Three CI flakes hit while landing the credential-pool restore fix; all three
were timing/wall-clock races in the tests, not product bugs (each passes
locally and the assertions are correct):
- test_entire_tree_is_sigkilled_not_just_parent: _terminate_host_pid SIGKILLs
synchronously, but the test's 4s budget after a 1s in-function SIGTERM grace
left almost no slack for the kernel to tear down 3 processes + reparent the
children to zombies under loaded-CI scheduling. Widen the wait to 15s and
make the liveness predicate tolerant of vanished-pid / zombie races. The
assertion never weakens: every tree member must end up dead or zombie.
- test_session_resume_follows_compression_tip: appended messages got
time.time() timestamps (~now) while the test forced session started_at into
the past, so the get_compression_tip MAX(m.timestamp) tiebreaker depended on
wall-clock ordering. Pass explicit, well-separated message timestamps so the
chain resolution is deterministic by construction.
- test_non_retryable_exhaustion_arms_cooldown: asserted the short (5s)
exhaustion cooldown with a tight +1.0s slack, which false-fails when
wall-clock jitter between the 'before' snapshot and the cooldown computation
exceeds a second on a loaded runner. Widen to +30s — still cleanly below the
60s rate-limit window it must distinguish from.
The existing test_anthropic_stream_parser_valueerror_retries_before_delivery
asserted mock_replace.call_count == 1 — i.e. it passed precisely because the
buggy OpenAI rebuild was invoked on the Anthropic path. Repoint it to assert
the corrected close+rebuild-Anthropic behavior (#28161).
interruptible_streaming_api_call() has three connection-pool cleanup
sites that called _replace_primary_openai_client() unconditionally.
For api_mode=anthropic_messages this has two consequences:
1. _replace_primary_openai_client() fails (OPENAI_API_KEY unset on
Anthropic-only configs), so dead connections are never purged.
2. The stale-stream detector's outer-poll site (L1977) is the only
mechanism that can interrupt the worker thread while it blocks in
for event in stream:. Because the Anthropic client is never closed,
the thread stays blocked until the 900 s httpx read-timeout fires,
producing a visible 15-minute hang for Telegram/gateway users on
claude-opus-4-7.
Fix: mirror the existing interrupt-path pattern (L1989-1997) at all
three cleanup sites — if api_mode == "anthropic_messages", call
_anthropic_client.close() + _rebuild_anthropic_client() instead of
_replace_primary_openai_client(). _rebuild_anthropic_client() handles
both direct Anthropic and Bedrock-hosted Claude correctly, unlike the
inline build_anthropic_client() calls in open PR #14430.
PR #14430 (open) covers only the outer stale-detector site (L1977).
PR #23678 (open) covers only the inner retry sites (L1774, L1833).
This PR covers all three sites and uses _rebuild_anthropic_client()
for Bedrock parity.
Fixes#28161
Same bug class as the Anthropic fix (#26293): the OpenAI/aggregator client is
built without max_retries, so the SDK default of 2 applies. The SDK's own 1-2s
backoff ignores Retry-After and retries inside hermes's outer conversation loop,
burning request slots against a rate-limited bucket. Set max_retries=0 at the
single create_openai_client chokepoint (covers init, switch_model, recovery,
restore, request-scoped). auxiliary_client builds its own clients and is not
wrapped by the loop, so it keeps SDK retries.
The Anthropic SDK clients were built without max_retries, so the SDK
default (max_retries=2) retried 429/5xx with its own backoff that ignores
Retry-After — double-retrying inside hermes's outer loop and burning
request slots against a bucket that won't refill for minutes. Set
max_retries=0 on all Anthropic/AnthropicBedrock client constructions so
the outer conversation loop (which already honors Retry-After) owns retry.
Also raise the Retry-After cap in the conversation loop from 120s to 600s.
Anthropic Tier 1 input-token buckets reset in ~171s, so the 120s cap made
hermes retry before the reset window and re-trip the limit.
Refs #26293
A persistent upstream 401 on a single-entry OAuth pool (common for Claude
Max subscribers) made the credential-pool recovery spin forever:
try_refresh_current() re-mints a fresh token and reports success on every
401, so recover_with_credential_pool returned True and the retry loop
continue'd without ever incrementing retry_count or reaching the
auth-failover block. The configured fallback_model never activated and the
agent appeared to hang.
Cap consecutive successful same-entry refreshes (keyed by provider +
pool-entry id) at 2; once exceeded, treat the credential as unrecoverable
and return not-recovered so the loop falls through to
_try_activate_fallback. The 429/billing paths already rotate-or-fall-through
correctly (mark_exhausted_and_rotate returns None on a single entry), so
only the auth-refresh branch needed the cap.
Co-authored-by: Hermes Agent <hermes@nousresearch.com>
When every provider in the fallback chain fails non-retryably back-to-back
(e.g. HTTP 400/402/429 across distinct providers), the within-turn walk is
already bounded — _fallback_index advances monotonically and the loop aborts
when the chain exhausts. The damaging mode is cross-turn: restore_primary_
runtime resets _fallback_index=0 every turn, so a client that re-submits
immediately replays the entire chain, re-marshaling the full (potentially
80k-token) context once per provider every turn with no throttle on the
non-rate-limit path. On constrained hosts this exhausts memory/swap.
Rate-limit/billing failures already arm a 60s cooldown via _rate_limited_until;
the gap was the non-rate-limit case. Now, when the chain exhausts on a non-
rate-limit failure with a non-empty chain, arm a short (5s) cooldown on the
same _rate_limited_until gate (max(), never shrinking an existing window).
The next turn's restore stays gated and does NOT reset the index, so the
chain isn't replayed until the cooldown clears. No new state, no thread sleep,
no false-trip on legitimately long chains (those walk normally within a turn).
Tests: tests/run_agent/test_24996_fallback_exhaustion_cooldown.py
The agent's image-rejection fallback strips images and retries text-only when
a provider rejects image content, which is what lets the gateway drain its
queued messages. The fallback only fires on a hardcoded phrase list, and the
OpenRouter wording — HTTP 404 'No endpoints found that support image input' —
was missing. For OpenRouter-routed non-vision models the fallback never fired,
the retry loop re-sent the same rejected request until exhaustion, and every
subsequent message (including plain text) stayed queued behind the stuck turn.
Add the phrase to _IMAGE_REJECTION_PHRASES (the 404 already passes the 4xx
gate). Add a positive test and a guard test so the sibling OpenRouter
'no endpoints ... data policy / guardrail' 404s do NOT get their images
stripped.
Fixes#21160. Reported by @liu14goal14-ux; PR #21198 by @ygd58.
When a MoA preset is selected, each reference model's answer now renders in the
CLI as a thinking-style block labelled with its source model, BEFORE the
aggregator responds — so the mixture-of-agents process is visible instead of a
silent pause. The aggregator's response (and its tool actions) follow as normal.
Mechanism (shared seam, all surfaces):
- MoAChatCompletions/MoAClient take an optional reference_callback and emit
'moa.reference' (index/count/label/text) per reference, then 'moa.aggregating'
(aggregator label) once. agent_init wires this to the agent's
tool_progress_callback, which every surface already consumes — so the events
reach CLI/TUI/desktop/gateway with no new plumbing.
- CLI _on_tool_progress renders 'moa.reference' as a labelled '┊ ◇ Reference
i/n — <model>' header + a thinking-style preview (reusing _emit_reasoning_
preview), and 'moa.aggregating' as a spinner transition. Display-only; never
touches message history (cache-safe).
Turn-scoped reference cache: the agent loop calls the facade once per tool-loop
iteration, but the advisory message view is identical across iterations within a
turn, so references are now run AND displayed once per user turn (keyed by the
advisory view's signature) instead of re-running/re-spamming on every iteration.
This also cuts reference API cost from O(iterations) back to O(turns).
Verified live via interactive PTY on the opus-gpt preset (gpt-5.5 + opus refs):
reference blocks render once per turn, labelled by model, before the aggregator;
fresh blocks on each new turn; aggregator tool actions still execute.
Follow-up: TUI/desktop rich rendering + gateway batched-summary already receive
the events via tool_progress_callback; their surface-specific renderers are a
separate change.
MoA was calling reference and aggregator models through a bare
call_llm(provider=slot["provider"], model=slot["model"]) with a forced
temperature and a forced max_tokens (the preset's hardcoded 4096). That left
base_url/api_key/api_mode unresolved — so the auxiliary auto-detector guessed
the API surface instead of using the provider's real runtime, and the 4096 cap
truncated long aggregator syntheses.
A MoA slot is just a model selection and must be called the same way any model
is called elsewhere. Each slot is now resolved through resolve_runtime_provider
(the canonical provider→api_mode/base_url/api_key resolver the CLI, gateway, and
delegate_task all use) via a new _slot_runtime() helper, and the resolved
endpoint is passed into call_llm. So a reference/aggregator gets its provider's
actual API surface — MiniMax → anthropic_messages, GPT-5/o-series →
max_completion_tokens, custom endpoints → their base_url — identical to how that
model is handled as the acting model.
MoA also no longer imposes its own output cap: max_tokens defaults to None
(omitted → the model's real maximum) for references and is passed through from
the caller for the aggregator. The preset's hardcoded 4096 is gone. The
max_tokens preset config field is left in place (config/web/desktop unchanged);
it is simply no longer applied as a forced cap.
Tests: slots route through resolve_runtime_provider with resolved base_url/
api_key; resolution errors fall back to bare provider/model; neither call
carries an output cap even when the preset config still contains max_tokens.
When automatic fallback activates a provider that differs from the
primary, try_activate_fallback() cleared the primary's pool (to avoid
cross-provider base_url contamination, #33163) but never loaded the
fallback provider's own pool. The fallback then ran with no pool, so
rate_limit/billing/auth recovery couldn't rotate its credentials.
After clearing a mismatched pool, load_pool(fb_provider) and attach it
when it has credentials, so provider-specific rotation continues to
work on the fallback target.
switch_model() swapped model/provider/base_url/api_key but never
refreshed agent._credential_pool, which stays bound to the original
provider. recover_with_credential_pool() then sees a pool.provider !=
agent.provider mismatch and short-circuits — so a 429/401 on the new
provider gets no rotation and falls through to fallback instead.
Reload load_pool(new_provider) inside switch_model when the provider
changes (or the pool is missing). The reload is inside the protected
swap block and the pool is added to the rollback snapshot, so a failed
client rebuild restores the original pool.
Fixes#16678, #52727.
Follow-up to the #13971 fix: a genuine native Ollama provider reached
through a reverse proxy carries no ollama/:11434 URL signature, so the
restricted detection would miss it. Add provider=="ollama" as an
explicit True case (idea from #14789, @Tranquil-Flow) and cover both it
and the #13971 LiteLLM-proxy-to-zai false-positive with E2E tests.
The _is_ollama_glm_backend() function was too broad: any local endpoint
running a GLM model was treated as Ollama, triggering the stop->length
misreport heuristic introduced in 8011aa3. This caused false truncation
detection on sglang, vLLM, LM Studio, and other non-Ollama servers that
correctly report finish_reason.
When a GLM model on sglang/vLLM returned finish_reason='stop', the agent
mistakenly reclassified it as 'length' if the response didn't end with
a whitelisted punctuation character (ASCII or CJK). This particularly
affected Chinese-language responses and Markdown-formatted text.
Root cause: the is_local_endpoint() fallback assumed any local GLM
endpoint = Ollama. But many non-Ollama servers also run on localhost.
Fix: remove the is_local_endpoint() catch-all. Only detect Ollama via
its distinctive signatures (port 11434, 'ollama' in URL). All other
local servers are assumed to report finish_reason correctly.
This is the correct tradeoff because:
- False negatives (Ollama at custom port, heuristic not triggered) only
mean the user sees a truncated response — same as having no heuristic
- False positives (non-Ollama server, heuristic wrongly triggered) inject
spurious continuation messages into the conversation — strictly worse
Adds two tests:
- sglang GLM response is NOT reclassified as truncated
- Ollama GLM on port 11434 still triggers the heuristic as before
Co-authored-by: Hermes Agent <hermes@nousresearch.com>
Force redact_sensitive_text(force=True) on the browser_type text arg so
recognized credentials (API keys, tokens, JWTs) are masked in tool
progress, previews, callbacks, and return payloads even when the global
security.redact_secrets opt-out is set — a typed credential reaching chat
history is a security boundary, not log hygiene. Normal typed text matches
no pattern and stays fully readable for debuggability.
Tests assert the API-key-shaped secret is masked across every surface and
that normal text passes through unchanged.
Stopping a turn while the model is streaming (stop/esc to redirect) raised
InterruptedError, set final_response to the throwaway "waiting for model
response" sentinel, and persisted messages WITHOUT the assistant text that
was already streamed to the screen. The next turn then had no record of the
half-finished reply, so the model appeared to "forget" what it just said.
Recover the on-screen text from _current_streamed_assistant_text in the
InterruptedError branch and append it as the assistant turn (and surface it
as final_response). The metadata sentinel is kept only when nothing was
streamed yet, preserving the ACP/client suppression behavior.
Completes the partial-stream recovery from 397eae5d9 (which wired the same
_current_streamed_assistant_text salvage into the connection-failure twin
but missed the user-interrupt path). The lossy handler dates to c98ee9852.
