feat(ai): LLM metrics tracking and AI span inspector#3213

Merged

ericallam merged 30 commits intomainfrom

feature/tri-7773-llm-cost-tracking

Mar 17, 2026

Commits on Mar 17, 2026

feat: add automatic LLM cost tracking for GenAI spans
ericallam
committed
feat: add friendlyId to LlmModel, TRQL llm_usage integration, and seed-on-startup
ericallam
committed
feat: add metadata map to llm_usage_v1 for cost attribution by user/tenant
ericallam
committed
New AI streamText span sidebar
ericallam
committed
A bunch of improvements around extracting AI data
ericallam
committed
Fixed a bunch of coderabbit issues
ericallam
committed
fix some typescript errors
ericallam
committed
some fixes
ericallam
committed
A bunch of fixes
ericallam
committed
Copy raw properties footer behind isAdmin

samejr
authored and
ericallam
committed
Nicer Tools tab icon and tabs scroll behaviour

samejr
authored and
ericallam
committed
Display details as a property table instead of pills

samejr
authored and
ericallam
committed
Use proper paragraph component

samejr
authored and
ericallam
committed
Style improvements to the messages

samejr
authored and
ericallam
committed
Renamed llm_usage_v1 to llm_metrics_v1, added some additional fields to track as well
ericallam
committed
Create AI/LLM metrics dashboard, with support for a model filter
ericallam
committed
Lots more example content in the spans seed

samejr
authored and
ericallam
committed
New model logos

samejr
authored and
ericallam
committed
Default wider side panel to better view markdown

samejr
authored and
ericallam
committed
Render JSON object output in our code block component rather than as a string

samejr
authored and
ericallam
committed
Use streamdown for system prompts

samejr
authored and
ericallam
committed
toolCall doesn’t get wrapped in chat div

samejr
authored and
ericallam
committed
fixed failing tests because of icon name changes
ericallam
committed
strip crumbs
ericallam
committed
give the llm model registry a short window to become ready when the server first tries to enrich spans so we don't miss enriching the first Xms of spans every time a server boots
ericallam
committed
A few comments
ericallam
committed
Add env vars for configuring the llm_metrics_v1 flush scheudler settings
ericallam
committed
as string bad
ericallam
committed
register default model prices
ericallam
committed
add llm_metrics to the get_query_schema tool desc
ericallam
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(ai): LLM metrics tracking and AI span inspector#3213

feat(ai): LLM metrics tracking and AI span inspector#3213
ericallam merged 30 commits intomainfrom
feature/tri-7773-llm-cost-tracking

Commits on Mar 17, 2026

feat: add automatic LLM cost tracking for GenAI spans

feat: add friendlyId to LlmModel, TRQL llm_usage integration, and seed-on-startup

feat: add metadata map to llm_usage_v1 for cost attribution by user/tenant

New AI streamText span sidebar

A bunch of improvements around extracting AI data

Fixed a bunch of coderabbit issues

fix some typescript errors

some fixes

A bunch of fixes

Copy raw properties footer behind isAdmin

Nicer Tools tab icon and tabs scroll behaviour

Display details as a property table instead of pills

Use proper paragraph component

Style improvements to the messages

Renamed llm_usage_v1 to llm_metrics_v1, added some additional fields to track as well

Create AI/LLM metrics dashboard, with support for a model filter

Lots more example content in the spans seed

New model logos

Default wider side panel to better view markdown

Render JSON object output in our code block component rather than as a string

Use streamdown for system prompts

toolCall doesn’t get wrapped in chat div

fixed failing tests because of icon name changes

strip crumbs

give the llm model registry a short window to become ready when the server first tries to enrich spans so we don't miss enriching the first Xms of spans every time a server boots

A few comments

Add env vars for configuring the llm_metrics_v1 flush scheudler settings

as string bad

register default model prices

add llm_metrics to the get_query_schema tool desc