Releases: oracle/wayflow
WayFlow 26.1.2
WayFlow 26.1.2 is published!
WayFlow 26.1.2 is here! This release makes WayFlow safer to run in production, easier to connect to external models and MCP servers, more compatible with Agent Spec, and more reliable across multi-agent and tool-calling workflows.
What's new
Production readiness & security
• Safer server defaults: A2AServer and OpenAIResponsesServer now require API keys on non-loopback hosts, and A2AServer.run(api_key=...) enforces bearer-token auth.
• Stronger MCP transport security: MCP HTTP and mTLS transports now verify certificates and hostnames by default.
• Scoped authless MCP: Local/test unauthenticated MCP usage now has an explicit authless_mcp_enabled() opt-in.
Models, MCP & remote calls
• Gemini support: Added GeminiModel for Google Vertex AI and Google AI Studio.
• OAuth for MCP clients: MCP clients can now use OAuth-based authorization.
• Configurable retries: New RetryPolicy object lets remote components configure retries, backoff, and request timeouts.
• OpenAI-compatible improvements: Custom TLS certificates, embedding API keys, and per-token logprobs are now supported.
Agent Spec, flows & multi-agent reliability
• Better Agent Spec support: Improved tracing compatibility, customizable component loading policies, and conversion support for summarization transforms and database datastores.
• Easier shared state: New VariableStep simplifies reading from and writing to variables in flows.
• More reliable transforms: Summarization transforms now serialize correctly and work directly on Swarm and ManagerWorkers.
Bug fixes & quality
• Improved OpenAI-compatible and Llama tool-call handling.
• Fixed OpenAI/vLLM assistant history replay for Responses API requests.
• Fixed OCIGenAIModel generation config forwarding.
• Fixed manager-worker resume behavior to avoid replaying already-materialized tool results.
• Tool results are now formatted more consistently across templates and model adapters.
Links
Artifacts:
Documentation:
New features ( read more here )
Improvements ( read more here )
Bug fixes ( read more here )
Possibly breaking changes ( read more here )
External contributions
We want to thank @qiuosier and @gotsysdba for their contributions to this release
Contributors
Assets 2
wayflow-26.1.1
WayFlow 26.1.1 is a patch release.
It includes bug fixes related to MCP session handling, MCP tool union input types, compatibility issues with Gemini models and swarm execution.
It also includes a backported feature to support MCP tool output streaming, particularly useful when tool execution is long but streamable.
The wheels are published both in our private PyPI Artifactory and on the public PyPI repository.
- To install from public PyPI:
pip install "wayflowcore==26.1.1" --force-reinstall --no-cache-dir
Bug fixes
New features
wayflow-26.1.0
We are excited to announce WayFlow 26.1.0!
This release delivers brings functionalities to easily deploy agents to production, improvements to multi-agent patterns, easier extensibility and safer template rendering.
You can install it from Pypi with:
pip install "wayflowcore==26.1.0"
What's new:
Server deployment & integrations
- Run your agents as a server: New built-in server makes it easy to deploy, run, and integrate your agents in production environments.
- OpenAI Responses API and A2A supported: Effortlessly connect WayFlow with OpenAI’s latest response APIs and agent-to-agent (A2A) workflows.
Multi-agent patterns improvements
- Refined Swarm coordination: Improved Swarm prompts and added HandoffMode to better guide delegation and enforce handoff behavior when required.
- More powerful multi-agent composition: Swarm and ManagerWorkers now support multiple tool calls per turn, and you can now build hierarchical ManagerWorkers and integrate Swarm/ManagerWorkers directly into Flows.
Security and compliance
- Stricter Jinja template rendering: No access to object attributes, only safe dictionary and loop properties. Templates that previously worked may now raise security errors.
Extensibility
- WayFlow Plugins: Extend or create Tools, Steps, and seamlessly integrate custom components in serialization and deserialization.
Breaking changes
- Stricter environment for Jinja templates rendering
- Deprecated methods and parameters removed:
- Use conversation.execute() instead of deprecated Agent/Flow.execute().
- Replace ChatHistoryContextProvider with FlowContextProvider using GetChatHistoryStep.
- Use begin_step (passing a Step) instead of begin_step_name.
- Tool names: Now enforce valid names, no whitespaces or special characters a deprecation warning is raised; in the next cycle, an error will be thrown.
Bug fixes & quality
- Default input values for agents now respected.
- Internal sanitization of agent and flow names prevents LLM API errors for multi-agent workflows.
New features
Improvements
Bug fixes
Possibly breaking changes
wayflow-25.4.3
Stricter environment for jinja templates rendering
We now use a stricter version of the SandboxedEnvironment for rendering jinja templates. No access to object attributes is allowed, only key-based access to python dictionaries and main jinja LoopContext properties are allowed.
Check the guide on How to write secure prompts with Jinja templating
You can install it with pip install "wayflowcore==25.4.3" .
Explore further:
Patch release wayflow-25.4.2
This is a patch release addressing the following issues:
- Fixed a bug where it is unable to load an API key into
OpenAICompatibleModelvia an environment variable - Fixed a bug where API URLs (for LLMs) are sometimes not correctly used internally
- Fixed an import bug where the oci optional dependency was treated as a mandatory, global dependency
You can install it with pip install "wayflowcore==25.4.2" .
Explore further:
wayflow-25.4.1
WayFlow is here: Build advanced AI-powered assistants with ease!
With this release, WayFlow provides all you need for building AI-powered assistants, supporting structured workflows, autonomous agents, multi-agent collaboration, human-in-the-loop capabilities, and tool-based extensibility. Modular design ensures you can rapidly build, iterate, and customize both simple and complex assistants for any task.
You can install it with pip install "wayflowcore==25.4.1" .
Explore further: