Commit Graph

189 Commits

Author SHA1 Message Date
Aaron Perez
9645960016 [SKV-4350] Add OLLAMA_SUPPORTS_VISION env var, update docs (#4351)
Co-authored-by: Suchintan <suchintan@users.noreply.github.com>
2025-12-22 20:20:31 -05:00
pedrohsdb
8b8e93a616 Pedro/add gemini 3 flash (#4323) 2025-12-17 20:52:29 -08:00
Stanislav Novosad
1eca20b78a Batch LLM artifacts creation (#4322) 2025-12-17 20:15:26 -07:00
pedrohsdb
f594474b9e fix(llm): strip static prompt from cached Vertex AI requests to preve… (#4321) 2025-12-17 17:25:36 -08:00
Suchintan
9add06c9a5 GPT 5.2 Support, update models (#4287)
Co-authored-by: Suchintan Singh <suchintan@skyvern.com>
2025-12-12 18:22:25 -05:00
pedrohsdb
99a477e0dd always capture llm artifacts (#4284) 2025-12-12 10:30:36 -08:00
pedrohsdb
f1c9271085 conditionally log CancelledError as debug for speculative steps (#4252) 2025-12-11 12:13:17 -08:00
pedrohsdb
c939d34603 set extract-actions thinking budget to 512, others to 1024 (#4249) 2025-12-09 12:57:10 -08:00
pedrohsdb
33bea92a5a fix: add speculative metadata support to non-router LLM handler (#4241) 2025-12-09 09:37:30 -08:00
Stanislav Novosad
0f00a86909 Move LLMAPIHandler to its own file (#4216) 2025-12-05 16:36:59 -07:00
Stanislav Novosad
a2c9d88bbb SDK: fix import time regression (#4208) 2025-12-05 15:52:59 +00:00
Shuchang Zheng
43e89d2991 add support for gpt 5.1 (#4197) 2025-12-04 17:08:16 -05:00
pedrohsdb
ce01f2cb35 fix: prevent Vertex cache contamination across different prompt templates (#4183) 2025-12-03 11:13:27 -08:00
Stanislav Novosad
4ac82ec25b Silence annoying OpenAI client shutdown error (#4157) 2025-12-01 18:02:17 -07:00
pedrohsdb
342629002a add drop_params to router fallback call to handle unsupported parameters (#4152) 2025-12-01 14:32:39 -06:00
pedrohsdb
3f11d44762 Pedro/fix vertex cache leak (#4135) 2025-11-29 07:39:05 -06:00
Suchintan
2eeca1c699 Add invalid response to log to help us better debug it (#4142)
Co-authored-by: Suchintan Singh <suchintan@skyvern.com>
2025-11-29 00:58:10 -05:00
Suchintan
d6aed0d0be Fix openai flex usage (#4141)
Co-authored-by: Suchintan Singh <suchintan@skyvern.com>
2025-11-29 00:15:38 -05:00
Mohamed Khalil
b7ecdaafb7 Add schema validation and default value filling for extraction results (#4063)
Co-authored-by: Suchintan <suchintan@users.noreply.github.com>
2025-11-28 15:10:41 +08:00
pedrohsdb
9785822e24 add magnifex openai flex config (#4106) 2025-11-26 11:21:38 -08:00
pedrohsdb
85fe9d69a5 prevent cached_content leak to non-extract prompts (#4089) 2025-11-25 08:51:37 -08:00
pedrohsdb
ae38b9096f fix(llm): prevent cached_content from being passed to non-Gemini models (#4086) 2025-11-24 18:24:45 -08:00
pedrohsdb
b52982d3c8 Pedro/thinking budget tests (#4072) 2025-11-21 21:22:12 -08:00
LawyZheng
7c189818d9 allow extract result to be non dict (#4069) 2025-11-22 10:36:43 +08:00
LawyZheng
6358b8b1d7 raise exception when non dict response (#4057) 2025-11-21 15:19:06 +08:00
pedrohsdb
d277882b72 handle list-wrapped llm responses (#4056) 2025-11-20 20:31:00 -08:00
pedrohsdb
46383c316d restore vertex cache credentials (#4050) 2025-11-20 14:54:53 -08:00
Celal Zamanoglu
5fc9435ef3 improve validations on parameter run ui (#4000)
Co-authored-by: Jonathan Dobson <jon.m.dobson@gmail.com>
2025-11-20 19:44:58 +03:00
pedrohsdb
bc6d7affd5 use explicit vertex credentials for cache manager (#4039) 2025-11-19 17:05:49 -08:00
LawyZheng
0b47dff89d fix cua engine (#4036) 2025-11-20 02:24:00 +08:00
pedrohsdb
f7e68141eb add vertex gemini 3 pro config (#4025) 2025-11-18 16:13:51 -08:00
pedrohsdb
d1c7c675cf cleaned up fallback router (#4010) 2025-11-17 12:08:19 -08:00
pedrohsdb
b7e28b075c parallelize goal check within task (#3997) 2025-11-13 17:18:32 -08:00
pedrohsdb
d88ca1ca27 Pedro/vertex cache minimal fix (#3981) 2025-11-12 10:40:52 -08:00
Marc Kelechava
ab162397cd Support Google Cloud Workload Identity for vertex models (#3956) 2025-11-10 15:56:57 -08:00
pedrohsdb
44528cbd38 Pedro/fix explicit caching vertex api (#3933) 2025-11-06 14:47:58 -08:00
pedrohsdb
0e0ae81693 Improve LLM error message when LLM is down (#3874) 2025-10-31 11:41:07 -07:00
pedrohsdb
46ee020b5d making gpt5 models have temp 1 (#3849) 2025-10-29 09:11:08 -07:00
pedrohsdb
5d7d668252 point flash and flash lite to stable (#3816) 2025-10-24 16:45:58 -07:00
greg niemeyer
de5a55bd66 add claude 4.5 haiku support (#3763)
Co-authored-by: Suchintan <suchintan@users.noreply.github.com>
2025-10-20 02:23:22 +00:00
pedrohsdb
bcb3414561 magnifex qwen3 featherless implementation (#3764) 2025-10-18 10:44:28 -07:00
Shuchang Zheng
770ddadc2f fix drop_params bug (#3756) 2025-10-17 12:00:34 -07:00
greg niemeyer
9b2bbda3c8 add support for claude sonnet 4.5 (#3692) 2025-10-12 12:57:52 -04:00
greg niemeyer
cb35d966ac fix claude output tokens (#3695) 2025-10-12 11:30:46 -04:00
Shuchang Zheng
ea92ca4c51 support openrouter qwen model (#3630) 2025-10-06 18:55:52 -07:00
pedrohsdb
bb48db6288 Updating Gemini flash pointers in registry and frontend to preview-09-2025 (#3584) 2025-10-01 15:41:14 -07:00
Jonathan Dobson
2196d46a47 Revert "Add endpoint for browser sessions history" (#3538) 2025-09-26 16:14:52 -04:00
Jonathan Dobson
1f585a184b Add endpoint for browser sessions history (#3537) 2025-09-26 16:07:27 -04:00
pedrohsdb
f40a2392c8 adding new gemini flash preview models (#3536) 2025-09-26 11:45:22 -07:00
pedrohsdb
dd9d4fb3a9 Pedro/prompt caching (#3531) 2025-09-25 15:04:54 -07:00