LangSmith
Each AI call, except its content. Model, prompt, tokens, cost and status. Prompt version, time to first token, tags, feedback scores and thread. Never inputs, outputs or error messages.
Access
- Kind
- A read-only key you paste
- What to give
- API key
- Why that one
- A LangSmith API key for the workspace. Valistry lists projects and reads the LLM runs' tokens and cost, with an explicit field list that carries no input and no output.
What it attributes on its own
- From the source
- day; project; model as LangSmith priced it; runs; input and output tokens; what LangSmith says they cost; runs that errored; latency and time to first token; prompt and settings; feedback scores
- Needs a rule or a trace
- the provider's invoice — LangSmith computes a cost, it does not bill you; team, workflow and agent unless a rule maps the project; outcome — feedback scores are a later slice
The addresses it calls
- https://api.smith.langchain.com
Valistry calls only these addresses. It blocks any other call and logs every call.
Connecting it
- Create a workspace service key. Switch to the workspace Valistry should read, then Create API Key. Open ↗
- Find your region. Look at your sign-in address. Pick its region's API address below.
LangSmith shows a cost; your provider bills it. Connecting both LangSmith and a provider it traces counts the same call twice — connect one. Pick LangSmith to see cost by project. Pick the provider to match the invoice.