* feat(moa): expose MoA presets as selectable virtual models
Reconstructed onto current main (PR #46081's base had diverged with no common
ancestor, marking the PR dirty so CI never dispatched). MoA is now a virtual
provider: each named preset is a selectable model under provider 'moa', and the
preset's aggregator is the acting model that answers and calls tools.
Reference models fan out in parallel via a bounded ThreadPoolExecutor (the same
batch pattern delegate_task uses) — all references dispatched at once, collected
when every one finishes, then handed to the aggregator. Output order is
preserved, failures and the MoA-recursion guard stay isolated per reference.
- Removed the old mixture_of_agents model tool and moa toolset.
- Added moa as a virtual provider in the provider/model inventory.
- /moa is shortcut behavior over model selection (default preset / named preset
/ one-shot prompt).
- Dashboard + Desktop manage named presets; presets appear in model pickers.
- Parallel reference fan-out in agent/moa_loop.py with regression test.
* fix(moa): thread moa_config through _run_agent to _run_agent_inner
The reconstructed gateway MoA wiring declared moa_config on _run_agent (the
profile-scoping wrapper) and used it inside _run_agent_inner, but the wrapper
never forwarded it — _run_agent_inner had no such parameter, so the runtime hit
NameError: name 'moa_config' is not defined on the compression-failure session
sync path. Add moa_config to _run_agent_inner's signature and forward it from
both wrapper call sites (multiplex and non-multiplex). Caught by
tests/gateway/test_compression_failure_session_sync.py on CI shard test(4).
* fix(moa): classify moa as a virtual provider in the catalog
The moa virtual provider has no PROVIDER_REGISTRY/ProviderProfile entry, so
provider_catalog() fell through to the default auth_type="api_key" with no
env vars — tripping two catalog invariants:
- test_provider_catalog: api_key providers must expose a credential env var
- test_provider_parity: every hermes-model provider must be desktop-configurable
moa already declares auth_type="virtual" in HERMES_OVERLAYS; consult that
overlay as an auth_type fallback so the catalog reports moa as virtual (no real
credential, no network endpoint). Exempt virtual providers from the desktop
parity union check the same way 'custom' is exempt — derived from the catalog,
not a hardcoded slug, so future virtual providers are covered too.
When an explicit aux provider cannot build a client before any request is
sent (missing raw env key, exhausted/unavailable OAuth or credential-pool
auth, resolver returning (None, None)), call_llm raised a misleading
"no API key was found" error and bypassed the configured fallback_chain
entirely. A provider authenticated through Hermes auth / the credential
pool (e.g. ollama-cloud) whose pool entry is exhausted hit this path, so
compression failed instead of routing to the configured fallback.
Adds _try_configured_fallback_for_unavailable_client() and wires it into
both sync and async call_llm before the raise, and into the startup
compression feasibility check.
Salvaged from #51835 by @herbalizer404.
These 7 test sites assert rotation behavior (fork, child sessions, lock
contention, logging session-context follows id rotation, boundary hooks fire
on rotation). Pin each builder to in_place=False explicitly so they keep
exercising the retained rotation fallback regardless of the global default
(flipped to True in #38763). Rotation stays a working opt-out fallback and
deserves continued coverage — these are NOT deleted.
Pinned sites:
- test_compression_concurrent_fork._build_agent_with_db
- test_compression_logging_session_context._build_agent_with_db
- test_compression_rotation_state._build_agent_with_db
- test_compression_boundary_hook._make_agent (2 helpers: CompressionBoundaryHook + SessionCompressEvent)
- test_compression_concurrent_sessions._build_agent_with_db
Make verification closure the default coding behavior after landed file edits while keeping bounded retries and config/env switches for users who need to disable it.
Add compression.minimum_context_floor config key that allows users
to lower the compression threshold floor below the hardcoded 64K
default, preventing infinite tool-call loops on models whose
structured output degrades well before 64K tokens.
- agent/model_metadata.py: add get_configurable_minimum_context()
helper with 16K hard safety limit
- agent/context_compressor.py: accept minimum_context_floor param,
thread it through _compute_threshold_tokens
- agent/conversation_compression.py: use compressor's floor for
aux model context validation
- agent/agent_init.py: read compression.minimum_context_floor from
config and pass to ContextCompressor
- gateway/run.py: cache-busting includes new key
Salvaged from #31686 by @Tranquil-Flow onto current main.
Resolves conflicts with in-place compaction (#38763) and max_tokens
threshold computation (#43547) that landed after the original PR.
