LLM audit · Client Path Docs

The LLM audit page at Settings > LLM > Audit gives you a full log of every LLM call made by Client Path on your workspace's behalf. It is useful for understanding costs, debugging unexpected results, and verifying that routing is working as intended.

What the audit log shows

Each entry records:

Timestamp: when the call was made.
Pipeline role: which step in the research pipeline made the call (web search, synthesis, scoring, outreach, etc.).
Provider and model: which LLM was used.
Tokens in / out: input and output token counts.
Latency: time from request to response.
Status: success or error.
Fallback: whether this call fell back to a secondary provider because the primary failed or was unavailable.

The LLM audit page showing a table of recent calls with provider labels, token counts, and fallback badges

Fallback rate

The fallback rate is shown in the audit page summary. It is the percentage of calls that used a fallback provider instead of the primary.

A high fallback rate means your primary provider is frequently unavailable or erroring. Causes include:

Rate limiting from the provider.
Network issues between Client Path and the provider.
An invalid or expired API key.

Check the provider's status page and verify your API key if the fallback rate is consistently above 5-10%.

Using the audit for cost management

The token counts in the audit log let you estimate costs per run or per pipeline role. If a particular role is consuming disproportionate tokens, you may want to switch it to a more cost-efficient model via LLM routing.

Log retention

Audit entries are retained for 90 days by default.