README.md

---
license: apache-2.0
tags:
- agentcap
- agentcap-captures
---

# dacorvo/transformers-coding-session-captures

HTTP captures of agent ↔ model interactions — one parquet row per
`/v1/chat/completions` call. Produced by
[agentcap](https://github.com/huggingface/agentcap).

Native session traces for the same runs live in companion datasets
named `transformers-coding-session-<agent>-traces`. They're all grouped under the
[transformers-coding-session Collection](https://huggingface.co/dacorvo)
alongside this dataset. Join on `run_id`.

## Loading

```python
from datasets import load_dataset

ds = load_dataset("dacorvo/transformers-coding-session-captures", split="train")
```

## Schema

| column | description |
|---|---|
| `run_id` | agentcap run id; matches the per-run folder in the traces dataset |
| `request_id` | UUID minted by the capture proxy |
| `model` | Model id from the captured request body |
| `captured_at` | Epoch seconds when the request was captured |
| `request` | Raw OpenAI request body, JSON-stringified |
| `response` | Raw OpenAI response body, JSON-stringified (or `{"stream": true, "raw": ...}` for SSE) |
| `served_by` | Per-response `X-Served-By` header (HF Router sub-provider routing) |
| `served_build_info` | Per-response `X-Build-Info` header |
| `served_model` | Per-response body-echoed `model` |
| `provider` | Derived from the proxy upstream URL (constant per file) |
| `upstream_url` | Proxy upstream URL at capture time (constant per file) |

`request` and `response` are JSON strings; consumers `json.loads(...)`
them. To recover per-message token ranges, render `request.messages`
through the model's chat template yourself —
`transformers.AutoTokenizer.apply_chat_template`.