helm/ulysses
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
ulysses/src
History
onemorethan0 8ae2b5f58c
fix(llm): suppress thinking mode for qwen3/gemma4 on Ollama /v1 endpoint (#3228) 
* fix(llm): suppress thinking for qwen3/gemma4 on Ollama /v1 compat endpoint

When using qwen3, QwQ, gemma4, or other thinking models via Ollama's
OpenAI-compatible /v1 endpoint, the model routes all output into its
<think>...</think> reasoning block. Since Odysseus strips thinking
content from round_response and only accumulates native tool_calls,
this produces a round with 0 chars, 0 native calls, 0 tool blocks —
the agent appears to silently do nothing.

Root cause: Odysseus classifies the /v1 endpoint as provider="openai"
(not "ollama"), so the payload is built as a standard OpenAI payload
without any Ollama-specific options. Ollama's /v1 endpoint accepts
"think": false as a top-level parameter to suppress extended thinking,
but this was never sent.

Fix:
- Add _is_ollama_openai_compat_url() to detect local Ollama /v1 URLs
- Inject "think": false in both stream_llm and llm_call_async for
  thinking models (qwen3, QwQ, gemma4, DeepSeek-R1, etc.) on this
  endpoint

Verified with qwen3:14b on Ollama 0.24: with think=False the model
correctly emits native tool_calls in a single streaming chunk and
the agent executes bash/file/web tools as expected.

* fix(llm): extend _is_ollama_openai_compat_url to match localhost on any port

Per reviewer feedback on PR #3228:

1. Generalize host detection to mirror _is_ollama_native_url: match any
   localhost/127.0.0.1/0.0.0.0/::1 host (not just port 11434) so that
   custom OLLAMA_HOST ports and container remaps are also covered.

