release: 0.11.7

[stainless-app]

Automated Release PR

0.11.7 (2026-06-01)

Full Changelog: v0.11.6...v0.11.7

Features

• examples: OpenAI Agents SDK local-sandbox tutorials (sync + async + temporal) (#377) (a66d239)

Performance Improvements

• tracing: bounded-concurrency span export (#374) (7b32a0d)

Chores

• back-merge release 0.11.6 into next (#384) (13d3eab)

---

This pull request is managed by Stainless's GitHub App.

The semver version number is based on included commit messages. Alternatively, you can manually set the version number in the title of this pull request.

For a better experience, it is recommended to use either rebase-merge or squash-merge when merging this pull request.

🔗 Stainless website
📚 Read the docs
🙋 Reach out for help or questions

Greptile Summary

This release (0.11.7) adds three new OpenAI Agents SDK local-sandbox tutorial examples (sync, async-base, and Temporal variants) plus a matching CLI template (sync-openai-agents-local-sandbox), and replaces the serial span-export loop in AsyncSpanQueue with a bounded-concurrency dispatch model that lets up to _concurrency (default 3) HTTP batch requests be in-flight at once while still preserving the per-span START-before-END ordering guarantee.

• Tracing (span_queue.py): Drain loop now dispatches each START/END batch as its own asyncio.Task; an asyncio.Semaphore bounds actual in-flight HTTP calls; END tasks snapshot in-flight START tasks as a barrier so ordering is preserved; shutdown waits for queue.join() then cleans up straggler tasks.
• Tutorials + CLI template: Three new openai_agents_local_sandbox examples demonstrate the UnixLocalSandboxClient (shell commands on the host) in sync, async-base, and Temporal modes; a corresponding sync-openai-agents-local-sandbox Jinja2 template is wired into the agentex init command.

Confidence Score: 5/5

Safe to merge — the span queue concurrency change is well-reasoned, correctly bounded, and fully covered by new tests; the tutorial additions are self-contained examples.

The bounded-concurrency span export is the most load-bearing change: the backpressure loop, semaphore, and END barrier all compose correctly, and the new test suite validates concurrency cap, serial behavior, throughput improvement, and per-span ordering. The tutorials and CLI template are additive and isolated. The only gap is the missing sgp_base_url in the Temporal tutorial, which is a minor example inconsistency, not a runtime failure.

workflow.py in the Temporal tutorial (sgp_base_url omission), and the self-noted re-enqueue ordering caveat in span_queue.py if retries are ever enabled by default.

Important Files Changed

Filename	Overview
src/agentex/lib/core/tracing/span_queue.py	Core change: replaces serial export with bounded-concurrency dispatch; logic is sound but the re-enqueue note acknowledges a START-before-END breakage when retries are enabled — not an issue at max_retries=1 (default).
tests/lib/core/tracing/test_span_queue.py	New TestAsyncSpanQueueConcurrency suite covers concurrency cap, serialization at concurrency=1, speed improvement, and per-span END-waits-for-START ordering — comprehensive coverage of the new behavior.
src/agentex/lib/cli/commands/init.py	Adds SYNC_OPENAI_AGENTS_LOCAL_SANDBOX to TemplateType enum and wires it into project_files and the interactive prompt; straightforward and consistent with existing template entries.
examples/tutorials/10_async/10_temporal/120_openai_agents_local_sandbox/project/workflow.py	Temporal tutorial workflow is well-structured; SGPTracingProcessorConfig call omits sgp_base_url (minor inconsistency with other tutorials), and the module-level tracing config runs unconditionally even when credentials are absent.
examples/tutorials/10_async/00_base/120_openai_agents_local_sandbox/project/acp.py	Async-base ACP handler with multi-turn state via adk.state; correct pattern for loading/saving conversation history, clean tracing integration.
examples/tutorials/00_sync/050_openai_agents_local_sandbox/project/acp.py	Sync ACP handler; guards tracing config behind SGP_API_KEY check (consistent with template), clean implementation.

Sequence Diagram

sequenceDiagram
    participant DL as _drain_loop
    participant D as _dispatch()
    participant RS as _run_send (task)
    participant S as asyncio.Semaphore
    participant P as Processor (HTTP)

    loop Drain items from queue
        DL->>DL: "backpressure check (len(inflight) >= concurrency?)"
        DL->>DL: collect batch (linger window)
        DL->>D: dispatch(starts, START)
        D->>RS: "create_task(_run_send(starts, barrier=[]))"
        D->>D: inflight.add(task), inflight_starts.add(task)
        DL->>DL: re-check backpressure for ENDs
        DL->>D: dispatch(ends, END)
        D->>RS: "create_task(_run_send(ends, barrier=snapshot(inflight_starts)))"
        D->>D: inflight.add(task)
    end

    par START send
        RS->>S: acquire semaphore slot
        S-->>RS: acquired
        RS->>P: on_spans_start(spans)
        P-->>RS: HTTP response
        RS->>S: release slot
        RS->>RS: queue.task_done() [finally]
        RS->>D: done_callback - inflight.discard, inflight_starts.discard
    and END send (waits for START barrier)
        RS->>RS: "await gather(*barrier, return_exceptions=True)"
        RS->>S: acquire semaphore slot
        S-->>RS: acquired
        RS->>P: on_spans_end(spans)
        P-->>RS: HTTP response
        RS->>S: release slot
        RS->>RS: queue.task_done() [finally]
        RS->>D: done_callback - inflight.discard
    end

Loading

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
examples/tutorials/10_async/10_temporal/120_openai_agents_local_sandbox/project/workflow.py:65-70
The other two tutorials (sync and async-base) both read `SGP_CLIENT_BASE_URL` from the environment and pass it as `sgp_base_url`. Omitting it here means the Temporal tutorial always uses the default production URL and cannot be pointed at a custom SGP endpoint via env var, which breaks parity with the other examples.

```suggestion
add_tracing_processor_config(
    SGPTracingProcessorConfig(
        sgp_api_key=os.environ.get("SGP_API_KEY", ""),
        sgp_account_id=os.environ.get("SGP_ACCOUNT_ID", ""),
        sgp_base_url=os.environ.get("SGP_CLIENT_BASE_URL", ""),
    )
)
```

_{Reviews (3): Last reviewed commit: "release: 0.11.7" | Re-trigger Greptile}

[stainless-app]

Release version edited manually

The Pull Request version has been manually set to 0.11.6 and will be used for the release.

If you instead want to use the version number 0.12.0 generated from conventional commits, just remove the label autorelease: custom version from this Pull Request.

[stainless-app]

🤖 Release is at https://github.com/scaleapi/scale-agentex-python/releases/tag/v0.11.7 🌻