hermes-agent

Files

Teknium 7cefa87ea7 fix(agent_init): reserve Gemini's default maxOutputTokens in the compressor when max_tokens is unset

The native generateContent adapter never runs uncapped: when
model.max_tokens is unset it sends maxOutputTokens=65,535
(GEMINI_DEFAULT_MAX_OUTPUT_TOKENS) because Gemini treats an omitted cap
as a low internal default. The context compressor's trigger is
pct×(window − max_tokens), and constructing it with max_tokens=None
reserved 0 — so on a 128K Gemma window the trigger landed at 98,304
while the real safe input budget was 65,537, and the provider 400'd
before compaction fired.

Live repro (real imports, temp HERMES_HOME, native Gemini base_url,
window=131072, max_tokens unset):
  before: compressor.max_tokens=None, threshold_tokens=98304,
          wire maxOutputTokens=65535 → trigger ABOVE the safe budget
  after:  compressor.max_tokens=65535, threshold_tokens=64000 → below it

Scoped to the native Gemini wiring (provider names + native base_url via
is_native_gemini_base_url; the /openai compat endpoint is excluded). The
generic provider-default reservation gap remains tracked in #63839.

Reported by @Artemonim in #57275 (residual claim 4).

2026-08-31 12:22:55 -07:00

lsp

fix(lsp): abort diagnostics waits after transport death

2026-08-25 01:48:43 +05:30

monitoring

feat(cron): acked failure signatures stop re-pinging — durable incidents + ack CLI (salvage #94692 ) (#95017 )

2026-08-25 14:02:40 -07:00

pet

feat(pet): gate Unicode placeholders to kitty and Ghostty

2026-08-28 23:38:59 -05:00

proxy_sources

…

secret_sources

fix(update): avoid cryptography self-lock on Windows

2026-08-15 13:28:14 -07:00

transports

fix(xai): request-local alias provenance + collision-safe wire aliasing

2026-08-31 10:09:04 -07:00

verify

Integrate verify subsystem with the existing verification stack

2026-08-07 10:11:05 -07:00

__init__.py

…

account_usage.py

…

acp_openai_bridge.py

refactor(acp): share one OpenAI bridge between the ACP clients

2026-08-26 10:10:11 -07:00

agent_init.py

fix(agent_init): reserve Gemini's default maxOutputTokens in the compressor when max_tokens is unset

2026-08-31 12:22:55 -07:00

agent_runtime_helpers.py

fix(cron): isolate lazy imports from stale modules

2026-08-31 09:58:51 -07:00

anthropic_adapter.py

fix(auth): keep the borrowed claude_code row out of token authority and carry the spent-rotation verdict through resolution

2026-08-29 18:34:35 -07:00

anthropic_credentials.py

fix(auth): carry the spent-rotation verdict across processes via a durable sidecar registry

2026-08-29 18:34:35 -07:00

anthropic_endpoints.py

refactor(anthropic): split the adapter godfile into four modules

2026-08-29 18:34:35 -07:00

anthropic_message_convert.py

refactor(anthropic): split the adapter godfile into four modules

2026-08-29 18:34:35 -07:00

async_utils.py

…

aux_accounting.py

…

auxiliary_client.py

fix(compression): dead Codex summary streams fail over in 60s instead of stacking 5-minute waits

2026-08-31 11:52:51 -07:00

azure_identity_adapter.py

fix(secrets): scope-aware Azure credential presence reads in identity diagnostics

2026-08-02 10:02:33 -07:00

backend_identity.py

…

background_review.py

feat: allow configured background review tools

2026-08-29 19:10:06 -07:00

battery.py

…

bedrock_adapter.py

fix(bedrock): recover from server-side cachePoint rejections per placement

2026-08-29 11:33:05 +05:30

billing_links.py

…

billing_usage.py

chore: remove unused imports and dead locals (ruff F401/F841 sweep)

2026-07-29 11:53:39 -07:00

billing_view.py

chore: remove unused imports and dead locals (ruff F401/F841 sweep)