2. Add tests/test_llm_core_ollama_thinking.py covering:
   - _is_ollama_openai_compat_url for all positive/negative URL cases
     including IPv6, non-default port, native /api path, and real OpenAI
   - Payload injection: think:false set for Ollama /v1 thinking model,
     not set for non-thinking model, not set for real OpenAI endpoint,
     and set for localhost on a non-default port (the new case)
2026-06-09 07:35:15 +02:00
..
search refactor(search): make src analytics a service shim (#2264) 2026-06-04 18:57:24 +02:00
action_intents.py fix(calendar): route read requests to agent (#2452) 2026-06-05 09:24:04 +01:00
agent_loop.py Prepare tested main sync cleanup 2026-06-09 09:34:42 +09:00
agent_runs.py Handle incomplete detached agent streams 2026-06-01 16:54:11 +09:00
agent_tools.py refactor(tools): consolidate duplicated _truncate and get_mcp_manager into src/tool_utils (#3478) 2026-06-09 01:05:30 +02:00
ai_interaction.py Prepare tested main sync cleanup 2026-06-09 09:34:42 +09:00
api_key_manager.py Stop API key save() from writing other providers' keys as plaintext (#1944) 2026-06-04 04:47:13 +01:00
app_helpers.py fix: inside_base_dir raises TypeError on a non-string path instead of failing closed (#1619) 2026-06-03 09:00:04 +09:00
app_initializer.py feat(memory): add provider interface (#72) 2026-06-04 16:26:11 +01:00
assistant_log.py Odysseus v1.0 2026-05-31 23:58:26 +09:00
auth_helpers.py fix(auth): gate api tokens from user routes (#2992) 2026-06-07 12:55:01 +02:00
bg_jobs.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
bg_monitor.py Ignore non-string background stream deltas (#1549) 2026-06-03 14:11:45 +09:00
builtin_actions.py fix(memory): only delete memories the model explicitly drops in tidy (#3455) 2026-06-08 18:54:45 +02:00
builtin_mcp.py Add native Windows compatibility layer 2026-06-01 15:09:47 +09:00
caldav_sync.py fix(caldav): skip the prune when any object fails to parse (#3454) 2026-06-08 18:59:14 +02:00
caldav_writeback.py fix(caldav): disable redirects on the sync/write-back DAVClient (SSRF) (#2663) 2026-06-07 05:05:24 +01:00
chat_handler.py Scope vision model resolution by owner (#3009) 2026-06-07 12:39:02 +02:00
chat_helpers.py fix(chat): keep balanced trailing ')' when extracting URLs (#3406) 2026-06-08 21:33:29 +02:00
chat_processor.py Fix calendar routing and user-local time context (#408) 2026-06-04 13:20:04 +01:00
chatgpt_subscription.py Fix model endpoint route test regressions 2026-06-09 10:16:38 +09:00
chroma_client.py fix: ChromaDB unreachable blocks app startup for 30-60s (#326) (#476) 2026-06-01 22:22:41 +09:00
cleanup_service.py Replace cleanup service datetime.utcnow calls (#1494) 2026-06-03 14:14:27 +09:00
config.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
constants.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
context_budget.py feat: adapt agent_input_token_budget to the model context window (#1170) (#1230) 2026-06-03 00:13:53 +09:00
context_compactor.py fix(context_compactor): return original messages when compaction summary fails (#2174) 2026-06-07 16:40:16 +02:00
cookbook_serve_lifecycle.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
copilot.py feat(provider): add GitHub Copilot provider with device-flow auth (#1480) 2026-06-04 21:13:14 +02:00
database.py Odysseus v1.0 2026-05-31 23:58:26 +09:00
deep_research.py fix(research): avoid double split() call and potential IndexError (#2229) 2026-06-07 16:46:21 +02:00
document_actions.py fix: document tidy crashes on a duplicate with NULL timestamps (#1772) 2026-06-03 13:23:01 +09:00
document_processor.py fix(documents): restore PDF library metadata and preview (#2483) 2026-06-07 23:23:27 +02:00
email_thread_parser.py Ignore non-string email thread bodies (#1654) 2026-06-03 14:06:31 +09:00
embedding_lanes.py fix: split Chroma embedding lanes (#3046) 2026-06-06 03:17:19 -06:00
embeddings.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
endpoint_resolver.py fix(models): query v1 models for llama-server endpoints (#3380) 2026-06-09 01:09:02 +02:00
event_bus.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
exceptions.py Odysseus v1.0 2026-05-31 23:58:26 +09:00
generated_images.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
goal_based_extractor.py fix(deep-research): wrap fetched webpage content in untrusted-context sandbox 2026-06-06 03:37:10 -06:00
integrations.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
llm_core.py fix(llm): suppress thinking mode for qwen3/gemma4 on Ollama /v1 endpoint (#3228) 2026-06-09 07:35:15 +02:00
markitdown_runtime.py fix: is_markitdown_format crashes on a non-string path (#1618) 2026-06-03 09:00:10 +09:00
mcp_manager.py feat: Add plan mode to the chat agent (#638) 2026-06-05 16:32:25 +02:00
