Skip to content

python_v0.1.86

Choose a tag to compare

View all tags
@Tomas2D Tomas2D released this 02 Nov 18:06
· 9 commits to main since this release
python_v0.1.86
This tag was signed with the committer’s verified signature.
Tomas2D Tomáš Dvořák
GPG key ID: 0850021F3A50BCCF
Verified
Learn about vigilant mode.
3bae010
This commit was signed with the committer’s verified signature.
Tomas2D Tomáš Dvořák
GPG key ID: 0850021F3A50BCCF
Verified
Learn about vigilant mode.

BREAKING CHANGE

  • beeai-framework[transformers] (and [all]) no longer installs torch, peft or
    outlines. Users of TransformersChatModel must install them separately, e.g. the PyTorch build for
    their hardware followed by pip install "beeai-framework[transformers]" peft outlines.

Bug Fixes

  • adapters: report cached token usage in the LangChain chat adapter (#1906)
  • internals: accept any value for a JSON Schema property without a type (#1903)
  • memory: include tool calls and results in the SummarizeMemory prompt (#1902)
  • tools: keep MCP image and audio payloads out of the model text (#1901)
  • deps: install torch, peft and outlines with the transformers extra (#1905)
  • tools: route OpenAPITool parameters by their location (#1885)
  • deps: install llama-index-core with the all extra (#1897)
  • tools: send OpenAPITool JSON request bodies as JSON (#1882)
  • adapters: keep chat model options when cloning LiteLLM and LangChain models (#1890)
  • internals: keep falsy defaults in JSONSchemaModel (#1883)
  • python: keep a whitespace-only last line buffered in LinePrefixParser (#1884)
  • python: don't send the implicit temperature=0 to Anthropic next to top_p or reasoning_effort (#1878)
  • tools: omit unset optional query parameters in OpenAPITool (#1873)
  • internals: treat a JSON Schema type list as a union (#1872)
  • adapters: price cached prompt tokens at the cache rate (#1874)
  • adapters: parse OpenAI chat messages by their role (#1868)
  • emitter: forward descendant run events to parent contexts (#1867)
  • tools: preserve false and zero JSON results (#1866)
  • adapters: support OpenAI permission round trips (#1743)
  • internals: release abort signal listeners when runs finish (#1844)
  • adapters: replace retired default models in the Python Anthropic, Gemini and Vertex AI chat models (#1857)
  • emitter: let child emitters be collected while the parent lives (#1820)
  • deps: update brotli for CVE-2026-111497 (#1851)
  • deps: pin litellm to 1.89.7 (#1848)
  • deps: update Python dependencies with security advisories (#1846)
  • deps: stop installing torch with the transformers extra (#1845)
  • adapters: mark the required MCP prompt arguments correctly (#1836)
  • adapters: deliver the canceled status to A2A streaming subscribers (#1831)
  • deps: keep agentstack installs on MCP 1.x (#1830)
  • memory: roll back failed TokenMemory.add() across Python and TypeScript (#1819)
  • agents: prevent infinite loop when max_iterations is 0 (#1821)
  • adapters: close the remaining cancellation gaps in the A2A server (#1811)
  • stop PromptTemplate.fork customizers from mutating the original (#1817)
  • memory: enforce token budget after TokenMemory sync (#1802)
  • agents: isolate ReActAgent clones from original state (#1795)
  • adapters: preserve memory when cloning AgentStackAgent (#1804)
  • tools: kill POSIX shell process groups on cancellation and timeout (#1803)
  • adapters: cancel only the requested task in the A2A server (#1794)
  • litellm: forward stop_sequences to the provider as stop (#1791)
  • tools: honor the country in OpenMeteoTool geocoding (#1800)
  • tools: make GrepTool's context_lines return the surrounding lines (#1799)
  • adapters: preserve parameters and memory when cloning A2AAgent (#1797)
  • tools/mcp: increment session refs on MCPTool.clone() (#1792)
  • internals: forward-compat fixes for Python 3.14 (#1801)
  • workflows: honor the per-run start option (#1782)
  • tools: reap shell subprocesses on cancellation (#1784)
  • middleware: add StreamToolCallMiddleware.destroy() and skip unused emitters (#1778)
  • agents: release per-run emitters in RequirementAgent (#1769)
  • agents: enforce only_before in ConditionalRequirement (#1775)
  • tools: prevent path traversal in LocalPythonStorage (#1734)
  • tools: prevent SSRF and base URL override in OpenAPITool (#1748)
  • tools: discover all MCP tool pages (#1728)
  • adapters: make the ACP adapter work with acp-sdk 1.0 and uvicorn 0.37 (#1739)
  • memory: keep the full summary text when the chat model streams (#1727)
  • tools: preserve validated nested arguments in decorated functions (#1741)

Features

  • adapters: default MiniMax text models to M3.1 Flash (#1815)
  • backend: add FallbackChatModel (#1806)
  • examples: add ACGS tool-call middleware example (#1862)
  • adapters: deprecate the AgentStack integration (#1864)
  • examples: add on-demand skill loader tool example (#1840)
  • adapters: add MiniMax regional (CN) endpoint selection (#1576)
  • adapters: forward execution limits in the serve adapters (#1737)

New Contributors

Full Changelog: python_v0.1.85...python_v0.1.86

Contributors

leninworld, Revinand, and 18 other contributors
Assets 2
Loading