Repository navigation
AI observability install guide for Atomic Agents? #21461
Description
Atomic Agents (https://github.com/eigenwise/atomic-agents, Python, MIT) is an agent framework built on Instructor and Pydantic. Since 2.15.0 it emits OpenTelemetry GenAI semantic-convention spans natively (invoke_agent, execute_tool, chat with gen_ai.* attributes and usage), with only opentelemetry-api as a dependency, so it works with PostHog's OTLP traces endpoint (/i/v0/ai/otel) or PostHogSpanProcessor without a provider instrumentor. Our guide already carries a PostHog recipe: https://eigenwise.github.io/atomic-agents/guides/observability.html
Would you take an installation guide alongside contents/docs/ai-observability/installation/instructor.mdx and mirascope.mdx? I can draft the onboarding component and the docs page, and run the sample against PostHog Cloud before opening the PR so the screenshot and attribute list are real.
One gap to be upfront about: 2.15.0 does not emit gen_ai.input.messages / gen_ai.output.messages, so the LLM generation text would not render in the AI view yet; that is planned as an opt-in. Token usage, model, latency, errors and the agent/tool tree do come through. I maintain the project.
Activity
posthog commentedon Oct 21, 2026
PostHog self-driving picked up this issue and researched it. Somebody may already be working on it, so ask before you start a fix.
Report (team access only): https://us.posthog.com/project/2/inbox/reports/01a15f93-6162-76bb-9c36-fdb871750120?link_source=github_comment
marco-g-pm commentedon Oct 21, 2026
Hey Kenny,
Marco here from the AI Observability team! That would be awesome and I agree it would work great with our OTel setup.
Some notes:
2.15.0not sending input or output messages (maybe as a callout at the setup level?)Let me know if you have any questions!
Eigenwise commentedon Oct 24, 2026
Thanks Marco. The linked drafts are up:
The setup step calls out the 2.15.0 input/output message gap. I checked the exact Python snippets offline through real Instructor/OpenAI clients and captured OTLP exports, including span parentage, token usage and the absent message attributes. No live PostHog ingestion is claimed.
They remain drafts for the rendered page/navigation checks and full repository checks. Merge order is shared content, website page, then context-mill once the docs URL is live.