teknium1 8899aeff53 fix(oneshot): --usage-file ledger reports auxiliary LLM spend ...

`hermes -z --usage-file` copied only the main-loop result, so title generation,
vision, compression, web_extract and background-review calls — recorded per task
in session_model_usage — never reached the pipeline ledger the flag advertises
as "so pipelines can always account for spend". The Insights page already folds
those rows in (#23270); the ledger is now consistent with it.

- SessionDB.auxiliary_usage_by_task(session_id): per-task sums over the
  session's compression lineage (aux calls bill to the id the turn started with
  while compression mints child ids mid-turn).
- oneshot snapshots aux usage before the turn and attaches the delta after it,
  so a resumed session's earlier runs are not re-billed.
- The report gains `auxiliary` (totals + `by_task`) and
  `total_including_auxiliary`; every existing key keeps its main-loop meaning.
- The auto-title upgrade runs on a daemon thread and can still be writing when
  the turn returns: title_generator tracks in-flight upgrade threads and
  oneshot joins them (bounded) before reading — no sleep, no eager read.

Fixes #112848. Direction shared with #112852 (@KoNit-K), which folded aux into
the headline counters; this keeps them backward compatible instead.

Co-authored-by: KoNit-K <124019182+KoNit-K@users.noreply.github.com>

2026-09-16 17:03:25 -07:00

..

lsp

fix(lsp): attach reused multi-root client outside state lock

2026-09-16 16:50:07 -07:00

monitoring

refactor(redact): one secret-pattern source; a2a, gateway chat and monitoring egress scrub through redact_for_egress

2026-09-13 05:07:50 -07:00

pet

…

proxy_sources

refactor(secrets): every private-credential file is written by utils.atomic_json_write(mode=0o600)

2026-09-13 05:07:11 -07:00

secret_sources

fix(utils): atomic writers refuse to resurrect a deleted named profile home

2026-09-16 00:32:15 -07:00

transports

fix(codex): healthy app-server sessions survive long post-tool reasoning silence

2026-09-16 16:53:20 -07:00

vault_backends

fix: skip Bitwarden URIs marked match=Never when collecting fill origins

2026-09-15 04:56:01 -07:00

verify

refactor(verify): one honest return type for the compose live-state probe (str | None)

2026-09-06 20:50:02 +05:30

__init__.py

…

account_usage.py

fix(usage): preserve routed profile context in Nous account fetches

2026-09-16 11:06:06 -07:00

acp_openai_bridge.py

…

activity_tracking.py

test: make tests/ mirror the source tree; drop issue numbers from filenames

2026-09-13 09:18:02 -07:00

agent_init.py

fix(auth): named profiles no longer inherit the root profile's auth.json (#111724)

2026-09-16 14:34:59 -07:00

agent_runtime_helpers.py

fix(agent): keyless OpenCode placeholder blanks Authorization under the paid profile too

2026-09-15 04:51:15 -07:00

AGENTS.md

fix(agent): hint-preview truncation log names the real remedy; invariant tests

2026-09-15 18:19:26 -07:00

anthropic_adapter.py

refactor(agent): one home for the stream parse-error markers

2026-09-14 20:01:19 +05:30

anthropic_credentials.py

fix(auth): named profiles no longer inherit the root profile's auth.json (#111724)

2026-09-16 14:34:59 -07:00

anthropic_endpoints.py

fix(anthropic): DeepSeek thinking replay contract for third-party proxies

2026-09-06 05:37:03 -07:00

anthropic_message_convert.py

fix(anthropic): DeepSeek thinking replay contract for third-party proxies

2026-09-06 05:37:03 -07:00

api_error_summary.py

refactor(agent): one home for the stream parse-error markers

2026-09-14 20:01:19 +05:30

api_request_hooks.py

…

async_utils.py

…

aux_accounting.py

…

auxiliary_client.py

fix(aux): only treat a standalone reasoning field name as a reasoning rejection; let parameter rungs chain

2026-09-16 17:01:52 -07:00

auxiliary_health.py

fix(multiplex): a served profile's turn sees its own cwd, approvals, redaction and tool policy

