How to Add Custom JSON Metadata in Phoenix Prompt Management

·Oct 31, 2025 09:24 PM·

Hello! I'm new to Phoenix and am currently doing a bake-off of Langfuse, Phoenix, etc. for work. I would like to maintain some metadata on prompts (title, description, roles, etc) to use when driving some UI, thereby externalizing this configuration for PMs to manage without needing a deploy. I can see how to do this in Langfuse, but I don't see how to provide user-supplied, arbitrary JSON metadata in Phoenix's prompt management. Can someone show me how or suggest a workaround? My best idea so far is to use YAML front mater in the prompt itself and strip this out in the application layer...

10 comments

· Sorted by Oldest

RunLLM
APP
·
·
Phoenix currently does not support attaching arbitrary user-supplied JSON metadata directly to prompts. While prompt versioning, tagging, and descriptions are supported, the ability to add custom metadata fields (like `roles` or arbitrary JSON) is still under development and not available in the UI or SDKs as of now. The only metadata fields available are those built into the prompt/version model, and prompt labels/metadata are marked as "under construction" in the docs and issues <https://arize.com/docs/phoenix/prompt-engineering/how-to-prompts/create-a-prompt#using-the-playground|[745000de>, <https://github.com/Arize-ai/phoenix/blob/main/docs/prompt-engineering/how-to-prompts/create-a-prompt.md|97ce3a10>, <https://github.com/Arize-ai/phoenix/issues/6290|6290>]. Your YAML front-matter workaround is a reasonable approach for now—store metadata in the prompt template and parse it out at runtime. Alternatively, you could encode metadata in the prompt description or naming convention, but this is less flexible. Would you like more detail or code examples for these workarounds? Sources: - <https://arize.com/docs/phoenix/prompt-engineering/how-to-prompts/create-a-prompt#using-the-playground|745000de> - <https://github.com/Arize-ai/phoenix/blob/main/docs/prompt-engineering/how-to-prompts/create-a-prompt.md|97ce3a10> - <https://github.com/Arize-ai/phoenix/issues/6290|6290>
💡 Hint: Mention RunLLM in the thread for followups.
Nathan P.
·
·
Bummer. That's a point for Langfuse 😞
Roger Y.
·
·
We actually already have metadata for prompts that are stored in the database (see query below), but they’re not yet exposed in the UI. We can enable this for you pretty quickly and see how it works for your use case.
Nathan P.
·
·
That would be great...although I'm not using Cloud. Due to regulatory requirements, we must be self-hosted. Is there a version of the Docker image, or something like that? (Feel free to move to DM...)
Roger Y.
·
·
yea. let me work this feature into the UI and when we release it it’ll be a new docker image you can download
Nathan P.
·
·
Sweet! I'm moving forward with the YAML-in-prompt approach so as not to be blocked, but will happily rip that out later...
Roger Y.
·
·
do you have a screenshot just so i understand exactly what you’re looking for in terms of UI? you can DM me if you like
Xander
External
·
·
Thanks for the feedback Nathan P.. I've filed an issue here.
👍1
Roger Y.
·
·
we have opened a PR for this issue
Roger Y.
·
·
this update has been released in v12.10
2

RunLLM
APP
·
·
Phoenix currently does not support attaching arbitrary user-supplied JSON metadata directly to prompts. While prompt versioning, tagging, and descriptions are supported, the ability to add custom metadata fields (like `roles` or arbitrary JSON) is still under development and not available in the UI or SDKs as of now. The only metadata fields available are those built into the prompt/version model, and prompt labels/metadata are marked as "under construction" in the docs and issues <https://arize.com/docs/phoenix/prompt-engineering/how-to-prompts/create-a-prompt#using-the-playground|[745000de>, <https://github.com/Arize-ai/phoenix/blob/main/docs/prompt-engineering/how-to-prompts/create-a-prompt.md|97ce3a10>, <https://github.com/Arize-ai/phoenix/issues/6290|6290>]. Your YAML front-matter workaround is a reasonable approach for now—store metadata in the prompt template and parse it out at runtime. Alternatively, you could encode metadata in the prompt description or naming convention, but this is less flexible. Would you like more detail or code examples for these workarounds? Sources: - <https://arize.com/docs/phoenix/prompt-engineering/how-to-prompts/create-a-prompt#using-the-playground|745000de> - <https://github.com/Arize-ai/phoenix/blob/main/docs/prompt-engineering/how-to-prompts/create-a-prompt.md|97ce3a10> - <https://github.com/Arize-ai/phoenix/issues/6290|6290>
💡 Hint: Mention RunLLM in the thread for followups.
Nathan P.
·
·
Bummer. That's a point for Langfuse 😞
Roger Y.
·
·
We actually already have metadata for prompts that are stored in the database (see query below), but they’re not yet exposed in the UI. We can enable this for you pretty quickly and see how it works for your use case.
Nathan P.
·
·
That would be great...although I'm not using Cloud. Due to regulatory requirements, we must be self-hosted. Is there a version of the Docker image, or something like that? (Feel free to move to DM...)
Roger Y.
·
·
yea. let me work this feature into the UI and when we release it it’ll be a new docker image you can download
Nathan P.
·
·
Sweet! I'm moving forward with the YAML-in-prompt approach so as not to be blocked, but will happily rip that out later...
Roger Y.
·
·
do you have a screenshot just so i understand exactly what you’re looking for in terms of UI? you can DM me if you like
Xander
External
·
·
Thanks for the feedback Nathan P.. I've filed an issue here.
👍1
Roger Y.
·
·
we have opened a PR for this issue
Roger Y.
·
·
this update has been released in v12.10
2