[florentinl]

Description

Allow services without both Datadog credentials to send AI Guard evaluations through the local Agent when it advertises the proxy route. Services with both keys continue calling AI Guard directly. Discovery is shared across clients, evaluations wait within the Agent timeout for pending discovery, and transient failures can be retried on the next evaluation. Includes Unix socket support and transport telemetry.

Original PR: #21569 by @eliottness. Eliott Bouhana remains credited as co-author.

Testing

All 561 AI Guard API and Agent-helper tests pass on Python 3.12, including concurrent startup, bounded waits, invalidation, transient recovery, a forked-child regression, advertised endpoint normalization, native HTTP requests through reverse-proxy paths, unsupported /info responses, cross-client isolation, and background-thread resource failures. All local repository checks pass, including typing, security, spelling, and ast-grep. Import analysis found no new cycles or dependency-direction violations.

Discovery performance has not been benchmarked.

Risks

Requires an Agent that supports the AI Guard proxy route. Missing credentials now produce evaluation-time errors instead of preventing client initialization. Evaluations fail if discovery cannot complete within the Agent timeout or no supported route exists.

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-09T13:42:04.923542Z	59f1205	New commits
🔒 Security Review	✅ Completed 2026-11-09T13:44:19.617271Z	59f1205	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[cit-pr-commenter-54b7da]

Codeowners resolved as

Resolved from the full PR diff against main using the target branch CODEOWNERS file.
CODEOWNERS team requests not listed below are not required by the current file set.

.cursor/rules/ai-guard.mdc                                              @DataDog/asm-python
ddtrace/aiguard/_api_client.py                                          @DataDog/asm-python
ddtrace/aiguard/_constants.py                                           @DataDog/asm-python
ddtrace/internal/agent.py                                               @DataDog/apm-core-python
docs/configuration.rst                                                  @DataDog/python-guild
releasenotes/notes/aiguard-agent-data-path-6a3f91b74c2d5e08.yaml        @DataDog/apm-python
tests/aiguard/api/conftest.py                                           @DataDog/asm-python
tests/aiguard/api/test_agent_route.py                                   @DataDog/asm-python
tests/aiguard/api/test_api_client.py                                    @DataDog/asm-python
tests/tracer/test_agent.py                                              @DataDog/apm-sdk-capabilities

[cit-pr-commenter-54b7da]

Dependency direction analysis

⚠️ Existing dependency direction violations

There are 83 dependency direction violations that already exist on the base branch and have not been changed by this PR.

Show existing violations (showing 5 of 83 highest severity)

ddtrace.internal.tracemethods -×-> ddtrace.trace  (internal-core -> product:tracing, score=120)
ddtrace.llmobs._telemetry -×-> ddtrace.trace  (product:llmobs -> product:tracing, score=118)
ddtrace.llmobs._integrations.vertexai -×-> ddtrace.trace  (product:llmobs -> product:tracing, score=118)
ddtrace.llmobs._integrations.google_adk -×-> ddtrace.trace  (product:llmobs -> product:tracing, score=118)
ddtrace.llmobs._integrations.base -×-> ddtrace.trace  (product:llmobs -> product:tracing, score=118)

To see all violations, download the layers-base.json and layers-pr.json artifacts from this CI job and run:

uv run --script scripts/import-analysis/layers.py compare layers-base.json layers-pr.json

[datadog-prod-us1-6]

 

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 59f1205 | Docs | View more details | Give us feedback!}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 452e8f64db

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[pr-commenter]

Benchmarks

Benchmark execution time: 2026-11-09 14:01:54

Comparing candidate commit 59f1205 in PR branch florentin.labelle/aiguard-agent-data-path with baseline commit 5449321 in branch main.

📊 Benchmarking dashboard

Found 0 performance improvements and 0 performance regressions! Performance is the same for 619 metrics, 9 unstable metrics, 1 flaky benchmarks without significant changes.

Explanation

This is an A/B test comparing a candidate commit's performance against that of a baseline commit. Performance changes are noted in the tables below as:

• 🟩 = significantly better candidate vs. baseline
• 🟥 = significantly worse candidate vs. baseline

We compute a confidence interval (CI) over the relative difference of means between metrics from the candidate and baseline commits, considering the baseline as the reference.

If the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD), the change is considered significant.

Feel free to reach out to #apm-benchmarking-platform on Slack if you have any questions.

More details about the CI and significant changes

You can imagine this CI as a range of values that is likely to contain the true difference of means between the candidate and baseline commits.

CIs of the difference of means are often centered around 0%, because often changes are not that big:

---------------------------------(------|---^--------)-------------------------------->
                              -0.6%    0%  0.3%     +1.2%
                                 |          |        |
         lower bound of the CI --'          |        |
sample mean (center of the CI) -------------'        |
         upper bound of the CI ----------------------'

As described above, a change is considered significant if the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD).

For instance, for an execution time metric, this confidence interval indicates a significantly worse performance:

----------------------------------------|---------|---(---------^---------)---------->
                                       0%        1%  1.3%      2.2%      3.1%
                                                  |   |         |         |
       significant impact threshold --------------'   |         |         |
                      lower bound of CI --------------'         |         |
       sample mean (center of the CI) --------------------------'         |
                      upper bound of CI ----------------------------------'

Unstable benchmarks

These benchmarks have a confidence interval too wide to call a change; treat them as noise rather than signal.

scenario:coreapiscenario-context_with_data_listeners

• unstable execution_time [-771.467ns; +624.659ns] or [-7.150%; +5.790%]

scenario:coreapiscenario-core_dispatch_50_listeners

• unstable execution_time [-1911.336ns; +1795.480ns] or [-9.475%; +8.901%]

scenario:coreapiscenario-core_dispatch_exception_listeners

• unstable execution_time [-1854.129ns; +1693.039ns] or [-9.514%; +8.688%]

scenario:coreapiscenario-core_dispatch_listeners

• unstable execution_time [-371.678ns; +345.445ns] or [-8.549%; +7.945%]

scenario:coreapiscenario-core_dispatch_no_args_listeners

• unstable execution_time [-219.963ns; +215.746ns] or [-8.059%; +7.904%]

scenario:coreapiscenario-core_dispatch_with_results_1_listener

• unstable execution_time [-90.372ns; +85.897ns] or [-6.146%; +5.842%]

scenario:coreapiscenario-core_dispatch_with_results_50_listeners

• unstable execution_time [-4558.839ns; +4855.833ns] or [-8.973%; +9.558%]

scenario:coreapiscenario-core_dispatch_with_results_listeners

• unstable execution_time [-1015.459ns; +888.683ns] or [-9.394%; +8.221%]

scenario:packagesupdateimporteddependencies-import_many_stdlib_cached

• unstable execution_time [-53.070µs; +56.828µs] or [-9.114%; +9.759%]

Known flaky benchmarks

These benchmarks are marked as flaky and will not trigger a failure. Modify FLAKY_BENCHMARKS_REGEX to control which benchmarks are marked as flaky.

Known flaky benchmarks without significant changes:

• scenario:packagespackageforrootmodulemapping-cache_off

[datadog-prod-us1-6]

More details

Concurrent evaluations share pending Agent discovery, and transient discovery failures remain retryable. Runtime behavior could not be verified because required dependencies were unavailable.

Was this helpful? React 👍 or 👎

_{Open Bits AI session}

_{🤖 Bits Code Review · Commit 9ed8af8}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9ed8af8795

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".