This alpha introduces the embedded Codex harness in the OpenAI Agents SDK for JavaScript and TypeScript. Install and run the starter Use Node.js 24 with its bundled npm. Check the installation guide for your platform's requirements. Platform Starter archive Checksums Linux x64 Starter archive SHA-256 checksums macOS Apple silicon Starter archive SHA-256 checksums Windows x64 Starter archive SHA-25…
Minor Changes 7b05470: feat: Remove the Runloop sandbox provider and its public API; see the migration guide for supported alternatives. Patch Changes 4521cf5: fix: refresh WebSocket dependencies and development tooling. 980423f: fix: preserve Twilio audio format defaults when session audio configuration is omitted (#2072). Updated dependencies [989513a] Updated dependencies [7b05470] Updated depe…
Patch Changes Updated dependencies [4521cf5] Updated dependencies [989513a] Updated dependencies [1163b0c] Updated dependencies [9687b54] Updated dependencies [7b05470] Updated dependencies [976579e] Updated dependencies [980423f] Updated dependencies [0dd1ef8] @openai/agents-realtime@0.21.0 @openai/agents-core@0.21.0 @openai/agents-openai@0.21.0
Patch Changes 4521cf5: fix: refresh WebSocket dependencies and development tooling. 9687b54: fix: Reject WebRTC connections when the server reports an error before the initial session configuration is acknowledged (#466). 980423f: fix: preserve Twilio audio format defaults when session audio configuration is omitted (#2072). 0dd1ef8: fix: update the Realtime browser bundler to Vite 8. Updated depe…
Patch Changes 989513a: fix: Preserve bounded nested numeric token usage breakdowns across streamed runs, usage accounting, and saved run state (schema 1.23, with backward reads of earlier snapshots; #2082); usage detail values now require narrowing to numbers before arithmetic. 1163b0c: fix: Preserve image detail when reading user messages from OpenAI Conversations sessions (#2078). Updated depend…
Minor Changes 989513a: fix: Preserve bounded nested numeric token usage breakdowns across streamed runs, usage accounting, and saved run state (schema 1.23, with backward reads of earlier snapshots; #2082); usage detail values now require narrowing to numbers before arithmetic. 7b05470: feat: Remove the Runloop sandbox provider and its public API; see the migration guide for supported alternatives…
Release v0.21.0
Minor Changes edfa410: fix: Use current Vercel authentication configuration for restored sessions and disable legacy authentication fallback when current authentication is configured; retain per-create authentication choices for live sessions. Keep restoration credentials runtime-only and honor per-run resume overrides. Patch Changes 2f3591b: fix: preserve sticky permission bits through manifest n…
Patch Changes Updated dependencies [40f7de2] Updated dependencies [836cb50] Updated dependencies [05241c7] Updated dependencies [4137680] Updated dependencies [8ee5055] Updated dependencies [df20529] Updated dependencies [b89905c] Updated dependencies [d8fa6c3] Updated dependencies [9e32e88] Updated dependencies [2f3591b] Updated dependencies [8567e09] @openai/agents-core@0.20.0 @openai/agents-rea…
Patch Changes df20529: fix: Preserve audio interruption after generation with transport-appropriate buffer clearing and response-specific playback duration. Updated dependencies [40f7de2] Updated dependencies [836cb50] Updated dependencies [05241c7] Updated dependencies [4137680] Updated dependencies [8ee5055] Updated dependencies [b89905c] Updated dependencies [d8fa6c3] Updated dependencies [9e32…
Patch Changes Updated dependencies [40f7de2] Updated dependencies [836cb50] Updated dependencies [05241c7] Updated dependencies [4137680] Updated dependencies [8ee5055] Updated dependencies [b89905c] Updated dependencies [d8fa6c3] Updated dependencies [9e32e88] Updated dependencies [2f3591b] Updated dependencies [8567e09] @openai/agents-core@0.20.0
Minor Changes 836cb50: fix: default automatic MCP tool discovery to 64 pages; set maxListPages explicitly for larger listings 8ee5055: fix: Require trusted Python 3 for UnixLocal file operations and reject disabled protection. Honor runAs for file tools and replace edited files without modifying other hardlinks. Updates now require writable parent directories and preserve ordinary permissions and…
Release v0.20.0
Minor Changes 38cee6d: fix: require Daytona SDK ^0.201.0 for patched OpenTelemetry dependencies while preserving polling-based sandbox lifecycle behavior. Patch Changes 742423e: fix: treat a blank CODEX_API_KEY as unconfigured so OPENAI_API_KEY can be used 9c94138: fix: honor the invoking Runner's sensitive-data tracing policy in Codex spans, including mixed CommonJS and ESM applications. 817161d:…
Patch Changes 817161d: build: Generate package metadata with native Node.js TypeScript support. ec8bb33: fix: Preserve post-construction agent storage settings through model requests and cloning. 7bc7368: docs: Remove the beta label from the sandbox agent documentation. 7f135af: fix: update compatible runtime and development dependencies. 7efc536: fix: update compatible workspace dependencies whil…
Minor Changes 39decd7: fix: bind conditional tool approvals to isolated normalized execution input in core and Realtime, re-evaluate current policies on durable approval resumes, reject uncopyable normalized values before conditional approval, and preserve invalid-input handling (#1914, #1915). 3515c29: fix: redact default function-tool and Realtime approval parse failures; require explicit invoca…
Patch Changes c55d64d: fix: add an optional rollback item budget for Responses session compaction. 817161d: build: Generate package metadata with native Node.js TypeScript support. 194fea9: fix: omit supported generation reasoning from default OpenAI trace exports while preserving original spans and custom destinations ec8bb33: fix: Preserve post-construction agent storage settings through model r…
Minor Changes 39d5e9e: fix: bound agent tool streaming callbacks to 1024 pending events by default, with onStreamMaxPendingEvents to adjust the limit or restore unlimited buffering with null. f0f3066: fix: reject started checkpoints without completed initial input validation, including older snapshots without completion evidence; restart with the original input and context. aeef9e1: fix: bind resu…
Release v0.19.0
What's Changed feat: add server-wide MCP tool guardrails by @seratch in #1754 feat: allow custom output guardrail blocked messages by @seratch in #1745 feat(sandbox): allow labels on Docker sandbox containers by @seratch in #1740 feat: support image results in web search tools by @seratch in #1843 fix: enforce selected callable type alias contracts by @seratch in #1749 fix: preserve serialized app…
Key Changes Docker file API migration Docker file APIs now run entirely inside a running container. This changes compatibility for applications that relied on host-side file access: images need /bin/sh and GNU filesystem utilities, including realpath, and access uses the container's default user or explicit runAs. Use a compatible image and suitable file permissions. Editor updates accept source f…
What's Changed fix: preserve model settings for GPT-5 and newer models by @seratch in #1850 Documentation & Other Changes docs: document v0.17.1 SDK updates by @seratch in #1742 chore: update versions by @github-actions[bot] in #1852 Full Changelog: v0.17.1...v0.17.2
Key Changes Output-guardrail replay safety Serialized output-bearing approval checkpoints now fail closed with UserError when the SDK cannot prove which response owns a pending terminal tool output. Continue with the live RunState when possible, or start a new run from safe input instead of replaying ambiguous serialized items. When an output guardrail rejects a completed function-tool result used…
What's Changed feat(core): add model call timeouts by @seratch in #1679 feat(sandbox): add run-scoped sandbox working directories by @seratch in #1683 feat(sandbox): allow Docker sandboxes to disable networking by @seratch in #1695 feat(extensions): add Modal sandbox resource options by @seratch in #1693 fix(core): honor exact call approval decisions by @seratch in #1694 fix(sandbox): validate vie…
Key Changes Deterministic SDK testing utilities This release adds provider-neutral testing utilities through @openai/agents/testing, @openai/agents-core/testing, @openai/agents/realtime/testing, and @openai/agents-realtime/testing. ScriptedModel, scriptedSandboxSession(), and ScriptedRealtimeTransport let applications exercise runner, sandbox, and Realtime workflows without live model, sandbox, We…
Key Changes OpenAI client compatibility and the new default model This release requires openai 7.2 or later for applications that supply their own OpenAI client. Agents without an explicit model now use gpt-5.6-luna with reasoning.effort: "none" and text.verbosity: "low". Configure an explicit model or set OPENAI_DEFAULT_MODEL when an application needs a different default. MCP v2 negotiation with…
What's Changed feat: add session history transactions by @seratch in #1572 fix: resolve model-visible tool name collisions consistently by @seratch in #1565 fix: preserve approved tool results across output guardrails by @seratch in #1562 fix: preserve completed tool guardrail results by @seratch in #1578 fix: preserve inline compaction items across turns by @seratch in #1567 fix: preserve causal…
What's Changed feat: preserve sandbox environment secret references by @seratch in #1547 fix: preserve raw realtime transport event payloads by @seratch in #1550 fix: reject ephemeral paths during sandbox archive hydration by @seratch in #1553 fix: redact endpoint credentials from MCP transport errors by @seratch in #1551 fix: redact MCP URL credentials from external metadata by @seratch in #1552…
What's Changed fix(core): export lifecycle and agent tool helper types by @seratch in #1535 fix(sandbox): support native Windows paths in path grants by @seratch in #1538 fix(sandbox): allow Docker workdirs within path grants by @seratch in #1540 fix(sandbox): preserve verified Docker sessions across resumes by @seratch in #1541 Documentation & Other Changes docs: updates for v0.14.0 by @seratch i…
Key Changes Programmatic Tool Calling support The SDK now supports Programmatic Tool Calling, allowing supported models to generate hosted JavaScript that coordinates eligible tools and reduces their intermediate results. Program calls are preserved across streaming, sessions, replay, and serialized RunState. Refer to a simple example to see how it works. Sensitive model and tool data logging is n…