2026-07-29 11:53:39 -07:00

bounded_response.py

…

browser_provider.py

feat(browser): integrate Browser Use CLI 3.0

2026-08-10 10:45:44 -07:00

browser_registry.py

fix(plugins): isolate ownership by profile

2026-08-12 19:13:32 -07:00

chat_completion_helpers.py

fix(xai): request-local alias provenance + collision-safe wire aliasing

2026-08-31 10:09:04 -07:00

codex_headers.py

fix(cron): isolate lazy imports from stale modules

2026-08-31 09:58:51 -07:00

codex_responses_adapter.py

fix(codex): preserve assistant image slots in replay

2026-08-28 04:58:14 -07:00

codex_runtime.py

feat(agent): context size anchors on provider-reported usage — estimation shrinks to the last turn

2026-08-28 07:51:31 -07:00

coding_context.py

fix: system prompt no longer references tools/skills the session can't use; hermes-agent skill is always kept

2026-08-24 20:25:10 -07:00

command_token_source.py

fix(auth): bound the no-TTL key_cmd token cache; docs for key_cmd

2026-08-15 03:16:21 -07:00

compaction_display.py

fix(clients): hide compaction carriers across surfaces

2026-08-21 21:53:11 +05:30

context_breakdown.py

feat(agent): context size anchors on provider-reported usage — estimation shrinks to the last turn

2026-08-28 07:51:31 -07:00

context_compressor.py

fix(agent): cap compaction threshold floor at 85% of the context window

2026-08-31 12:19:29 -07:00

context_engine.py

Revert "fix: rewire DCP context engine to current main architecture"

2026-08-08 23:09:45 +05:30

context_references.py

feat(plugins): allow plugins to register custom @-prefix context references

2026-08-12 18:41:59 -07:00

conversation_compression.py

fix(compression): truncated summaries no longer become compaction checkpoints (port of earendil-works/pi#7048 )

2026-08-31 10:39:55 -07:00

conversation_loop.py

fix(compression): keep estimate seam positional-compatible for monkeypatched estimators

2026-08-30 20:40:43 -07:00

copilot_acp_client.py

refactor(acp): share one OpenAI bridge between the ACP clients

2026-08-26 10:10:11 -07:00

credential_persistence.py

fix(auth): make the Anthropic refresh commit part of the transaction

2026-08-29 18:34:35 -07:00

credential_pool.py

fix(auth): carry the spent-rotation verdict across processes via a durable sidecar registry

2026-08-29 18:34:35 -07:00

credential_sources.py

fix(auth): read .env as utf-8-sig in the dotenv-vs-shell detector

2026-08-08 12:32:23 -07:00

credits_tracker.py

refactor(credits): fold review findings for stealth free-tier fix

2026-08-22 04:20:48 +05:30

curator_backup.py

fix(curator): abort rollback when safety snapshot fails

2026-08-16 22:06:41 -07:00

curator.py

fix(curator): remove terminal from the consolidation fork (issue #96962 )

2026-08-31 10:08:13 -07:00

deadline.py

fix(terminal): bound env.execute wait so a wedged poll cannot disable every timer

2026-08-31 10:42:39 -07:00

delegation_context.py

fix(cron): don't let a cron job inherit a kanban worker's dispatcher identity

2026-08-06 03:16:26 +05:30

display.py

feat(delegation): live orchestration of running subagents via delegate_task action param

2026-08-13 09:34:36 -07:00

empty_response_guard.py

refactor(agent): move empty-response guard settings from env vars to config.yaml

2026-08-16 22:06:08 -07:00

error_classifier.py

fix(agent): classify "media exceeds size limit" as image_too_large

2026-08-28 04:58:06 -07:00

error_surface.py

fix(desktop): error card honors the classifier's retry verdict + failing-session identity (review feedback)