2026-09-11 19:39:12 -07:00

auxiliary_wire.py

fix: sanitize resolved auxiliary chat requests before dispatch

2026-09-07 06:02:22 -07:00

azure_identity_adapter.py

fix(secret-scope): mem0, langfuse and azure credential readers stop swallowing UnscopedSecretError

2026-09-13 05:19:48 -07:00

backend_identity.py

fix(agent): preserve route URL path identity

2026-09-07 07:06:58 -07:00

background_review.py

fix(agent): same-model review fork keeps the parent's affinity header and Portal conversation root (#109964)

2026-09-14 06:55:54 -07:00

battery.py

…

bedrock_adapter.py

fix(bedrock): omit sampling params for xAI Grok on Converse

2026-09-15 04:45:33 -07:00

billing_links.py

…

billing_usage.py

fix(usage): preserve routed profile context in Nous account fetches

2026-09-16 11:06:06 -07:00

billing_view.py

…

bounded_response.py

…

browser_provider.py

…

browser_registry.py

…

chat_completion_helpers_relay.py

fix(agent): preserve streamed refusals as text (port of anomalyco/opencode#43343)

2026-09-15 03:39:07 -07:00

chat_completion_helpers.py

fix: fallback activation picks the OpenCode per-model wire (muse-spark → Responses)

2026-09-16 14:24:30 -07:00

chat_completion_nonstream.py

fix: wait sixty seconds before provider silence notices

2026-09-07 05:24:15 -07:00

chat_completion_stream_monitor.py

fix: wait sixty seconds before provider silence notices

2026-09-07 05:24:15 -07:00

client_lifecycle.py

feat(free-tier): ruled behaviour for every welcome-api failure, with friendly copy and a fault-injecting rehearsal server

2026-09-15 20:44:42 +05:30

codex_headers.py

…

codex_responses_adapter.py

fix(responses): avoid invented empty assistant turns before tool replay

2026-09-16 14:23:20 -07:00

codex_runtime.py

fix(codex): drop the stale 'watchdog tripped' retirement cause and re-scope the reset test

2026-09-16 16:53:20 -07:00

coding_context.py

perf(prefix-cache): pin session-start workspace snapshot across compaction rebuilds

2026-09-06 22:45:31 +05:30

command_token_source.py

fix(multiplex): children spawned for a served profile start from that profile's env

2026-09-15 03:47:15 -07:00

compaction_display.py

…

compression_facade.py

…

context_breakdown.py

refactor(agent): /context breakdown and native-compaction retention use the canonical token estimator

2026-09-13 05:09:43 -07:00

context_compressor_summary.py

style: trim summary dispatch trailing whitespace

2026-09-07 05:59:08 -07:00

context_compressor.py

fix(compression): mark failed skill-tool results and drop the duplicated stub suffix

2026-09-16 16:59:43 -07:00

context_engine.py

feat(compression): provider-scoped model_thresholds keys ("<provider>:<substr>")

2026-09-11 02:07:48 -07:00

context_file_sources.py

fix(context): compute-host /context and session.context_breakdown carry the per-file manifest; report blocked files

2026-09-15 03:37:49 -07:00

context_references.py

fix(profiles): side-thread asyncio.run workers keep the caller's profile scope

2026-09-16 16:59:03 -07:00

conversation_compression_manual.py

fix(agent): manual /compress "nothing to compress" gate stays gateway-only

2026-09-13 05:20:26 -07:00

conversation_compression.py

fix(compression): publish the working-attempt marker only when the summarizer actually runs

2026-09-16 16:54:37 -07:00

conversation_loop.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

copilot_acp_client.py

fix: NT-namespace guard fires before every sibling resolve (checkpoint, ACP bridge, @file:)

2026-09-15 04:17:31 -07:00

credential_persistence.py

…

credential_pool_admin.py

fix(auth): named profiles no longer inherit the root profile's auth.json (#111724)

2026-09-16 14:34:59 -07:00

credential_pool.py

fix(auth): named profiles no longer inherit the root profile's auth.json (#111724)

