langchain

Updated the langchain-mistralai package from version 1.1.5 to 1.1.6, including updates to version files and dependency lock file.

Enhanced the release process documentation in AGENTS.md and CLAUDE.md with a complete step-by-step walkthrough covering version bumps, PR merging, workflow dispatch commands, and automated tag/release creation via the `mark-release` job.

Released version 0.2.6 of the langchain-openrouter package, updating version numbers across configuration files and lock files.

ChatOpenRouter now accepts a `default_headers` parameter to inject custom HTTP headers into requests to OpenRouter, enabling features like xAI's sticky-routing prompt cache. User-supplied headers take precedence over built-in app-attribution headers, with case-insensitive collision handling to prevent duplicate headers.

Mistral AI chat responses can now properly surface citation metadata so users can map cited text back to source documents. The update normalizes Mistral's citation chunks into standard LangChain text blocks while preserving reference information through both regular and streaming responses.

Added automatic conversion of Groq's context-length errors to the standard `ContextOverflowError` so application code can catch prompt-too-long errors consistently across all LLM providers. The change detects context overflow via multiple signals in the error message and response, then raises a `GroqContextOverflowError` that preserves the original error details for backward compatibility.

When using Anthropic's prompt caching with fallback models, Anthropic-specific cache control markers were incorrectly being sent to non-Anthropic models on retry attempts, causing failures. This fix sanitizes those markers during fallback retries while keeping the primary call unchanged.

Updated code to comply with ruff preview linting rules, including converting tuple membership checks to set literals for better performance and removing unnecessary intermediate variables.

Fixed Ruff linting issues across langchain_v1 and standard-tests by replacing lambda functions with operator.itemgetter() and removing unnecessary variable assignments in generator functions.

Replaced mypy with ty for faster type checking in the text-splitters package. Includes stricter type validation that now raises a clear error when SentenceTransformersTokenTextSplitter receives incompatible parameters, and refactors several modules to satisfy ty's stricter analysis while preserving all existing behavior.

Added repository owner checks to 14 GitHub Actions workflows to prevent write-capable automation and CI jobs from running on forks, which was causing unwanted mutations and wasted resources. Also removed the dead `v03_api_doc_build` workflow that depended on an unset secret.

Multi-provider model profile refresh PR summaries now wrap each provider's changes in collapsible `<details>` toggles, making them easier to skim. Single-provider summaries remain unchanged. Workflow templates also now link directly to the workflow file instead of plain text references.

Refreshed model profiles across six AI providers (Anthropic, Fireworks, Groq, HuggingFace, Mistral, OpenRouter), adding 35 new models like Claude Sonnet 5 and Qwen variants, removing 8 older models, and updating context/output token limits and capabilities for 33 existing models.

Upgraded the langsmith library from version 0.8.18 to 0.9.5 in the HuggingFace partners integration. This update includes bug fixes for sandbox connection error handling, improved client initialization, and new features like datasets v2 resource exposure.

Upgraded pytest from version 9.1.0 to 9.1.1 in the HuggingFace partners library, which includes bug fixes for RaisesGroup logic, parametrized fixture overrides, mypy typing errors, and conftest.py loading issues.

Upgraded the transformers dependency from version 5.2.0 to 5.3.0 in the HuggingFace partners library, which includes bug fixes for CLI errors, tokenizer backend handling, and support for additional model types like Qwen3.5 and EuroBERT.

Bumped the pytest testing framework from version 9.1.0 to 9.1.1 in the Fireworks partner library, which includes bug fixes for parametrized fixtures, RaisesGroup error messages, and conftest.py file loading.

Upgraded the langsmith library from version 0.8.18 to 0.9.5 in the Fireworks partner integration, bringing bug fixes for sandbox connection error handling, improved Google ADK integration, and new datasets v2 features.

Updated the langsmith library from version 0.8.18 to 0.9.5 in the Chroma partners package. This update includes bug fixes for sandbox connection error handling, support for datasets v2, improved resource access, and various other improvements to the langsmith SDK.

Upgraded the pytest testing framework from version 9.1.0 to 9.1.1 in the Chroma partners library, which includes bug fixes for parametrized fixtures, RaisesGroup logic, and conftest.py file loading.

Upgraded langsmith from version 0.8.18 to 0.9.5 in the xai partners library, bringing bug fixes for sandbox connection handling, improved client initialization, and new datasets v2 features.

Bumped the pytest testing framework from version 9.1.0 to 9.1.1 in the xai partners library. This patch release includes bug fixes for RaisesGroup error messages, parametrized fixture overrides, type checking in parametrize decorators, and conftest.py file loading.

Bumped the uv package manager version from 0.11.17 to 0.11.26 in the GitHub Actions setup configuration. This is an automated dependency update with verified mirror availability.

Bumped 11 minor and patch version updates across three directories (/libs/model-profiles, /libs/standard-tests, /libs/text-splitters), including pytest-asyncio, ruff, langsmith, beautifulsoup4, transformers, and other testing and utility packages to their latest stable versions.

Upgraded the GitHub Actions checkout action from version 6.0.2 to 7.0.0, which includes security improvements like blocking fork PR checkouts for pull_request_target workflows, module modernization to ESM, and dependency updates.

Upgraded the yq YAML processing tool to version 4.53.3, which includes new INI quote preservation features, multiple bug fixes for TOML and YAML encoding/decoding, and dependency updates.

Bumped the setup-python action to v6.3.0, which adds RHEL support, improves pip cache handling on Windows, and upgrades dependencies. Also updated aws-actions/configure-aws-credentials to v6.2.1 with a bug fix for account ID validation across authentication paths.

Updated the docstring for the output_config parameter to reflect Sonnet 5 model's support for adaptive thinking, including new configuration examples and clarified behavior differences from earlier Claude models.

Updated contribution guidelines to require release notes for feature and behavior-changing PRs. The PR template and agent/Claude documentation now consistently expect a user-visible release note section for clarity on what changed.

Released version 0.2.5 of the langchain-openrouter package, updating version numbers across package configuration files and dependency lock files.