LLM call recording
Add one line to the SDK and every OpenAI / Anthropic / Gemini / Mistral call is recorded — cost, latency, and full errors, traceable one by one.
Demo: Recent calls
Click a row to expand
| Time | Model | Cost | Latency | Status |
|---|---|---|---|---|
| 14:32:18 | gpt-5 | <$0.01 | 1.14 s | ● Success |
| 14:32:09 | claude-sonnet-4-6 | <$0.01 | 892 ms | ● Success |
| 14:31:54 | gemini-2.5-pro | — | 322 ms | ● Error |
| 14:31:42 | gpt-5-mini | <$0.01 | 412 ms | ● Success |
| 14:30:51 | mistral-large | <$0.01 | 716 ms | ● Success |
Prompt
Reply politely and concisely to a customer who says they cannot log in.
Completion
Sure, let me check the situation. Could you try resetting your password from the link below?
Servicesupport-bot
Useruser_a12f
Sessionsess_8f12
TokensInput 280 / Output 96
Latency1.14 s
4 providers supported
Wrap the OpenAI / Anthropic / Gemini / Mistral SDK clients. Argosvix records the rest automatically.
Per-call detail you can trace
Token counts, USD cost, latency, and errors — per call, in one view. Full prompt/completion storage is an opt-in Pro+ option (all 4 providers).
Dashboard + MCP + REST
Reach the same data from the web UI, from Claude Desktop / Cursor, or from curl.