2026-09-16 14:34:59 -07:00

credential_sources.py

…

credits_tracker.py

The free tier is created in one place, at boot, only behind HERMES_GUEST_ONBOARDING=1 (NS-847) (#107697)

2026-09-11 03:45:33 +05:30

curator_backup.py

fix: validate the skill name before opening its lock; key lock files on a digest

2026-09-15 19:03:33 -07:00

curator.py

fix(curator): exclude disabled review candidates

2026-09-16 16:58:22 -07:00

deadline.py

fix(process): stop late forks escaping deadline tree cleanup

2026-09-07 08:19:32 -07:00

delegation_context.py

fix(kanban): read the worker's card title from the board instead of a new env var

2026-09-15 06:08:20 -07:00

display.py

fix(gateway): report structured tool failures in run events

2026-09-15 18:42:10 -07:00

empty_response_guard.py

…

error_classifier.py

refactor: trim empty-frame salvage to the translated error only

2026-09-15 18:18:08 -07:00

error_surface.py

fix(free-tier): the desktop renders a free-tier refusal as its own card, not an OAuth re-login

2026-09-15 20:44:42 +05:30

errors.py

…

estop.py

…

fallback_cooldown.py

refactor(agent): move the entitlement rejection marker into agent/fallback_cooldown

2026-09-09 11:29:56 -07:00

fast_mode.py

…

file_safety.py

refactor(file-safety): fold vault/ and browser-profile/ into the existing protected-subpath table

2026-09-15 06:20:15 -07:00

gemini_native_adapter.py

docs(gemini): scope the base-URL normalization claim to the Google host and TTS

2026-09-15 03:54:01 -07:00

gemini_schema.py

feat(gemini): send full JSON Schema tool parameters via parametersJsonSchema

2026-09-14 06:44:54 -07:00

i18n.py

fix(multiplex): tool and memory-provider env reads stay inside the routed profile

2026-09-11 15:26:46 -07:00

image_gen_provider.py

…

image_gen_registry.py

…

image_routing.py

fix: failed probe credentials cannot fall through to configured auth

2026-09-07 08:08:04 -07:00

image_token_cost.py

fix(multiplex): key tool-side and agent-side memos by profile home

2026-09-12 01:35:05 -07:00

inline_tool_executors.py

fix(session_search): forward after/before/exclude_session_ids through the inline executor

2026-09-15 03:58:05 -07:00

insights.py

fix(insights): count tool usage per session before merging the two sources

2026-09-12 08:23:13 -07:00

interrupt_compat.py

…

interrupt_control.py

fix(agent): a /steer row is human input for every user-turn predicate

2026-09-09 13:08:25 +05:30

interrupt_scope.py

fix(console): cancel/timeout interrupts the command's agent and waits for the worker

2026-09-09 10:58:16 -07:00

iteration_budget.py

fix: keep budget checkpoints out of cancelled tool results

2026-09-07 08:28:43 -07:00

jiter_preload.py

…

kanban_stop.py

fix(kanban): scope the stop nudge to the dispatcher-owned worker

2026-09-15 18:41:19 -07:00

lazy_forward.py

…

learn_prompt.py

revert: remove Collective Wisdom V1 (#94266)

2026-09-11 11:54:49 -07:00

learning_graph_render.py

…

learning_graph.py

fix: validate the skill name before opening its lock; key lock files on a digest

2026-09-15 19:03:33 -07:00

learning_mutations.py

…

lmstudio_reasoning.py

…

manual_compression_feedback.py

…

markdown_tables.py

…

memory_manager.py

refactor(memory-plugins): background threads inherit the profile context; JSON sidecar reads and the holographic config.yaml write use core primitives

2026-09-13 05:19:48 -07:00

memory_provider.py

fix(multiplex): background threads and teardown paths carry the turn's profile scope

2026-09-15 03:47:15 -07:00

message_content.py

…

message_metadata.py

…

message_sanitization.py

refactor(agent): one outbound-kwargs sanitizer seam for the main loop and the summary