Closes#31600
Adds auxiliary.background_review.{provider,model} (default auto = main chat
model — unchanged). Set it to a different, cheaper model and the post-turn
self-improvement review runs there for ~3-5x lower cost.
Cache-aware by design: the main chat is warm in the prompt cache, so the
default full-history replay on the main model is cheap cache reads — left
exactly as-is. A different model can't reuse that cache (different key), so
when (and only when) routed to a different model the fork replays a compact
digest instead of the full transcript, minimising what it cold-writes on the
aux model. Same model -> full replay; different model -> digest.
Quality holds in benchmarks: memory capture identical, skill near-identical.
Nothing changes unless you opt in by naming a different model.
Co-authored-by: Hermes Agent <noreply@nousresearch.com>
The success/staged gating and op-expansion for mirroring built-in memory
writes to external providers lived in a standalone agent/memory_write_bridge.py
helper called inline from two core call sites (tool_executor.py,
agent_runtime_helpers.py). That left the mirror decision-making in the agent
loop, outside the memory-provider interface.
Fold it into a new MemoryManager.notify_memory_tool_write() entry point: the
loop now hands over the raw tool result + args and a metadata callback, and the
manager decides whether/what to mirror. Both core call sites collapse to a
single call; the orphan module is removed. No MemoryProvider ABC change.
Tests rewritten as behavior tests against the manager method.
Mirror built-in memory writes to external providers only after the native memory tool succeeds and is not staged for approval. Keep OpenViking's built-in memory mirroring add-only, since Hermes native memory entries do not yet have stable OpenViking file URIs for replace/remove.
Add a narrow viking_forget tool for exact user memory file deletion and document the current OpenViking write/delete behavior.
Adds test_413_retries_on_token_only_compression: same message count but
materially fewer tokens after compaction must count as progress and retry,
not abort. Fails on main without the salvaged fix, passes with it.
* fix(agent): strip stale reasoning_content when falling back to a strict provider
A reasoning primary (DeepSeek/Kimi/MiMo thinking mode) pins reasoning_content
on every assistant tool-call turn (a single space " " pad). api_messages is
built once under the primary; on a mid-session fallback to a strict
OpenAI-compatible provider (Mistral, Cerebras, Groq, SambaNova), those stale
pads were replayed verbatim and rejected with HTTP 400/422:
body.messages.2.assistant.reasoning_content: Extra inputs are not
permitted (input: ' ')
reapply_reasoning_echo_for_provider() only ever ADDED pads, so it never
reconciled history built under a reasoning primary against a strict fallback.
copy_reasoning_content_for_api() also leaked empty-string and 'reasoning'-only
shapes to non-pad providers.
Fix both sites: when the active provider does not enforce echo-back, strip
reasoning_content (empty, space-pad, or non-empty) entirely. Re-padding when
switching TO a reasoning provider is preserved. Covers the Cerebras 400 from
#45655 and the DeepSeek->Mistral 422 fallback report.
Refs #45655.
* test: update reasoning-replay tests for strict-provider stripping
test_explicit_reasoning_content_beats_normalized_reasoning_on_replay was
implicitly running on the OpenRouter fixture (non-pad); pin it to a reasoning
provider so the precedence it checks is observable. Add a positive
strict-provider test asserting reasoning_content is stripped on replay.
The background-review fork (fires ~every 10 turns) pins
review_agent.session_id = agent.session_id — the parent's LIVE id — for
prefix-cache parity, then calls close(). With session finalization now in
close(), that would end the still-active parent session mid-conversation.
Set _end_session_on_close = False on the fork so the real owner (CLI close /
gateway reset / cron) finalizes the session instead.
Follow-up to the #12029 fix.
When the main provider is the Codex app-server runtime (api_mode
codex_app_server), the gateway showed no verbose 'running X' tool-progress
breadcrumbs on Telegram while every other provider did. The app-server
session processes item/started notifications (command execution, file
changes, MCP/dynamic tool calls) but never surfaced them as Hermes
tool-progress events — the session was constructed without an on_event
hook, so the agent's tool_progress_callback was never invoked on this
route.
Add _codex_note_to_tool_progress() mapping item/started → (tool_name,
preview, args) for commandExecution / fileChange / mcpToolCall /
dynamicToolCall, and wire an on_event hook into CodexAppServerSession that
forwards mapped events to agent.tool_progress_callback('tool.started',
...) — the same signature the chat_completions path uses (tool_executor.py).
Non-tool items (agentMessage/reasoning) and non-item/started methods map
to None and are ignored.
