[misc] chore: update openai-agents requirement from <0.19,>=0.18.1 to >=0.18.1,<0.23

Closed
DE
2mo ago
+1-1

Updates the requirements on openai-agents to permit the latest version.

Release notes

Sourced from openai-agents's releases.

v0.22.0

Key Changes

Version 0.22.0 is a minor release because it adds substantial runtime hardening and tightens an existing provider configuration contract. Applications that combine an explicit openai_client with organization or project must move those values to the AsyncOpenAI client.

Highlights:

  • Redacts terminal function-tool output rejected by agent output guardrails from replayable and persisted SDK state.
  • Raises ModelBehaviorError for non-streaming Responses with terminal status failed or incomplete.
  • Rejects conflicting provider options when OpenAIProvider receives an explicit client.
  • Isolates usage accounting between independent RunState checkpoints while preserving nested-agent aggregation.
  • Expands agents registered through handoff(agent) in generated graphs.
  • Clarifies the existing shallow-copy behavior of Agent.clone() and RealtimeAgent.clone().

What's Changed

Documentation & Other Changes

New Contributors

Full Changelog: https://github.com/openai/openai-agents-python/compare/v0.21.1...v0.22.0

Changelog

Sourced from openai-agents's changelog.

0.22.0

Version 0.22.0 tightens failure handling and data isolation for several existing APIs. Applications that construct OpenAIProvider with an explicit client and also pass organization or project to the provider must remove those duplicate arguments.

Highlights:

  • When an agent-level output guardrail blocks final output produced directly by a terminal function tool, the SDK retains a replay-valid call/output pair only when validated fields permit safe reconstruction. The original function_call_output payload is replaced with the fixed text "Output withheld by an output guardrail." in session history, RunState, and streamed result state, and payload-bearing current-response guardrail metadata is cleared or replaced. If the current response contains reasoning or another unsupported shape, the SDK discards the complete current-response suffix instead. Earlier accepted turns and guardrail results remain available. See Output guardrails.
  • Non-streaming OpenAI Responses calls now raise ModelBehaviorError when the returned response has terminal status failed or incomplete, matching the existing streamed terminal-event handling. This applies to OpenAIResponsesModel and the Responses path in AnyLLMModel. See Exceptions.
  • [OpenAIProvider][agents.models.openai_provider.OpenAIProvider] now also raises UserError when openai_client is combined with organization or project. The existing conflicts with api_key, base_url, and websocket_base_url are unchanged. Configure these values on the explicit AsyncOpenAI client instead. See API keys and clients.
  • Each RunResult.to_state() checkpoint now owns an independent usage snapshot. A resumed result starts with the checkpoint totals and adds its own model calls without mutating the source result or sibling checkpoints. Nested Agent.as_tool() resumes continue to aggregate post-resume usage into the active outer run. See Usage in RunState checkpoints.
  • Agent visualization now recursively expands the tools, MCP servers, and downstream handoffs of a target registered with handoff(agent), matching direct Agent entries in an agent's handoffs list. See Generating a graph.
  • The Agent.clone() and RealtimeAgent.clone() API guidance now states their existing shallow-copy behavior precisely: list attributes that are not overridden remain the same list objects. Pass a new list when the clone must own the container independently. See Cloning/copying agents.

0.21.0

Version 0.21.0 requires openai v3 and moves the Agents SDK's OpenAI HTTP integrations to HTTPX2. Applications that use the default OpenAI client do not need to change their client setup, but applications that customize the OpenAI HTTP layer may need to migrate transport-facing code.

Highlights:

  • The required OpenAI dependency is now openai>=3.0.0,<4. A clean core installation uses HTTPX2 and no longer installs legacy httpx as a direct dependency.
  • The default OpenAI provider, Voice provider, Responses WebSocket support, tracing exporter, and provider retry normalization now use HTTPX2. Their existing Agents SDK public configuration and runtime behavior remain unchanged.
  • Applications that pass http_client= to AsyncOpenAI should migrate custom clients, transports, authentication, event hooks, mock transports, timeout values, URLs, requests, responses, and transport exception handling from httpx to httpx2. Prefer the OpenAI Python SDK's DefaultAsyncHttpx2Client when the application needs the OpenAI client's defaults plus custom HTTP options. See Custom HTTP clients with openai v3.
  • The Agents SDK does not convert arbitrary legacy HTTPX objects to HTTPX2. The OpenAI Python SDK's temporary legacy-client compatibility path requires an explicit httpx installation and should be treated as a migration bridge.
  • Local MCP HTTP customization continues to follow the installed MCP package: MCP Python SDK v1 supplies and uses legacy httpx, while MCP Python SDK v2 uses httpx2. Ordinary MCP connections do not need application changes. See MCP Python SDK v1 and v2.
  • Public provider-neutral testing utilities now cover Agent model, Sandbox session, Realtime session, and Voice pipeline workflows without provider or process dependencies. See Testing for recipes and guidance on when to keep the real provider adapter or integration boundary.

0.20.0

Version 0.20.0 includes a potentially breaking MCP dependency migration for applications that customize local MCP HTTP transports. It also updates the SDK default model used when an agent or run does not explicitly select one.