2026-09-14 20:35:28 +05:30

micro_compaction.py

…

moa_loop.py

fix: stat signatures cover inode + ctime everywhere config identity is cached

2026-09-15 06:29:50 -07:00

moa_trace.py

…

model_metadata.py

feat(models): add stealth/union-alpha to the OpenRouter free-tier catalog

2026-09-16 10:52:32 -07:00

models_dev.py

fix(utils): atomic writers refuse to resurrect a deleted named profile home

2026-09-16 00:32:15 -07:00

moonshot_schema.py

…

native_compaction.py

refactor(agent): /context breakdown and native-compaction retention use the canonical token estimator

2026-09-13 05:09:43 -07:00

nous_rate_guard.py

feat(free-tier): ruled behaviour for every welcome-api failure, with friendly copy and a fault-injecting rehearsal server

2026-09-15 20:44:42 +05:30

nous_wire.py

feat(nous): anthropic_wire=auto decides a session's wire from its first response

2026-09-06 09:11:04 -07:00

onboarding.py

refactor(config): collapse thin _load_config copies onto the canonical readers

2026-09-13 05:09:06 -07:00

oneshot.py

…

opencode_affinity.py

…

outbound_webhooks.py

fix(multiplex): tool and memory-provider env reads stay inside the routed profile

2026-09-11 15:26:46 -07:00

periodic_scheduler.py

fix(agent): preserve periodic callback context

2026-09-16 11:06:06 -07:00

plan_prompt.py

…

plugin_llm.py

…

plugin_stream_hooks.py

…

portal_tags.py

…

process_bootstrap.py

fix(agent): LLM traffic honours CIDR and wildcard NO_PROXY entries like the platform adapters do

2026-09-13 05:09:43 -07:00

prompt_builder.py

fix: share the control-frame opener list and cover [System:/[IMPORTANT:/[PRIOR CONTEXT/[CONTEXT SUMMARY]

2026-09-15 19:04:59 -07:00

prompt_cache_boundary.py

…

prompt_cache_scope.py

fix(agent): same-model review fork keeps the parent's affinity header and Portal conversation root (#109964)

2026-09-14 06:55:54 -07:00

prompt_caching.py

…

provider_base.py

…

provider_media.py

feat(video): add OpenRouter Hailuo 3 Max provider

2026-09-12 13:44:52 -07:00

provider_projection.py

…

provider_registry.py

…

proxy_bypass.py

fix: bypass the proxy on loopback HTTP CDP discovery and keep NO_PROXY=* intact

2026-09-15 06:16:38 -07:00

rate_limit_credits.py

The free tier is created in one place, at boot, only behind HERMES_GUEST_ONBOARDING=1 (NS-847) (#107697)

2026-09-11 03:45:33 +05:30

rate_limit_tracker.py

…

reactions.py

…

reasoning_effort.py

fix: clamp aux reasoning effort once before profile projection

2026-09-15 18:20:13 -07:00

reasoning_params.py

refactor(agent): drop the Optional import orphaned by the LM Studio helper removal

2026-09-14 20:35:28 +05:30

reasoning_summaries.py

fix(agent): streamed reasoning_details survive for replay continuity (port of earendil-works/pi#8605)

2026-09-13 21:27:35 -07:00

reasoning_timeouts.py

feat(models): DeepSeek V4.1 Flash on the Nous Portal and OpenRouter pickers

2026-09-10 09:22:31 -07:00

redact.py

fix: redact '/'- and '~'-led secrets whose segments cannot be a path

2026-09-15 18:59:55 -07:00

relay_llm.py

…

relay_runtime.py

feat(relay): migrate legacy HERMES_NEMO_RELAY_ATIF_*/ATOF_* vars into a validated relay-plugins.toml

2026-09-15 03:44:36 -07:00

relay_tools.py

…

repetition_guard.py

…

replay_cleanup.py

refactor(agent): fail loudly on a missing turn clock; single lowercase in is_dangerous_confirmation

2026-09-13 19:23:09 +05:30

retry_utils.py

fix(agent): explicit "retry after N s" wins over "resets in ..." in the shared reset table