Co-authored-by: jplew <462836+jplew@users.noreply.github.com>
When the active provider returns a 401/403 that survives its per-provider
credential-refresh attempt (revoked OAuth, blocked/expired key, or an
account pinned to a dead/staging inference endpoint), the conversation
loop now escalates to the configured fallback chain instead of dead-ending.
Before: the generic failover dispatch fired only for {rate_limit, billing};
auth/auth_permanent fell through to 'switch providers manually' advice and
never called _try_activate_fallback(). A user whose primary credential was
broken kept thrashing on the same dead credential every turn — the main
agent appeared 'stuck in fallback mode' while never actually failing over.
This also affected auxiliary tasks (compression, vision, title-gen), since
auto-resolved aux follows the main provider.
After: a persistent auth failure with a configured fallback chain switches
to the next provider (mirroring the rate-limit/billing failover path),
guarded one-shot per attempt by TurnRetryState.auth_failover_attempted.
When no fallback is configured the behavior is unchanged — it falls through
to the existing terminal handling and provider-specific troubleshooting
guidance.
Tests: test_auth_provider_failover.py — 401/403 classify as auth, the
gating condition fires only with a chain present + guard unset, the guard
blocks repeats, and non-auth (500) errors do not trigger auth failover.
Follow-up to the soft-archive durability fix. Reusing the rewind/undo active=0
flag for compaction-archived turns inherited the wrong search semantics: undo
rows are intentionally HIDDEN from session_search (the user took them back), but
compaction-archived turns must stay DISCOVERABLE — that is the whole point of
Teknium's "searchable / recoverable" requirement. As built, search_messages
defaulted to WHERE active=1, so after in-place compaction the pre-compaction
turns were in the FTS index but filtered out of the default search. (The earlier
"searchable" claim only held for a raw FTS query / include_inactive=True, not
the actual session_search tool.)
Empirically confirmed the gap: search 'HMAC' returned 2 hits before compaction,
1 after (only the summary's mention) — the originals were hidden.
Fix — a `compacted` flag distinct from `active`, giving a 3-way state:
- active=1, compacted=0 → live context (normal)
- active=0, compacted=1 → compaction-archived: OUT of live context, IN search
- active=0, compacted=0 → rewind/undo: OUT of live context, OUT of search
Changes:
- messages.compacted INTEGER NOT NULL DEFAULT 0 added to SCHEMA_SQL. Declarative
_reconcile_columns adds it on existing DBs — no version bump (plain column add).
- archive_and_compact: UPDATE … SET active=0, compacted=1 (was active=0 only).
- search_messages: default WHERE active=1 → (active=1 OR compacted=1), on BOTH
the main FTS5 path and the trigram CJK path. include_inactive=True still
returns everything. The short-CJK LIKE fallback already returns all rows
(no active filter) — unchanged.
- Docstrings on archive_and_compact + search_messages document the 3-way state.
Verified: after compaction, session_search default finds the archived originals
(ids 1 & 4); rewind/undo rows stay hidden by default (recoverable via
include_inactive); live context still excludes both. 322 in-place + hermes_state
tests and 46 session_search tests green; ruff clean. Mutation check: reverting
the search WHERE to active-only fails the new searchable test.
(Surfaced by the question "is search semantic or only FTS?" — answer: session
search is FTS5 keyword/BM25 only, no embeddings over the transcript; semantic
retrieval lives in the optional memory-provider layer. Tracing that confirmed
the active-only filter gap above.)
Teknium review: keeping one durable session id must NOT come at the cost of
destroying history. The prior in-place implementation used replace_messages,
which hard-DELETEs the pre-compaction turns (they also drop out of the FTS
index) — same id, but the original conversation is gone with no recovery path
and the summary becomes the only record. Rotation today is non-destructive
(the old session's full transcript survives under the old id); in-place must
match that durability contract, not weaken it.
Fix: compact in place by SOFT-ARCHIVING, reusing the existing messages.active
flag (the /undo soft-delete mechanic), instead of deleting:
- New SessionDB.archive_and_compact(session_id, compacted): in one atomic
write, UPDATE messages SET active=0 on the live turns, then insert the
compacted set as fresh active=1 rows. Nothing is deleted.
- The insert loop is extracted into a shared _insert_message_rows() helper so
archive_and_compact and replace_messages don't duplicate the 60-line
column/encoding block (extend-don't-duplicate).
- Agent in-place branch calls archive_and_compact instead of replace_messages.
Durability outcome (proven by test + E2E across repeated compactions):
- Live context load (get_messages_as_conversation / get_messages) filters
active=1, so a resume reloads ONLY the compacted set — compaction still
shrinks the live session.