mcp_oauth.py feat(mcp): add Streamable HTTP transport with OAuth 2.0 (#1033) 2026-06-05 02:40:52 +02:00
memory.py refactor(memory): canonicalize memory imports (#50) 2026-06-04 05:31:15 +01:00
memory_provider.py feat(memory): add provider interface (#72) 2026-06-04 16:26:11 +01:00
memory_vector.py fix: split Chroma embedding lanes (#3046) 2026-06-06 03:17:19 -06:00
model_context.py fix(models): query v1 models for llama-server endpoints (#3380) 2026-06-09 01:09:02 +02:00
model_discovery.py feat(platform): Add support for APFEL as part of the dependencies and models for the Cookbook. (#2657) 2026-06-07 17:28:02 +02:00
pdf_form_doc.py fix(forms): keep PDF-form export from dropping values when the label has '*' (#1407) 2026-06-03 14:24:07 +09:00
pdf_forms.py Odysseus v1.0 2026-05-31 23:58:26 +09:00
pdf_runtime.py Show a clear message when PyMuPDF is missing 2026-06-01 18:27:17 +09:00
personal_docs.py Skip malformed personal keyword index rows 2026-06-03 13:42:05 +09:00
preset_manager.py fix(presets): persist presets atomically to avoid corruption on crash (#2169) 2026-06-08 19:16:37 +02:00
prompt_security.py fix(security): harden untrusted_context_message against delimiter spoofing (#3086) 2026-06-07 22:15:50 +01:00
rag_manager.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
rag_singleton.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
rag_vector.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
rate_limiter.py Odysseus v1.0 2026-05-31 23:58:26 +09:00
readiness.py feat: add /api/ready readiness probe (DB, data dir, local-first) (#1200) 2026-06-02 23:33:22 +09:00
request_models.py Odysseus v1.0 2026-05-31 23:58:26 +09:00
research_handler.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
research_utils.py Treat non-string research summaries as low quality 2026-06-03 13:42:24 +09:00
secret_storage.py refactor(constants): single source of truth for data dir (#3368) 2026-06-08 09:58:52 +02:00
session_actions.py fix(sessions): keep fresh chats during auto tidy (#1871) 2026-06-09 01:06:20 +01:00
session_search.py feat(search): unify session transcript search (#2877) 2026-06-05 18:08:31 -06:00
settings.py feat(reminders): add generic webhook as a fourth reminder channel (#2952) 2026-06-05 22:47:57 +02:00
settings_scrub.py Harden note reminder dispatch ownership (#2999) 2026-06-07 12:52:27 +02:00
task_endpoint.py Harden session endpoint owner scope (#1308) 2026-06-03 02:40:22 +09:00
task_scheduler.py refactor(tools): consolidate duplicated _truncate and get_mcp_manager into src/tool_utils (#3478) 2026-06-09 01:05:30 +02:00
teacher_escalation.py fix(presets): scope expand-prompt model resolution to owner (#3477) 2026-06-08 21:12:02 +02:00
text_helpers.py fix: normalize Gemma 4 thought-channel output (#2224) 2026-06-04 19:26:58 +02:00
tls_overrides.py Support extra CA bundle for private-CA LLM providers (#769) 2026-06-04 13:18:50 +01:00
tool_execution.py refactor(tools): consolidate duplicated _truncate and get_mcp_manager into src/tool_utils (#3478) 2026-06-09 01:05:30 +02:00
tool_implementations.py Cookbook UI: Ollama browser, advanced serve fold, API tokens form, diagnosis toolbar, polish 2026-06-09 09:46:19 +09:00
tool_index.py Fix remaining CI regressions 2026-06-09 10:21:56 +09:00
tool_parsing.py fix(agent): stop treating illustrative Markdown fences as tool calls for native function-calling models (#3356) 2026-06-08 22:25:28 +02:00
tool_policy.py fix(agent): enforce guide-only tool policy (#3088) 2026-06-06 18:48:24 -06:00
tool_schemas.py Prepare tested main sync cleanup 2026-06-09 09:34:42 +09:00
tool_security.py feat: Add plan mode to the chat agent (#638) 2026-06-05 16:32:25 +02:00
tool_utils.py refactor(tools): consolidate duplicated _truncate and get_mcp_manager into src/tool_utils (#3478) 2026-06-09 01:05:30 +02:00
topic_analyzer.py Topics: hydrate session history before analysis 2026-06-02 20:44:27 +09:00
upload_handler.py fix(upload): configure chat attachment size limit (#2439) 2026-06-07 22:42:24 +02:00
upload_limits.py refactor(uploads): centralize upload byte-limits in upload_limits.py (#3364) (#3518) 2026-06-09 01:24:30 +02:00
url_safety.py fix: check_outbound_url crashes on a truthy non-string URL (#1623) 2026-06-03 08:59:49 +09:00
url_security.py Reapply "Merge branch 'main' of github.com:pewdiepie-archdaemon/odysseus" 2026-06-03 22:47:00 +09:00
user_time.py Fix calendar routing and user-local time context (#408) 2026-06-04 13:20:04 +01:00
visual_report.py security: sanitize rendered research-report HTML (#364) 2026-06-04 13:42:49 +01:00
webhook_manager.py fix(webhooks): redact IPv6 addresses in sanitized error messages (#3038) 2026-06-07 04:55:33 +01:00
youtube_handler.py fix: youtube transcript formatter crashes on a non-dict segment (#1745) 2026-06-03 13:29:08 +09:00