Find your project's connection details and see how AIO reads the traces your OpenLIT-instrumented application sends.
Prerequisites
- An AIO project. To create one, see Create a project.
- An application that's already instrumented with OpenLIT.
Get your project's connection details
- Go to the projects list, and in your project's row, select the connection icon (Connection info for followed by the project name). The Connection Info page opens.

- Next to Endpoint, select Copy endpoint, and save the value where you configure OpenLIT.
- Next to Project ID, select Copy project ID, and save the value too.
- Next to Tenant ID, select Copy tenant ID, and save the value too. If the field shows "—", AIO doesn't know the tenant ID, and the field has no copy button.
- Select Back to projects when you're done.
AIO also shows the same Connection info panel once, right after you create a project. Select Done to close it.
For the access key and secret key, see Create an API key.
Send OpenLIT traces to AIO
OpenLIT exports spans over the OpenTelemetry Protocol (OTLP). AIO applies the same endpoint and header requirements to OpenLIT as to any OpenTelemetry sender. For those requirements, see OpenTelemetry. AIO drops spans that arrive without the required project identity.
Spans usually show up in AIO within seconds. To find them, see Explore traces.
How AIO reads OpenLIT spans
AIO treats spans from OpenLIT instrumentation the same way it treats spans from the AIO Python SDK's own instrumentation. AIO maps only spans from OpenLIT's instrumentation scopes (openlit.instrumentation.*), so spans from other gen_ai emitters aren't mapped. On spans from other scopes, AIO still reads the conversation ID and marks database spans, and passes everything else through unchanged.
Span kinds
AIO marks any span that has a db.system.name or db.system attribute as a db span. For other OpenLIT spans, AIO sets the kind from the gen_ai.operation.name attribute, as shown in the following table.
gen_ai.operation.name |
Span kind |
|---|---|
chat |
llm |
execute_tool |
tool |
invoke_agent |
agent |
retrieval |
retrieval |
execute_task or invoke_workflow |
chain, unless a kind is already set |
Any value, when the span name starts with mcp |
mcp |
| Anything else | other, unless a kind is already set, such as db |
Attributes AIO reads
AIO picks up the following OpenLIT attributes:
- Provider:
gen_ai.provider.name. - Request and response models:
gen_ai.request.modelandgen_ai.response.model. AIO removes a leadingbedrock/from model names. - Token usage:
gen_ai.usage.input_tokens,gen_ai.usage.output_tokens,gen_ai.usage.cache_read.input_tokens, andgen_ai.usage.cache_creation.input_tokens. - Streaming:
gen_ai.request.stream. - Timing:
gen_ai.server.time_to_first_tokenandgen_ai.server.time_per_output_token, converted from seconds to milliseconds. - Finish reasons:
gen_ai.response.finish_reasons. - Tool calls:
gen_ai.tool.nameandgen_ai.tool.call.id. - System instructions and tool definitions:
gen_ai.system_instructionsandgen_ai.tool.definitions, from which AIO records a fingerprint.
AIO groups spans into a conversation by their shared gen_ai.conversation.id value. For more about how conversations and sessions work, see Track users and sessions.
AIO discards any cost that OpenLIT computes (gen_ai.usage.cost). To see how AIO estimates cost, see Cost tracking.
Troubleshooting
- "Project not found": The page shows a message such as "“support-bot” is not a project you can see." The page only opens projects that you can already see. Check the project in the URL, or ask for access. For details, see Project permissions.

Have a suggestion?