文件历史

512 次代码提交

作者 SHA1 备注 提交日期
James Sanford 2a54939951 Fix streaming tool calls with tests for many variants. (#1218)
* Recorded instance of streaming tool response variant "a".

This is the typical response, where "arguments":"" arrives
first in the stream, followed by "arguments":"{}"

The response data is a real capture from the OpenRouter API,
however some request and header data may be from other test fixtures.

* Recorded instance of streaming tool response variant "b".

This is a streaming response where the first arguments
you get is a fully formed "arguments":"{}"

The response data is a real capture from the OpenRouter API,
however some request and header data may be from other test fixtures.

* Test cases for streaming tool responses.

Note that the replays are marked as "read-only", as they are variants
seen in the wild where the streaming tool call argument fragments
arrive in a specific order.

* Fix streaming tool response variant "b", where "arguments":"{}" is what arrives first.

The previous code erroneously caused the first "arguments" to be duplicated,
by using "+=" even when being initially set.

This went unnoticed as many models stream "arguments":"" first.

When a more fully formed "arguments" fragment arrived first, it was causing
"Error: Extra data: line 1 column 3 (char 2)"

* Recorded instance of streaming tool response variant "c".

This was failing with "Error: unsupported operand type(s) for +=: 'NoneType' and 'str'"

The response data is a real capture from the OpenRouter API,
however some request and header data may be from other test fixtures.

* Test case for streaming tool response variant "c".

* Fix streaming tool response variant "c".

However, I'm not sure why arguments was initially not present or seen as None.
2025-08-11 14:17:52 -07:00
Simon Willison 5204a11f33 Allow -o option when calling tools, closes #1233 2025-08-11 13:44:26 -07:00
Simon Willison cbd3aab511 GPT-5 model IDs, refs #1229 2025-08-11 13:23:36 -07:00
Simon Willison 08094082f2 Toolbox.add_tool(), prepare() and prepare_async() methods
Closes #1111
2025-08-11 13:19:31 -07:00
Simon Willison 0863ed460e llm logs -l/--latest -q option, closes #1177 2025-06-17 23:23:45 -07:00
Simon Willison 3a96d52895 Better handling of before_call cancellation, closes #1148 2025-06-01 18:36:55 -07:00
Simon Willison d96ae4ed8d Fix --async logging to database, closes #1150 2025-06-01 17:38:26 -07:00
Simon Willison 65dde782b7 Fix mypy warning, closes #1146 2025-06-01 17:03:44 -07:00
Simon Willison 30e0c4abe8 ToolResult.exception for tool errors, now logged to DB
Closes #1104
2025-06-01 17:01:40 -07:00
Sukhbinder Singh f5a38d3342 Add -h shortcut for --help
Closes #983, #1134
2025-06-01 12:31:02 -07:00
Joe Freeman 94c62f45b1 Fix type of tools argument (#1144)
* Fix type of 'tools'
2025-06-01 12:06:20 -07:00
Simon Willison ed64fc3362 chain_limit/before_call/after_call for conversations
* chain_limit/before_call/after_call for conversations, closes #1088
* Docs for before_call/after_call including for model.conversation
2025-06-01 12:00:29 -07:00
Simon Willison b5d1c5ee90 Tools can now return attachments
Closes #1014

- llm.ToolOutput(output='...', attachments=[...]) for tools to return attachments
- New table: `tool_results_attachments`
- Table is populated when tools return attachments
- llm --tools-debug shows attachments returned by tools
- llm logs shows attachments returned by tools
2025-06-01 10:08:36 -07:00
Simon Willison f74e242442 Clarifying comment 2025-06-01 09:16:37 -07:00
Simon Willison fd2f0ffbcd Better error on llm templates show, closes #1053 2025-05-31 17:26:07 -07:00
Simon Willison 330a568683 Raise error if a non-Toolbox subclass is passed to register(), closes #1114 2025-05-30 21:37:41 -07:00
Simon Willison a3a2996fed Tools in templates (#1138)
Closes #1009
2025-05-30 17:44:52 -07:00
Simon Willison b858b0083e set_resolved_model() for async models, closes #1117 2025-05-28 07:39:57 -07:00
Simon Willison 301db6d76c responses.resolved_model column and response.set_resolved_model(model_id) method, closes #1117 2025-05-28 07:17:03 -07:00
Simon Willison 6bab712cdd Fix for module 'builtins' has no attribute 'instance_id, closes #1107 2025-05-27 13:13:28 -07:00
Simon Willison 7fb4245f9d Fix for mypy 2025-05-27 10:16:27 -07:00
Simon Willison 9e25055765 Turn incorrect tool names into errors, closes #1104 2025-05-27 09:29:05 -07:00
Simon Willison 2bc6d7679c New default tool, llm_time, closes #1103 2025-05-26 23:00:18 -07:00
Simon Willison d1c5c688e2 Error on llm -c with a Toolbox, closes #1101
Refs #1092
2025-05-26 21:51:16 -07:00
Simon Willison b4365ceb35 Move llm_version into llm.tools namespace 2025-05-26 21:44:33 -07:00
Simon Willison e4ecb86421 Log tool_instances to database (#1098)
* Log tool_instances to database, closes #1089
* Tested for both sync and async models
2025-05-26 21:01:55 -07:00
Simon Willison eafe141d49 Fix for Python 3.9, refs #1099 2025-05-26 19:54:57 -07:00
Simon Willison 1dc7a1d1f9 Monotonic ULIDs, refs #1099 2025-05-26 19:49:42 -07:00
Simon Willison 57459c0a6e Fix the tests I broke in #1096 2025-05-26 19:02:14 -07:00
Simon Willison 9bbb37fae0 New default llm_version tool, closes #1096
Refs https://github.com/simonw/llm/issues/1095#issuecomment-2910574597
2025-05-26 13:30:47 -07:00
Simon Willison 218bd10d6d Include dataclasses in plugin tool docs
Refs #1000