2026-09-13 05:09:43 -07:00

review_engine.py

fix: make review workers recognizable in live subagent viewers

2026-09-08 04:38:47 -07:00

review_idle_queue.py

…

runtime_cwd.py

…

sdk_transform_bypass.py

fix(agent): keep moved chat fields as slot placeholders so the wire body is byte-identical; one shared escape hatch

2026-09-15 19:23:53 -07:00

search_policy.py

…

secret_scope.py

fix: key the check_fn cache by the same served-profile predicate as the MCP registry scope

2026-09-15 04:56:40 -07:00

session_activity.py

refactor(gateway): iteration-progress formatter lives in agent/session_activity, tests trimmed to the invariant bar

2026-09-06 05:35:12 -07:00

session_persistence.py

fix(gateway): retain pre-admission interrupted input

2026-09-15 04:18:34 -07:00

shell_hooks.py

fix(utils): writers that published through mkstemp on main keep NEW files at 0600

2026-09-13 05:07:11 -07:00

side_question.py

…

skill_bundles.py

…

skill_commands.py

fix: validate the skill name before opening its lock; key lock files on a digest

2026-09-15 19:03:33 -07:00

skill_preprocessing.py

fix: keep Kanban worker scope out of descendant processes

2026-09-07 07:10:28 -07:00

skill_utils.py

fix: validate the skill name before opening its lock; key lock files on a digest

2026-09-15 19:03:33 -07:00

ssl_guard.py

…

ssl_verify.py

…

status_output.py

…

stream_delivery.py

…

stream_diag.py

…

stream_single_writer.py

…

subagent_lifecycle.py

…

subdirectory_hints.py

fix(agent): suppress subdirectory hint truncation warning

2026-09-15 18:19:26 -07:00

subscription_view.py

…

surface_switch.py

simplify(agent): surface switch — reuse flatten_message_text / agent_tool_names / one runtime-boundary split

2026-09-09 10:31:26 +05:30

system_prompt.py

fix: restore TestBridgeDispatch, trim #39797 tests, drop dead execution_guidance_text param

2026-09-15 04:13:13 -07:00

terminal_approval_batch.py

fix(approval): prepare batches for desktop session sources

2026-09-16 06:27:23 -05:00

terminal_env_provider.py

…

terminal_env_registry.py

…

think_scrubber.py

refactor(display): one think-tag list; ACP tool titles derive from agent/display previews

2026-09-13 05:20:26 -07:00

thinking_timeout_guidance.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

thread_scoped_output.py

…

title_generator.py

fix(oneshot): --usage-file ledger reports auxiliary LLM spend

2026-09-16 17:03:25 -07:00

tool_dispatch_helpers.py

Connector tools (Gmail, Linear, Notion, ...) are searchable and callable through tool_search for signed-in Nous users (#106842)

2026-09-10 02:21:16 +05:30

tool_executor.py

feat(approval): prepare GUI terminal asks before ordered execution

2026-09-16 06:27:23 -05:00

tool_guardrails.py

fix(guardrails): catch repeating multi-call cycles in the stall guard

2026-09-12 21:07:46 -07:00

tool_result_classification.py

…

trace_upload.py

refactor(state): fold the speculative-open retry into acquire's loop; keep real inode-swap tests

2026-09-06 22:56:51 +05:30

trajectory.py

fix(agent): keep trajectory serialization inside the best-effort handler

2026-09-12 08:25:14 -07:00

transcript_repair.py

…

transcription_provider.py

…

transcription_registry.py

…

tts_provider.py

…

tts_registry.py

…

turn_api_call.py

fix(free-tier): the desktop card body leaves the "To sign in" tail off; the button is the door

2026-09-15 20:44:42 +05:30

turn_api_error.py

feat(free-tier): ruled behaviour for every welcome-api failure, with friendly copy and a fault-injecting rehearsal server

2026-09-15 20:44:42 +05:30

turn_api_request.py

refactor(agent): one outbound-kwargs sanitizer seam for the main loop and the summary

