Is There an Equivalent to Span Attributes in Arize?

Eugene E. · 2025-06-17T16:21:39.577Z

hi everyone. i'm converting a langsmith-instrumented large project to arize. one convenience that langsmith has is that it displays what otel would call span attributes that are dictionaries in a convenient way in the dashboard. is there any equivalent way to do this in arize? or is json serialized strings the best we've got?

Is There an Equivalent to Span Attributes in Arize? | Arize AI Community

8 comments

· Sorted by Oldest

RunLLM
APP
·
Phoenix supports displaying span attributes that are dictionaries in a structured, nested JSON format in the dashboard, not just as raw JSON strings. The attributes are automatically un-flattened and presented as nested objects, making them more readable and convenient to explore in the UI, similar to LangSmith's approach. This is achieved by splitting attribute keys on the `.` separator and reconstructing them into nested dictionaries for display purposes, as described in <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> and implemented in the UI per <https://github.com/Arize-ai/phoenix/pull/1149|feat(trace): nested attributes>. For attributes that are JSON-serialized strings (such as certain metadata fields), Phoenix will attempt to parse and display them as structured JSON when possible, falling back to string display only if parsing fails. This behavior is documented in <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> and <https://github.com/Arize-ai/phoenix/pull/1107|feat(gql): span attributes as json string>. Would you like more detail or code examples on how to ensure your attributes are displayed this way? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> - <https://github.com/Arize-ai/phoenix/pull/1149|feat(trace): nested attributes> - <https://github.com/Arize-ai/phoenix/pull/1107|feat(gql): span attributes as json string>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
RunLLM can you show me a code example? if possible can it use span.set_attribute ?
RunLLM
APP
·
Yes, you can use `span.set_attribute` to create nested attributes by using dot notation in the key. Phoenix will display these as structured/nested objects in the UI. For example: ``` span.set_attribute("metadata.source", "search_engine") span.set_attribute("metadata.score", 0.98) span.set_attribute("metadata.details.author", "Alice") span.set_attribute("metadata.details.length", 1234) ``` This will appear in the Phoenix UI as a nested dictionary under the `metadata` key, with `source`, `score`, and a nested `details` object, as described in <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py>. Would you like a more complete example or further explanation? Sources: - <https://github.com/Arize-ai/phoenix/issues/7687|https://github.com/Arize-ai/phoenix/issues/7687> - <https://github.com/Arize-ai/openinference/blob/main/spec/semantic_conventions.md|https://github.com/Arize-ai/openinference/blob/main/spec/semantic_conventions.md>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
RunLLM are there utility methods to convert a python dictionary to this notation?
RunLLM
APP
·
Yes, Phoenix provides utility methods to flatten a nested Python dictionary into dot notation key-value pairs suitable for `span.set_attribute`. Use the `flatten` function from <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py>: ``` from phoenix.trace.attributes import flatten nested_dict = { "metadata": { "source": "search_engine", "details": {"author": "Alice", "length": 1234} } } for key, value in flatten(nested_dict): span.set_attribute(key, value) ``` Would you like more detail on how this function works or how to handle lists? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
RunLLM do i need to pip install something for this to work?
RunLLM
APP
·
Yes, you need to install the `arize-phoenix` package to access `phoenix.trace.attributes.flatten`. Install it with: ``` pip install arize-phoenix ``` This will provide the necessary modules, including the `flatten` utility. Would you like more details on installation or usage? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> - <https://arize-phoenix.readthedocs.io/en/latest/index.html|Arize Phoenix Python Reference>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
thank you, this worked perfectly, the llm did a good job here
:phoenix-party:1
🙌1

RunLLM
APP
·
Phoenix supports displaying span attributes that are dictionaries in a structured, nested JSON format in the dashboard, not just as raw JSON strings. The attributes are automatically un-flattened and presented as nested objects, making them more readable and convenient to explore in the UI, similar to LangSmith's approach. This is achieved by splitting attribute keys on the `.` separator and reconstructing them into nested dictionaries for display purposes, as described in <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> and implemented in the UI per <https://github.com/Arize-ai/phoenix/pull/1149|feat(trace): nested attributes>. For attributes that are JSON-serialized strings (such as certain metadata fields), Phoenix will attempt to parse and display them as structured JSON when possible, falling back to string display only if parsing fails. This behavior is documented in <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> and <https://github.com/Arize-ai/phoenix/pull/1107|feat(gql): span attributes as json string>. Would you like more detail or code examples on how to ensure your attributes are displayed this way? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> - <https://github.com/Arize-ai/phoenix/pull/1149|feat(trace): nested attributes> - <https://github.com/Arize-ai/phoenix/pull/1107|feat(gql): span attributes as json string>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
RunLLM can you show me a code example? if possible can it use span.set_attribute ?
RunLLM
APP
·
Yes, you can use `span.set_attribute` to create nested attributes by using dot notation in the key. Phoenix will display these as structured/nested objects in the UI. For example: ``` span.set_attribute("metadata.source", "search_engine") span.set_attribute("metadata.score", 0.98) span.set_attribute("metadata.details.author", "Alice") span.set_attribute("metadata.details.length", 1234) ``` This will appear in the Phoenix UI as a nested dictionary under the `metadata` key, with `source`, `score`, and a nested `details` object, as described in <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py>. Would you like a more complete example or further explanation? Sources: - <https://github.com/Arize-ai/phoenix/issues/7687|https://github.com/Arize-ai/phoenix/issues/7687> - <https://github.com/Arize-ai/openinference/blob/main/spec/semantic_conventions.md|https://github.com/Arize-ai/openinference/blob/main/spec/semantic_conventions.md>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
RunLLM are there utility methods to convert a python dictionary to this notation?
RunLLM
APP
·
Yes, Phoenix provides utility methods to flatten a nested Python dictionary into dot notation key-value pairs suitable for `span.set_attribute`. Use the `flatten` function from <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py>: ``` from phoenix.trace.attributes import flatten nested_dict = { "metadata": { "source": "search_engine", "details": {"author": "Alice", "length": 1234} } } for key, value in flatten(nested_dict): span.set_attribute(key, value) ``` Would you like more detail on how this function works or how to handle lists? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
RunLLM do i need to pip install something for this to work?
RunLLM
APP
·
Yes, you need to install the `arize-phoenix` package to access `phoenix.trace.attributes.flatten`. Install it with: ``` pip install arize-phoenix ``` This will provide the necessary modules, including the `flatten` utility. Would you like more details on installation or usage? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/src/phoenix/trace/attributes.py|src/phoenix/trace/attributes.py> - <https://arize-phoenix.readthedocs.io/en/latest/index.html|Arize Phoenix Python Reference>
💡 Hint: Mention RunLLM in the thread for followups.
Eugene E.
·
thank you, this worked perfectly, the llm did a good job here
:phoenix-party:1
🙌1