Highlights:

  • The SDK default model is now gpt-5.6-luna instead of gpt-5.4-mini. The default reasoning.effort="none" and verbosity="low" settings are unchanged.
  • Explicit agent models, run-level model overrides, and the OPENAI_DEFAULT_MODEL environment variable continue to take precedence over the SDK default.
  • Realtime input transcription settings now recognize gpt-transcribe, gpt-live-transcribe, and gpt-realtime-whisper. For low-latency gpt-live-transcribe sessions, nested audio.input.transcription settings can supply prompt, keywords, and multiple expected languages. The OpenAI client version pinned by this SDK supports the delay latency/accuracy level only with gpt-realtime-whisper. Use gpt-transcribe over WebSocket for transcription after a committed audio turn or for detected-language output. Setting audio.input.turn_detection=None explicitly disables automatic turn detection. See Input transcription settings.
  • Local MCP connections created by the Agents SDK now support MCP Python SDK v2 while retaining v1 compatibility through mcp>=1.19.0,<3. The Agents SDK adapts ordinary stdio, SSE, and Streamable HTTP connections automatically. With MCP v2 installed, these connections use mcp.Client(mode="auto") to probe the newest supported protocol and fall back to the legacy initialize handshake for older servers. If dependency resolution selects MCP v2, applications that supply custom httpx.Auth objects or httpx.AsyncClient factories must migrate those values to httpx2, or pin mcp<2 to retain the v1 HTTP stack. MCPServerStreamableHttp's params["ignore_initialized_notification_failure"] = True option also remains v1-only. See MCP Python SDK v1 and v2 for migration details.
  • Sandbox mount validation now rejects unsafe credential placement before sandbox or mount-helper side effects. Trusted applications can acknowledge mount-scoped or broad credential exposure for an exact in-container mount path without changing the storage capability tables. These acknowledgements are runtime-only and serialized sandbox state never grants credential authority by itself. At protected mount boundaries, the SDK returns a fresh redacted exception. If the source exception is an exact recognized SDK sandbox error and its approved structured fields validate, the replacement preserves that subtype and the validated safe fields. A recognized MountConfigError can also retain an SDK-generated safe validation message. Otherwise, the SDK returns a fresh generic redacted error. Provider-controlled or otherwise unapproved messages, command data, notes, context, causes, and source traceback state are not retained. See Mounts and remote storage and Resume from session state.
  • Retry policies can inspect stable replay-safety facts and explicitly set RetryDecision(approve_unsafe_replay=True) for a non-streaming request that the provider marked unsafe. This approval does not bypass aborts, emitted streamed output, or separate local-side-effect vetoes such as Programmatic Tool Calling. See Runner-managed retries.
  • Resumable RunState objects can now stage durable user input with add_input() before the next model call. Staged input survives serialization, runs through input guardrails, and produces one durable SDK input occurrence across local sessions and server-managed conversations. An explicitly approved unsafe replay can still resend the input to the provider and repeat provider-side work. See Add input before resuming.
  • Runtime reliability fixes align streamed and non-streamed output-guardrail session persistence, preserve FunctionTool subclasses during copying and namespacing, and raise an explicit error for unsupported Chat Completions audio output instead of silently completing an empty stream. The OpenAIResponsesCompactionSession wrapper attempts and awaits pre-compaction history recovery before cancellation reaches the caller. A VoicePipeline consumer now receives transcription-session close failures after a clean run, while an earlier turn failure retains precedence over a later close failure. RunState round trips now preserve local shell output, acknowledged computer safety checks, default-valued tool output fields, and Pydantic model or dataclass outputs encountered while traversing dictionaries, lists, or tuples. MCP conversion preserves free-form object schemas and image output, and serializes other raw content blocks such as audio and resource blocks as valid JSON text. MCPServerManager serializes overlapping lifecycle operations and applies finite default timeouts to connection and cleanup. Model replay removes server-owned created_by metadata from output items before using them as input.

0.19.0

This minor release does not introduce a breaking change. The minor version bump reflects a significant new OpenAI Responses feature area: Programmatic Tool Calling.

Highlights:

  • Added [ProgrammaticToolCallingTool][agents.tool.ProgrammaticToolCallingTool], which lets supported OpenAI Responses models generate JavaScript to coordinate tools eligible for Programmatic Tool Calling. It supports per-tool allowed_callers, structured outputs from FunctionTool instances, and integration with Runner streaming, guardrails, approvals, sessions, and RunState. See Programmatic Tool Calling for setup and constraints.
  • Added the public agents.decorators module and @tool as a shorter alias for the existing @function_tool decorator, alongside the existing guardrail decorators. FunctionTool instances now also support async callable objects.
  • SDK configuration now consistently accepts either typed settings objects or dictionaries across agents, runs, models, sessions, sandboxes, and voice pipelines, with validation for unknown settings.

... (truncated)

Commits
  • 4df9ecf release: 0.22.0 (#4523)
  • fb8fa1b fix(core): isolate usage between RunState checkpoints (#4479)
  • 9432f7e fix(visualization): expand handoff() targets in agent graphs (#4517)
  • 0486792 fix(core/extensions): reject terminal failed/incomplete responses in non-stre...
  • 36dbc36 docs: add testing resources to llms indexes (#4509)
  • ed644fc fix(core): redact blocked tool outputs from replay state (#4507)
  • 2c55603 test: make stream event ordering deterministic
  • 21a1f9b fix: avoid Windows integration bootstrap crashes
  • 32e4526 ci: change windows ci configuration for stability
  • ebb746d fix: keep Codex verification for development sandboxed (#4508)
  • Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

  • @dependabot rebase will rebase this PR
  • @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
  • @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
  • @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
  • @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
  • @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)

Summary by cubic

Widens the openai-agents dependency range to allow up to v0.22.0. v0.22.0 introduces a breaking change: OpenAIProvider now rejects organization or project when an explicit client is also provided, so move those values onto the AsyncOpenAI client.