Releases: oracle/agent-spec
Agent Spec 26.1.2
AgentSpec 26.1.2 is published!
AgentSpec 26.1.2 is here! This release makes Agent Spec easier to run across agent frameworks, safer for remote integrations, more complete for evaluation workflows, and more reliable for LangGraph, CrewAI, OpenAI Agents, and WayFlow interoperability.
What's new
Integrations & adapters
- New Agent Spec adapters: WayFlow, Microsoft Agent Framework, and OpenAI Agents SDK.
- Expanded LangGraph support: Swarm conversion, OCI GenAI models, async server tools, flow tracing, and async tracing.
- Improved CrewAI support, including flow conversion and MCP tools.
Evaluation, flows & state
- Agent Spec Evaluation is now available for framework-agnostic evaluation of agentic systems.
- Flow Builder makes programmatic flow creation and serialization easier.
- Datastores and summarization transforms help support persistent state and long-context workflows.
- CatchExceptionNode adds explicit exception-handling branches to flows.
Models, configuration & remote calls
- Generic LlmConfig supports provider-agnostic model configuration.
- Gemini configuration support and xAI provider support for OCI GenAI.
- RetryPolicy lets remote components configure retries, backoff, and request timeouts.
- URL allow lists are now available for RemoteTool and ApiNode.
Security & production readiness
- LangGraph MCP remote transports now validate HTTPS certificates by default.
- OpenAI-compatible endpoints can use certificate and mTLS configuration.
- Agent Spec loaders now support allowed and blocked component loading policies.
- MCP stdio transport is blocked by default in loaders unless explicitly allowed.
Bug fixes & quality
- More reliable LangGraph tracing, retry handling, template rendering, and HTTPS behavior.
- Improved OpenAI Agents code generation and model-setting validation.
- Component-level helpers such as to_json(), to_yaml(), from_json(), and from_dict() simplify serialization.
Links
Artifacts:
PyAgentSpec: https://pypi.org/project/pyagentspec/26.1.2/
Documentation:
🔗 Agent Spec 26.1.2 docs: https://oracle.github.io/agent-spec/26.1.2/
Language specification: https://oracle.github.io/agent-spec/26.1.2/agentspec/language_spec_26_1_2.html
Changelog: https://oracle.github.io/agent-spec/26.1.2/changelog.html
New features (read more here: https://oracle.github.io/agent-spec/26.1.2/changelog.html#new-features)
Improvements (read more here: https://oracle.github.io/agent-spec/26.1.2/changelog.html#improvements)
Possibly breaking changes (read more here: https://oracle.github.io/agent-spec/26.1.2/changelog.html#breaking-changes)
External contributions
We want to thank @spichen , @pullely-samuel and @kanak02rawat for their contributions to this release!
agent-spec-26.1.0
We’re excited to announce AgentSpec 26.1.0!
This release focuses on safer data handling with Sensitive Fields, richer orchestration with Swarm and ManagerWorkers, and easier integrations with new LangGraph and AutoGen adapters for Agent Spec.
You can install it today with pip install "pyagentspec==26.1.0" .
What's included:
Integrations and adapters
- PyAgentSpec now includes LangGraph and AutoGen adapters for easier interoperability.
- Native support for OpenAI Responses API and A2A.
Multi-agent and orchestration
- New patterns: Swarm and Manager–Workers.
- ParallelFlowNode and ParallelMapNode for concurrent steps and map-style workflows.
- AgentSpecialization to specialize an agent for certain goals or tasks.
Output and tooling
- Structured Generation in LLM Node for reliable, schema-aligned outputs.
- Tools with user confirmation, plus ToolBoxes and BuiltinTool for modular, reusable actions.
Security and compliance
- Fields support for safer handling of confidential data.
Platform and quality
- Python 3.14 compatibility.
Breaking changes
- Sensitive Fields introduces changes that may require updates.
New features
Bug fixes
agent-spec-25.4.1
Agent Spec is now available: Quickly build portable, framework and language-agnostic agents!
This initial release establishes the foundation of the Agent Spec ecosystem with the first version of the language specification, a Python SDK (PyAgentSpec) for simplified agent development, and a set of adapters that enable running Agent Spec representations on several popular, publicly available agent frameworks.
You can install it with pip install "pyagentspec==25.4.1" .
Explore further: