Automated Release PR

0.12.0 (2026-05-13)

Full Changelog: v0.11.0...v0.12.0

⚠ BREAKING CHANGES

• remove AgentexTracingProcessor from default tracing processors (#349)

Features

• internal/types: support eagerly validating pydantic iterators (2c528c6)
• remove AgentexTracingProcessor from default tracing processors (#349) (73eca7a)
• streaming: emit OTel metrics for ttft, tps, token counts (#347) (3bf7d1f)

Bug Fixes

• client: add missing f-string prefix in file type error message (dcb1cb4)
• render .env.example template in agentex init (#351) (6092595)
• tracing: make SGP processor stateless to stop dropping span closes (#354) (5e9f28d)
• wire SGP_CLIENT_BASE_URL and silence openai-agents tracer in templates (#352) (870324e)

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

• Bumps to 0.12.0 with a breaking change: removes AgentexTracingProcessor from default tracing processors, requires users to explicitly register processors. Also fixes a long-standing cross-pod Temporal bug by making SGPSyncTracingProcessor and SGPAsyncTracingProcessor stateless — both processors now build and flush a complete span on start and end rather than caching state in a _spans dict that could be missing on a different pod.
• Adds a new agentex.lib.core.observability module that emits OTel metrics (ttft, ttat, tps, token counters, request counts) from the streaming model path; cardinality is bounded to model and a small fixed status enum.
• Fixes a missing f-string prefix in _files.py and introduces EagerIterable to eagerly validate pydantic iterators upfront, preventing the repeated-serialization data-loss bug.

Important Files Changed

sequenceDiagram
    participant SM as TemporalStreamingModel
    participant Hooks as TemporalStreamingHooks(LLMMetricsHooks)
    participant Metrics as LLMMetrics (OTel)
    participant SGP as SGPAsyncTracingProcessor

    SM->>SM: "stream_start_perf = time.perf_counter()"
    SM->>SM: responses.create() [stream open]
    loop per delta event
        SM->>SM: bookmark first_token_at / last_token_at / first_answer_at
    end
    SM->>SGP: on_span_end(span) [stateless: build + upsert]
    SM->>Metrics: ttft_ms / ttat_ms / tps.record()
    SM->>Hooks: on_llm_end(response)
    Hooks->>Metrics: "requests.add(1, status=success)"
    Hooks->>Metrics: input_tokens / output_tokens / cached / reasoning .add()
    alt Exception raised
        SM->>Metrics: record_llm_failure(model, exc)
    end

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

---

### Issue 1 of 1
src/agentex/lib/cli/templates/temporal-openai-agents/project/workflow.py.j2:42-50
The `add_tracing_processor_config` call is made unconditionally, regardless of whether `SGP_API_KEY` and `SGP_ACCOUNT_ID` are set. Users who generate this template without SGP credentials will still register a tracing processor that initialises `SGPClient(api_key="", ...)`, producing auth errors on every trace event. The sibling `acp.py.j2` template correctly guards this call — the same pattern should be applied here.

```suggestion
SGP_API_KEY = os.environ.get("SGP_API_KEY", "")
SGP_ACCOUNT_ID = os.environ.get("SGP_ACCOUNT_ID", "")
SGP_CLIENT_BASE_URL = os.environ.get("SGP_CLIENT_BASE_URL", "")

# Setup tracing for SGP (Scale GenAI Platform)
# This enables visibility into your agent's execution in the SGP dashboard
if SGP_API_KEY and SGP_ACCOUNT_ID:
    add_tracing_processor_config(
        SGPTracingProcessorConfig(
            sgp_api_key=SGP_API_KEY,
            sgp_account_id=SGP_ACCOUNT_ID,
            sgp_base_url=SGP_CLIENT_BASE_URL,
        )
    )
```

_{Reviews (7): Last reviewed commit: "release: 0.12.0" | Re-trigger Greptile}