build(deps): bump @posthog/ai from 8.2.2 to 8.4.0

[dependabot]

Bumps @posthog/ai from 8.2.2 to 8.4.0.

Release notes

Sourced from @​posthog/ai's releases.

@​posthog/ai@​8.4.0

8.4.0

Minor Changes

• #4163 45f5488 Thanks @​jurajmajerik! - feat(ai): add a label option to Prompts.get() to fetch the prompt version a label (e.g. production) currently points to. Labeled fetches are cached separately, results carry the resolved label, and a warning is logged when the server does not resolve the requested label (older PostHog versions ignore the parameter and return the latest version). (2026-07-17)

@​posthog/ai@​8.3.1

8.3.1

Patch Changes

• #4071 182434f Thanks @​DerGeraetK! - fix(ai): declare $ai_cache_reporting_exclusive: false on OpenAI wrapper events so ingestion no longer double-bills cached input tokens for Claude models served through OpenAI-compatible hosts (e.g. OpenRouter). The flag stays unset when callers pass their own input or cache token counts through posthogProperties, so existing passthrough workarounds keep reporting correctly. (2026-07-16)

@​posthog/ai@​8.3.0

8.3.0

Minor Changes

• #4084 f4b453a Thanks @​marandaneto! - Add optional error callback support to the OpenAI Agents tracing processor. (2026-07-07)

@​posthog/ai@​8.2.3

8.2.3

Patch Changes

• #4076 68640de Thanks @​bamtheboozle! - fix: prevent unhandled promise rejection from crashing the host process when a streamed provider response errors mid-flight (OpenAI, Azure OpenAI, Anthropic wrappers) (2026-07-06)

Changelog

Sourced from @​posthog/ai's changelog.

8.4.0

Minor Changes

• #4163 45f5488 Thanks @​jurajmajerik! - feat(ai): add a label option to Prompts.get() to fetch the prompt version a label (e.g. production) currently points to. Labeled fetches are cached separately, results carry the resolved label, and a warning is logged when the server does not resolve the requested label (older PostHog versions ignore the parameter and return the latest version). (2026-07-17)

8.3.1

Patch Changes

• #4071 182434f Thanks @​DerGeraetK! - fix(ai): declare $ai_cache_reporting_exclusive: false on OpenAI wrapper events so ingestion no longer double-bills cached input tokens for Claude models served through OpenAI-compatible hosts (e.g. OpenRouter). The flag stays unset when callers pass their own input or cache token counts through posthogProperties, so existing passthrough workarounds keep reporting correctly. (2026-07-16)

8.3.0

Minor Changes

• #4084 f4b453a Thanks @​marandaneto! - Add optional error callback support to the OpenAI Agents tracing processor. (2026-07-07)

8.2.3

Patch Changes

• #4076 68640de Thanks @​bamtheboozle! - fix: prevent unhandled promise rejection from crashing the host process when a streamed provider response errors mid-flight (OpenAI, Azure OpenAI, Anthropic wrappers) (2026-07-06)

Commits

• 6326610 chore: update versions and lockfile [version bump]
• 45f5488 feat(prompts): Fetch prompts by label (#4163)
• 943d9b8 chore: update versions and lockfile [version bump]
• 182434f fix(ai): declare inclusive cache token reporting for OpenAI-compatible provid...
• 629e6c4 chore: delete eslint-disable comments that suppress nothing (#4146)
• 7a35382 chore: update versions and lockfile [version bump]
• f4b453a feat(ai): report OpenAI Agents tracing processor errors (#4084)
• 5720595 chore: update versions and lockfile [version bump]
• 68640de fix(ai): prevent unhandled rejection crashing host process on mid-stream prov...
• See full diff in compare view

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

[greptile-apps]

PR author is in the excluded authors list.

[capy-ai]

Capy auto-review is paused for this organization because the usage-cycle auto-review limit has been reached. Increase the limit or turn it off in billing settings to resume automatic reviews.