Refs https://github.com/simonw/llm/issues/997#issuecomment-2873497310
2025-05-26 10:05:10 -07:00
Simon Willison 7eb8acb767 llm.get_key() is now a documented utility, closes #1094
Refs #1093, https://github.com/simonw/llm-tools-datasette/issues/2
2025-05-26 09:39:40 -07:00
Simon Willison a87e4505ff Remove obsolete details: mechanism from ChainResults, closes #1087 2025-05-25 22:43:45 -07:00
Simon Willison bb336d33a0 Toolbox class for class-based tool collections (#1086)
* Toolbox class for class-based tool collections

Refs #1059, #1058, #1057
2025-05-25 22:42:52 -07:00
Simon Willison abc4f473f4 --td now pretty-prints JSON if possible, closes #1083 2025-05-25 12:59:48 -07:00
Simon Willison 7837335eea Remove colon in Token usage header for consistency
More consistent with other headings in logs now.
2025-05-24 16:55:05 -07:00
André Arko fc44fd6a0d llm chat now respects default model options (#985) 2025-05-24 14:24:53 -07:00
Arjan Mossel 5d6f96a908 Provide response_json type hint (#1077) 2025-05-24 13:46:36 -07:00
Simon Willison 3d25708acd Instantiate from spec utility function, refs #1059
Code by o3, docstring by Claude Sonnet 4
2025-05-24 08:21:33 -07:00
Simon Willison a7c51e2871 Don't execute fragments/attachments in !edit sessions, closes #1074 2025-05-23 23:44:24 -07:00
Abizer Lokhandwala 9dace69af7 bugfix: cast path to str in templates_edit (#1031) 2025-05-23 23:34:16 -07:00
Dan Turkel 88d3e11c65 Support fragments in chat through -f, --sf, and !fragments (#1048)
* add vscode and uv artifacts to gitignore
* !fragment feature
* Update fragment command to accept multiple fragments
* Add support for multiple fragments in !fragment command, factor out fragment processing
* Add support for fragment/system fragment arguments in chat command
* update docs

---------

Co-authored-by: Simon Willison <swillison@gmail.com>
2025-05-23 23:31:16 -07:00
Dan Turkel 1d5d73481a Filter IDs with --prefix in llm similar (#1052) 2025-05-23 22:54:55 -07:00
Mahesh Hegde d5f7bf9073 Support supports_tools parameters in openai compatible models (#1068) 2025-05-23 22:30:51 -07:00
Simon Willison 70092f2e78 llm logs --tools flag, closes #1072 2025-05-23 21:58:24 -07:00
Simon Willison 8a1fc5a90e llm logs --tool/-T option, closes #1013 2025-05-23 21:46:10 -07:00
Simon Willison 36477cf9e5 llm chat -c and llm -c carry forward tools, closes #1020 2025-05-23 21:10:51 -07:00
Simon Willison 3cbc37a470 Log tool plugin to database
Refs https://github.com/simonw/llm/issues/1020#issuecomment-2905968166
2025-05-23 15:59:43 -07:00
Simon Willison d2886d4692 Record which plugin a tool came from, including in DB - refs #1020 2025-05-23 15:44:12 -07:00
Simon Willison cb94047b46 Removed obsolete code I accidentally added in #1070 2025-05-23 15:22:32 -07:00