- The pre-compaction turns stay on disk at active=0, recoverable via
get_messages(include_inactive=True) / restore_rewound.
- They remain FTS-searchable: the messages_fts* triggers index on INSERT and
remove on DELETE only — they do NOT key on active, and active=0 is a
content-preserving UPDATE. session_search still finds them.
- Verified across TWO successive compactions: the 1st compaction's originals
are still recoverable + searchable after the 2nd (answers the "no recovery
path after the next compaction" concern directly).
message_count now reflects the LIVE (active/compacted) count, matching the
live load. replace_messages keeps its DELETE semantics (still correct for
/retry, /undo) and gains a docstring note pointing compaction at the
non-destructive method.
Tests: test_in_place_keeps_same_session_id strengthened to assert the 8
seeded originals survive at active=0 alongside the 2 compacted rows AND stay
FTS-searchable. Mutation check: swapping archive_and_compact back to a hard
DELETE fails the test, so the non-destructive contract is bound. 285
hermes_state + in-place tests green; rotation/persistence/compress-command/cli
suites green; ruff clean.
Parallel 3-reviewer cleanup of the in-place compaction code. Findings applied:
- perf: in-place mode no longer pre-flushes current-turn messages. The flush
ran INSERTs that the immediately-following replace_messages(compressed)
DELETE+reinsert discarded -- pure wasted writes per compaction. The
current-turn tail survives via the compressor's compressed output
(protect_last_n), not the flush. Verified no data loss; rotation still
pre-flushes (its old session row is preserved, so the flush is real there).
- quality: hoist the two shared post-write steps (update_system_prompt +
_last_flushed_db_idx = 0) below the if/else -- they ran in both branches
against agent.session_id. Removes the easiest divergence bug.
- quality: compute the compaction-boundary locals (_old_sid, _is_boundary,
_boundary_parent) ONCE instead of recomputing locals().get('old_session_id')
and the "_old_sid or agent.session_id or ''" chain three times.
- quality: initialize compacted_in_place up front and assign
agent._last_compaction_in_place directly, dropping the fragile
locals().get('compacted_in_place') reflection.
- reuse: parse the in_place config flag with utils.is_truthy_value (the
project's canonical truthy coerce) instead of a hand-rolled
str().lower() in {...} (agent_init already imports from utils).
Dropped as false positives / out of scope: gateway getattr of agent internals
(established session_id pattern), dual result-dict carry (mirrors history_offset
etc.), stringly-typed "compression" (codebase-wide convention, no constant).
Behavior-preserving: 7 in-place tests (incl. 2 new flush-guard tests) + 26
rotation/boundary/persistence/command tests green; mutation check confirms the
durable-replace guard still binds (removing replace_messages fails the test);
ruff clean. Added test_in_place_skips_redundant_preflush /
test_rotation_still_preflushes to guard the perf change.
Review (Codex + 3-agent parallel) found the first cut of in-place mode was
incomplete: it only updated the system prompt, so the persisted transcript
stayed 'full history + summary' and the next turn/resume reloaded the full
history and immediately re-compacted (a loop), and every downstream layer
that keyed off session-id rotation silently no-op'd. The session_id was
doing double duty as the 'compaction happened' signal. This wires the whole
path so removing rotation is actually complete:
Agent (agent/conversation_compression.py):
- In-place now DURABLY replaces the transcript: replace_messages(session_id,
compressed) on the same row (the canonical store the gateway reloads from),
not just update_system_prompt. Resume reloads the compacted set; no loop.
- Reset flush identity/cursor (_last_flushed_db_idx=0, _flushed_db_message_ids
cleared) so next-turn appends diff against the compacted transcript.
- Expose a rotation-independent signal: agent._last_compaction_in_place, and
in_place=True on the session:compress event.
- Fire the compaction-boundary hooks (context-engine on_session_start, memory
manager on_session_switch, reason='compression') in BOTH modes — in-place
passes the same id as parent so DAG/buffer state still checkpoints. Without
this, memory/context plugins miss every in-place compaction.
Gateway auto-compress (gateway/run.py):
- Read agent._last_compaction_in_place; set history_offset=0 on rotation OR
in-place (both return the compacted set, so slicing past the pre-compaction
length would drop everything). Carry compacted_in_place in the result dict.
- No extra rewrite needed: the agent shares the gateway's SessionDB, so its
replace_messages already updated the canonical store load_transcript reads.
