Shutdown-only staleness, no data
loss — but the protocol now matches at all three drain sites.
Test: concurrent flush during a stop-drain mid-apply must time out
(False), never report drained.
Follow-up to PR #64171. Two writer-death hardening gaps:
- queue_token_counts only spawned the writer when the thread object was
None, so a writer that died from an unexpected exception could never be
replaced: deltas piled up on the uncapped deque until a reader's flush
drained them synchronously. Respawn on 'not thread.is_alive()' instead.
(atexit re-registration on respawn is safe: unregister removes all equal
bound-method registrations and the drain hook is idempotent.)
- _coalesce_token_deltas ran outside the per-delta try/except in
_apply_token_batch, so a merge bug (e.g. an unclassified future kwarg
summing None + 0) would escape and kill the writer thread. Wrap it and
fall back to applying the raw batch — coalescing is an optimization,
never load-bearing.
Tests: coalesce-failure fallback + dead-writer respawn.
