litellm - 💡(How to fix) Fix [Bug]: Analytics API returns 0 tokens despite token usage data correctly stored in database [1 comments, 2 participants]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

What happened?

Environment

• Open WebUI version: v0.8.12
• Installation: Docker
• Backend: LiteLLM proxy → Azure AI Foundry (OpenAI-compatible)
• Database: SQLite (default)

Description

The Analytics API endpoint /api/v1/analytics/tokens returns 0 for all token counts across all models, despite token usage data being correctly stored in the chat_message table. The per-message token counter in the chat UI works correctly and displays accurate counts.

Steps to Reproduce

1. Configure Open WebUI with an OpenAI-compatible backend (LiteLLM or Azure OpenAI)
2. Enable usage tracking via environment variables:

DEFAULT_MODEL_METADATA: '{"capabilities": {"usage": true}}'
DEFAULT_MODEL_PARAMS: '{"stream_options": {"include_usage": true}}'

1. Send several chat messages — token counts appear correctly below each response in the UI
2. Navigate to Admin Panel → Analytics — Total Tokens shows 0
3. Call /api/v1/analytics/tokens directly — all models show input_tokens: 0, output_tokens: 0, total_tokens: 0

Verification — Data Exists in DB

Direct SQLite query confirms data is present and correctly structured:

SELECT 
    model_id,
    SUM(json_extract(usage, '$.prompt_tokens')) as prompt,
    SUM(json_extract(usage, '$.completion_tokens')) as completion,
    SUM(json_extract(usage, '$.total_tokens')) as total,
    COUNT(*) as messages
FROM chat_message
WHERE usage IS NOT NULL
AND usage != 'null'
AND json_extract(usage, '$.total_tokens') > 0
GROUP BY model_id;

Result:

('company-assist-fast', 110, 645, 755, 5)

Analytics API Response (same model)

{
  "model_id": "company-assist-fast",
  "input_tokens": 0,
  "output_tokens": 0,
  "total_tokens": 0,
  "message_count": 23
}

Note: message_count is correct (23) but all token values are 0. This confirms the analytics router can find the messages but fails to aggregate token data from the usage JSON column.

Sample usage JSON Stored in DB

{
  "completion_tokens": 195,
  "prompt_tokens": 57,
  "total_tokens": 252,
  "completion_tokens_details": {
    "accepted_prediction_tokens": 0,
    "audio_tokens": 0,
    "reasoning_tokens": 0,
    "rejected_prediction_tokens": 0
  },
  "prompt_tokens_details": {
    "audio_tokens": 0,
    "cached_tokens": 0
  }
}

Expected Behaviour

/api/v1/analytics/tokens should return aggregated token counts matching what is stored in the chat_message.usage JSON column.

Actual Behaviour

All token values return 0 despite data being present and correctly structured in the database. Message counts are correctly returned, indicating the query reaches the right rows but fails to extract the JSON token values.

Possible Cause

The analytics router query may not be correctly using json_extract() on the usage column, or there may be a mismatch in how JSON aggregation is handled for the usage field compared to how message counts are aggregated.

Steps to Reproduce

1. Configure Open WebUI with an OpenAI-compatible backend (LiteLLM or Azure OpenAI)
2. Enable usage tracking via environment variables:

DEFAULT_MODEL_METADATA: '{"capabilities": {"usage": true}}'
DEFAULT_MODEL_PARAMS: '{"stream_options": {"include_usage": true}}'

1. Send several chat messages — token counts appear correctly below each response in the UI
2. Navigate to Admin Panel → Analytics — Total Tokens shows 0
3. Call /api/v1/analytics/tokens directly — all models show input_tokens: 0, output_tokens: 0, total_tokens: 0

Relevant log output

What part of LiteLLM is this about?

UI Dashboard

What LiteLLM version are you on ?

v1.82.6

Twitter / LinkedIn details

No response