Uptate litellm -> 1.20.6 (fixes bugs, incl dropping API params)
0.1.180
None
0.1.179
* Task: in interactive mode, Agent should respond to LLM tool, without waiting for human input * run_batch_tasks: set stream = False in LLM config * parsing.utils.extract_content_from_path - util fn to extract content from (list of) path(s) or URL(s)
0.1.178
* `CSVGraphChatAgent`: Agent to chat with Neo4j graph built from a pandas dataframe. * Tweaks to the various web search tools `GoogleSearchTool`, `SciPhiSearchTool`, `MetaphorSearchTool`.
0.1.177
* `TableChatAgent` enhancement (improve agent tool handling, DONE condition) * `Task`: - when `interactive = True`, only user should be allowed to quit top level; - when `default_human_response = ""`, set `interative = False` - `done()`: first check if we are at top level task and non-interactive
0.1.176
* `DocChatAgent`: Allow `ingest_docs()` to be incremental; added `clear()` method to "forget all docs, clear vecdb collection" * `DocChatAgent`: `ingest_doc_paths()`, `ingest_docs()` take an optional `metadata` param which can be either: - a list of metadata-dicts of same size as the paths or docs - a single metadata-dict that is applied all the docs or paths