Google ADK v2.14.0 makes execution environments, A2A and evaluation stable

#agents #google #python #softwaredevelopment

Google ADK v2.14.0 makes execution environments, A2A and evaluation stable
📅 November 4, 2026 • Google ADK

RELEASE DATE AND SCOPE
Google ADK Python v2.14.0 is dated November 4, 2026. Its release tag was published on November 7, 2026 at 00:58 UTC, a separate timestamp worth distinguishing from the release date. Per the release notes, execution environments, A2A, evaluation and core runtime modules graduate from experimental. The change gives developers a stable API surface for these areas, rather than requiring them to build around experimental warnings.

EXECUTION BECOMES A STABLE SURFACE
Per the release notes, BaseEnvironment, LocalEnvironment, EnvironmentToolset, DaytonaEnvironment and E2BEnvironment are now stable. In total, 26 auth, environment, app, CLI and cache modules, plus 16 toolset and integration features, graduate from experimental. A streamed file read supports line ranges without loading the whole file into memory. A code executor accepts input and output files and runs skill scripts from validated paths and arguments instead of a shell command string.

These execution changes are relevant when deciding which operations an agent can invoke and how their inputs are represented.

EVALUATION ACROSS REPEATED RUNS
Per the release notes, MultiRunConfig provides options for the number of runs and parallelism, executing each evaluation case multiple times concurrently and combining scores or pass rates across runs. ToolArgsMatcher compares tool-call arguments through exact, case-insensitive, numeric-tolerance or LLM-backed semantic checks. EnvironmentSimulation uses the same matcher. For builders, this supports evaluation that distinguishes repeated-run outcomes from a single successful trace, while making the argument-matching rule an explicit part of the test.

INTEROPERABILITY AND OPERATIONS
Per the release notes, A2A integration is stable and a CLI command generates an agent card. A runner connects ADK agents to Microsoft Teams chats and channels. A classifier for workflow and tool-gating questions uses the OpenAI Decisions API, while a memory service adds vector-search retrieval with TTL expiration. AgentEngineEnvironment reports commands exceeding 570 seconds as timed out before the service deadline. The notes also report faster long multi-turn workflows and lower memory use, without specific figures.

IMPACT
The practical shift is a broader stable framework surface for execution backends, evaluation and cross-agent interoperability, not a guarantee that an application is production-ready. Check which experimental interfaces your application uses, whether evaluation repeats cases and declares its matching criteria, and whether script execution uses validated paths and arguments. Watch timeout handling, memory expiration and integration behavior in your own workflows.

Stable status helps establish an API baseline, but application-specific security and reliability still need verification.

Sources:
https://github.com/google/adk-python/releases/tag/v2.14.0 — https://github.com/google/adk-python/releases — https://newreleases.io/project/github/google/adk-python/release/v2.14.0 — https://google.github.io/adk-docs/