2026-08-21 15:24:03 -07:00

errors.py

…

estop.py

refactor(estop): drop redundant isinstance branch, fix stale docstring

2026-09-01 00:35:29 +05:30

file_safety.py

refactor(profiles): retire the cross-profile write guard — profiles are not isolated (maintainer decision); mirror lost-write guards (#32049 ) survive; patch/write_file schemas drop cross_profile (-83 tok/call) (#97165 )

2026-08-28 06:36:22 -07:00

gemini_native_adapter.py

fix(gemini): embed images in Gemini 3.x functionResponse.parts for multimodal tool results

2026-08-27 21:07:21 +05:30

gemini_schema.py

…

i18n.py

perf: use load_config_readonly() at read-only call sites in agent/

2026-07-29 15:28:15 -07:00

image_gen_provider.py

feat(media): opt-in upscale pass for image_generate and video_generate across FAL and Krea

2026-08-08 14:49:28 -07:00

image_gen_registry.py

fix(tools): dispatch image/video FAL strictly on the stored hermes tools selection

2026-08-19 16:10:01 -07:00

image_routing.py

vision_analyze diet + image routing: explicit aux vision backend becomes the de-facto route (reverses #29135 ) (#97339 )

2026-08-28 12:15:24 -07:00

insights.py

refactor(usage): simplify-pass follow-ups

2026-08-14 02:04:24 +05:30

interrupt_compat.py

fix(agent): separate cancellation diagnostics from tool output

2026-08-23 18:25:19 -07:00

iteration_budget.py

…

jiter_preload.py

…

kanban_stop.py

…

learn_prompt.py

fix(learn): process large sources incrementally

2026-08-08 05:09:43 -07:00

learning_graph_render.py

chore: remove unused imports and dead locals (ruff F401/F841 sweep)

2026-07-29 11:53:39 -07:00

learning_graph.py

…

learning_mutations.py

…

lmstudio_reasoning.py

…

manual_compression_feedback.py

fix(compression): /compress refusal no longer reports a successful rewrite

2026-08-19 22:41:05 -07:00

markdown_tables.py

…

memory_manager.py

fix(memory): tolerate bare-signature v2 providers when forwarding checkpoint requirement

2026-08-31 09:57:22 -07:00

memory_provider.py

refactor(memory): renumber checkpoint API — v1 is the implicit historical contract, v2 opts into fail-closed checkpoints

2026-08-25 03:55:55 -07:00

message_content.py

…

message_metadata.py

fix: create message_metadata module and update chat_completion_helpers

2026-08-15 01:04:19 -07:00

message_sanitization.py

fix(compression): route-aware stale-thinking charge parity between compaction trigger and tail walks (#84371 )

2026-08-30 20:40:43 -07:00

moa_loop.py

fix(prompt-caching): tool-using sessions no longer 400 behind LiteLLM Anthropic proxies (#89886 )

2026-08-29 12:06:48 +05:30

moa_trace.py

feat(langfuse): widen tracing to errors, sessions, subagents, and MoA fan-out

2026-08-13 23:10:16 +05:30

model_metadata.py

fix(model_metadata): parse Google's 'supports up to N' context-limit phrasing

2026-08-31 12:20:02 -07:00

models_dev.py

refactor(models): simplify-pass follow-ups on the refresh path

2026-08-14 03:31:22 +05:30

moonshot_schema.py

…

native_compaction.py

fix(gateway): preserve native compaction capability on resume

2026-08-30 05:16:10 -07:00

nous_rate_guard.py

…

onboarding.py

…

oneshot.py

…

outbound_webhooks.py

fix(hooks): flush outbound queue at interpreter exit

2026-08-02 15:01:11 -07:00

plan_prompt.py

feat: /plan graduates from bundled skill to built-in command on every surface

2026-08-29 19:14:15 -07:00

plugin_llm.py

fix(plugins): complete task-routed LLM integration

2026-08-12 16:25:20 -07:00

plugin_stream_hooks.py

feat(plugins): add streaming output observer hooks

2026-08-12 18:38:25 -07:00

portal_tags.py

…

process_bootstrap.py

docs: note httpcore pin dependency in _enable_happy_eyeballs

2026-08-24 21:46:05 -07:00

prompt_builder.py

refactor(skills): shipped-set slim — 15 to optional, github 6-way merge, pdf absorbs OCR, channel-gated teams pipeline (index −26%) (#98539 )

2026-08-30 04:53:39 -07:00

prompt_cache_boundary.py

fix(caching): prevent whitespace-only text blocks in prompt cache prefix splits

2026-08-29 11:33:05 +05:30

prompt_cache_scope.py

refactor(cache): never-raising scope resolver shared by both call sites

2026-08-15 11:09:56 +05:30

prompt_caching.py

docs: note the MEASURED_1H_PROVIDERS exception in effective_cache_ttl's docstring

2026-08-29 12:23:41 +05:30

provider_projection.py

feat(agent): fold an agent-as-provider's own tool work back into the turn

2026-08-26 10:10:11 -07:00

rate_limit_tracker.py

…

reactions.py

…

reasoning_effort.py

fix(nebius): route effort through canonical clamp_effort — hand-rolled map inverted the ladder

2026-08-29 20:39:44 +05:30

reasoning_summaries.py

fix(reasoning): keep gpt-5.x summary parts as separate blocks on the chat wire

2026-08-06 22:02:37 -05:00

reasoning_timeouts.py

fix: follow-up for salvaged PRs #93250 + #96234

2026-08-27 03:24:20 -07:00

redact.py

fix(redact): keep dotted config-key scans linear past the keyword pre-gate

2026-09-01 00:25:45 +05:30

relay_llm.py

feat(relay): activate configured dynamic plugins

2026-08-19 08:52:01 -07:00

relay_runtime.py

fix(relay): complete native plugin cutover

2026-08-19 08:52:03 -07:00

relay_tools.py

fix(relay): bypass managed execution for nested calls inside managed callbacks

2026-08-08 23:46:17 -07:00

repetition_guard.py

fix(agent+discord): guard truncated-response continuation loops and cap Discord split delivery (#86581 )

2026-08-16 01:55:51 -07:00

replay_cleanup.py

…

retry_utils.py

refactor: single shared Retry-After parser

2026-07-29 10:13:50 -07:00

review_engine.py

feat: every subagent's prompt embeds the workspace's project context files

2026-08-23 19:04:37 -07:00

runtime_cwd.py

…

secret_scope.py

fix(gateway): classify relay routing stamps as process-global deployment config

2026-08-21 13:21:46 +10:00

session_activity.py

fix(gateway): split turn-hold expiry from idle-timeout failure path

2026-08-29 13:02:13 +05:30

shell_hooks.py

feat(hooks): pre_tool_call content transformation via modify directive

2026-08-15 23:03:52 -07:00

side_question.py

feat: /btw rides the background-review cache-parity fork for full-context answers

2026-08-29 08:23:47 -07:00

skill_bundles.py

feat(observability): add Relay skill metrics

2026-07-29 12:20:34 -07:00

skill_commands.py

fix: skill supporting-file listings no longer repeat the absolute path per line

2026-08-30 01:24:14 -07:00

skill_preprocessing.py

perf: use load_config_readonly() at read-only call sites in agent/

2026-07-29 15:28:15 -07:00

skill_utils.py

refactor(skills): shipped-set slim — 15 to optional, github 6-way merge, pdf absorbs OCR, channel-gated teams pipeline (index −26%) (#98539 )

2026-08-30 04:53:39 -07:00

ssl_guard.py

fix(ssl_guard): tolerate truststore SSLContext.get_ca_certs() NotImplementedError on Windows

2026-08-08 15:10:24 +05:30

ssl_verify.py

…

stream_diag.py

…

stream_single_writer.py

…

subagent_lifecycle.py

fix(agent): separate cancellation diagnostics from tool output

2026-08-23 18:25:19 -07:00

subdirectory_hints.py

fix(paths): display_hermes_home renders POSIX separators on Windows — kills ~/AppData\Local\hermes chimeras in tool schemas and user-facing messages (#97137 )

2026-08-28 05:45:00 -07:00

subscription_view.py

…

system_prompt.py

fix(prompt): skip bundled AGENTS.md for desktop launch cwd

2026-08-31 10:10:25 -07:00

terminal_env_provider.py

feat(terminal): pluggable terminal environment backends via plugin registry

2026-08-24 20:10:44 -07:00

terminal_env_registry.py

feat(terminal): pluggable terminal environment backends via plugin registry

2026-08-24 20:10:44 -07:00

think_scrubber.py

…

thinking_timeout_guidance.py

…

thread_scoped_output.py

fix(agent): stop thread output descriptor leaks

2026-08-15 00:33:01 -07:00

title_generator.py

Port from can1357/oh-my-pi#7306 : reject answer-shaped auto-title output

2026-08-16 22:10:02 -07:00

tool_dispatch_helpers.py

fix(tool-search): bridge batch barrier, listing truncation, source indexing (salvage #92693 , part 1)

2026-08-25 15:24:15 -07:00

tool_executor.py

refactor(desktop-tools): consolidate preview + project, diet the desktop_ui suite (3,861 → 2,293 tok/call, −41%) (#97659 )

2026-08-28 23:10:01 -07:00

tool_guardrails.py

refactor(tools): remove expired BFL FLUX 3 promo core tools (migration v39); FLUX 3 stays via video_gen/FAL for subscribers (#94599 )

2026-08-25 02:45:10 -07:00

tool_result_classification.py

…

trace_upload.py

fix(state): close leaked SessionDB connections on exception paths (#83226 )

2026-08-14 21:41:26 -07:00

trajectory.py

…

transcript_repair.py

fix(desktop): preserve streamed assistant text and unify atomic persistence (#95514 )

2026-08-29 11:33:05 +05:30

transcription_provider.py

feat(plugins): add pre_transcription hook and STT prompt threading

2026-08-12 19:01:30 -07:00

transcription_registry.py

fix(plugins): isolate ownership by profile

2026-08-12 19:13:32 -07:00

tts_provider.py

…

tts_registry.py

fix(plugins): isolate ownership by profile

2026-08-12 19:13:32 -07:00

turn_context.py

fix(context): fail closed when preflight compression stalls

2026-08-31 11:18:53 -07:00

turn_finalizer.py

fix(desktop): preserve streamed assistant text and unify atomic persistence (#95514 )

2026-08-29 11:33:05 +05:30

turn_retry_state.py

fix(agent): narrow #69078 image-corrupt recovery to the classifier route

2026-08-28 03:46:24 -07:00

turn_summary.py

…

usage_pricing.py

fix(pricing): support Gemini context-tiered rates in pricing snapshot (#93469 )

2026-08-24 03:23:07 -07:00

verification_evidence.py

fix(agent): reject masked verification results

2026-08-16 00:14:58 -07:00

verification_stop.py

fix: make verify_on_stop opt-in everywhere (default False, not auto) (#84383 )

2026-08-12 01:15:25 -07:00

verify_hooks.py

…

vertex_adapter.py

refactor(vertex): fold _creds_cache_key into _sa_snapshot (simplify pass)

2026-08-29 12:33:44 +05:30

video_gen_provider.py

refactor(video_generate): capability-gated dynamic schema (~814 → 458/377 tok/call) (#97095 )

2026-08-28 05:10:33 -07:00

video_gen_registry.py

fix(tools): dispatch image/video FAL strictly on the stored hermes tools selection

2026-08-19 16:10:01 -07:00

web_search_provider.py

refactor: remove the Tavily web backend; keyless ring is exa/parallel/firecrawl/keenable

2026-08-31 00:56:41 -07:00

web_search_registry.py

refactor: remove the Tavily web backend; keyless ring is exa/parallel/firecrawl/keenable

2026-08-31 00:56:41 -07:00