Manual /compress (gateway/slash_commands.py):
- The throwaway /compress agent has no _session_db, so rewrite_transcript is
the durable write. Previously gated behind 'if rotated:' which treated
'id unchanged' as the #44794 data-loss failure case and SKIPPED the rewrite
— making /compress a silent no-op in in-place mode. Now rewrites on rotated
OR in_place; the data-loss guard still fires only for the genuine
no-rotation-AND-not-in-place failure.
Hygiene auto-compress already writes _compressed to the same id
unconditionally (its agent has no _session_db, can't rotate) — correct for
in-place, no change.
Tests (tests/run_agent/test_in_place_compaction.py):
- Assert the DURABLE transcript IS the compacted set after reload
(get_messages_as_conversation == compacted), message_count==2, flush
identity reset, and the rotation-independent signal set on in-place /
unset on rotation. Rotation regression guard unchanged.
Verified: 64 tests green across in-place + rotation/persistence/boundary/
concurrent/failure-sync/command/cli suites; E2E both modes (durable replace,
gateway offset=0, rotation preserves old transcript); ruff clean. Still
default-off.
Context compression today rewrites the message list AND rotates the
session id — it ends the session, forks a parent_session_id child, and
renumbers the title (name -> name #2). That moving identity key is the
root cause of a whole bug cluster: /goal lost (#33618), pending response
lost at the split (#14238), orphan sessions (#33907), TUI sid desync
(#36777), FTS search gaps + duplicate sidebar entries (#45117), null
continuation cwd (#42228), and title-rename dead-ends (#48989). It also
forced a large defensive apparatus (compression lock, contextvar/env/
logging triple-sync, orphan finalization, gateway SessionEntry
re-propagation, tip projection) whose only job is surviving a
mid-conversation id change.
Add a compression.in_place config flag (default False during rollout).
When True, compaction rewrites the transcript and rebuilds the system
prompt but keeps the SAME session_id: no end_session, no child row, no
title renumber, no contextvar/logging re-sync, no memory/context-engine
session-switch. The conversation keeps one durable id for life, like
Claude Code / Codex. Compaction is lossy by design — the pre-compaction
transcript is summarized away, not archived.
The rotation path is unchanged when the flag is off (moved verbatim into
an else branch). Staged rollout: this PR ships the option behind a
default-off flag for live validation; a follow-up flips the default and
deletes the now-redundant rotation machinery, superseding the 14 open
band-aid PRs in this area.
- hermes_cli/config.py: add compression.in_place (default False), documented
- agent/agent_init.py: resolve the flag -> agent.compression_in_place
- agent/conversation_compression.py: branch compress_context() on the flag
- tests/run_agent/test_in_place_compaction.py: in-place invariants +
rotation regression guard + config default
The pre-flush of current-turn messages (#47202) runs in BOTH modes, so no
boundary data loss. Prompt-cache invariant preserved: the system-prompt
rebuild is the same single sanctioned invalidation that already happens
during compaction — no NEW invalidation. Message alternation preserved.
The image-too-large reactive shrink (try_shrink_image_parts_in_messages)
conflated two independent constraints: it always rejected a resize whose
re-encoded bytes were >= the original, even when the shrink was driven by a
PIXEL-DIMENSION cap (Anthropic many-image 2000px) rather than the byte budget.
Downscaled screenshot PNGs routinely re-encode LARGER in bytes, so the
dimension-correct result was discarded and the image left oversized -> the
provider re-rejected on retry and the session wedged forever.
Fix: track which constraint triggered the shrink (bytes vs dimension) and gate
the accept on the SAME axis.
* dimension path: accept the result as long as it is now within max_dimension,
regardless of byte size (verify via Pillow; fall back to the byte gate only
when the re-encode can't be decoded).
* bytes path: still require bytes to shrink, but ALSO re-check the per-side cap
when it's active — _resize_image_for_vision returns a best-effort, possibly
over-cap blob when it exhausts its halving budget on a very-high-aspect
image, so a byte-shrink alone can leave it over the dimension cap and
re-brick on retry.
Extend the unshrinkable-oversized guard to the pixel axis so a partial shrink
doesn't burn the one-shot retry.
Single shared agent path -> fixes CLI, TUI, and gateway alike.
Adds a real-Pillow runnable proof (repro_48013_image_shrink_brick.py) that
reproduces the issue's per-image table (bricks 3/5 before, passes 5/5 after)
plus unit invariants for the dimension and bytes accept/reject paths,
partial-progress accounting, and the bytes-path still-over-cap regression
surfaced by adversarial review.
Closes#48013