2026-09-14 20:35:28 +05:30

turn_author.py

feat(agent): a2a_key names a bot author's turns

2026-09-10 10:45:57 -07:00

turn_context_compaction.py

refactor(compression): every compaction gate asks real usage first; rough estimates only decide whether to wait

2026-09-06 13:21:17 -07:00

turn_context.py

fix(gateway): retain pre-admission interrupted input

2026-09-15 04:18:34 -07:00

turn_empty_response.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

turn_explainers.py

fix(gateway): pin the remaining persistence-failure commands to the failing profile

2026-09-16 21:15:30 +05:30

turn_facade_lease.py

fix: keep the carried pre-admission input across the waited lease reload

2026-09-15 04:18:34 -07:00

turn_facade.py

refactor: carry the unadmitted user message from the lease sibling, not the facade

2026-09-15 04:18:34 -07:00

turn_failure_copy.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

turn_final_response.py

fix(agent): persist promoted clean-stop reasoning; pin the length negative

2026-09-13 14:46:42 -07:00

turn_finalizer.py

fix(kanban): delegate_task children never act on the parent's HERMES_KANBAN_TASK

2026-09-16 10:43:48 -07:00

turn_iteration_prep.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

turn_liveness.py

fix(agent): isolate periodic scheduler callbacks from blocking siblings (#102574)

2026-09-09 12:17:57 +05:30

turn_loop_errors.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

turn_overflow.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

turn_preflight_gate.py

refactor(compression): every compaction gate asks real usage first; rough estimates only decide whether to wait

2026-09-06 13:21:17 -07:00

turn_preflight.py

refactor(compression): every compaction gate asks real usage first; rough estimates only decide whether to wait

2026-09-06 13:21:17 -07:00

turn_recovery.py

fix(free-tier): review round 2 — route-gate the allowance verdict, keep policy/billing 403s, pool the provision RPC, guard the retry race

2026-09-15 20:44:42 +05:30

turn_request_assembly.py

fix: remove dedicated user-facing output cap controls

2026-09-07 06:15:43 -07:00

turn_response_check.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

turn_response_intake.py

fix: Responses-wire length continuation raises the output cap and drops reasoning after an empty fragment

2026-09-16 14:23:55 -07:00

turn_retry_state.py

feat(free-tier): ruled behaviour for every welcome-api failure, with friendly copy and a fault-injecting rehearsal server

2026-09-15 20:44:42 +05:30

turn_stop_gates.py

…

turn_summary.py

…

turn_tool_round.py

…

turn_tool_validation.py

fix(ux): plain-language, actionable user-facing messages (core)

2026-09-15 04:12:13 -07:00

turn_truncation.py

fix: seed Responses budget escalation from the observed ceiling when no cap is configured

2026-09-16 14:23:55 -07:00

turn_usage.py

fix(sessions): token-accounting guard stamps the agent's real source; trim salvage

2026-09-15 18:23:07 -07:00

usage_anchor.py

feat(compression): usage anchor survives DB reloads and process restarts (salvage #99585)

2026-09-06 13:21:17 -07:00

usage_pricing.py

fix(deepseek): deepseek-flash is the canonical Flash id; retired names fold onto it

2026-09-10 02:44:26 -07:00

vault_login_classifier.py

fix(vault): 2FA review follow-ups — per-digit fill only for an unmistakable maxlength=1 widget; honour otpauth digits/period/algorithm

2026-09-10 11:48:01 -07:00

vault_store.py

fix(vault): bind manager logins to every saved web origin

2026-09-15 04:56:01 -07:00

verification_evidence.py

refactor(sqlite): one open_db/transaction layer for every small store; plugin DBs use the WAL fallback

2026-09-13 05:08:29 -07:00

verification_stop.py

…

verify_hooks.py

…

vertex_adapter.py

…

video_gen_provider.py

feat(video): add OpenRouter Hailuo 3 Max provider

2026-09-12 13:44:52 -07:00

video_gen_registry.py

…

vision_message_prep.py

fix(vision): honor routed tool message veto

2026-09-15 04:48:20 -07:00

web_search_provider.py

…

web_search_registry.py

…