adding better exporting

README.md

@@ -37,7 +37,7 @@ The global context limit defaults to 256k tokens and can be changed with `MAX_CO
 
 Inference requests are limited per client IP to 2 requests per second, 100 requests per five hours, and $10 of reported upstream cost per five hours. Railway's `X-Real-IP` header is used to identify clients.
 
-Set `DATABASE_URL` in `.env` (or the process environment) to a Postgres connection string. The proxy creates its `requests` table and index automatically on startup. Every incoming request and its complete response—including streaming responses and errors—is stored with headers, status, model, client IP, and timestamp. Successful inference requests also store a normalized training conversation containing the full chat history and generated assistant output, including reasoning, tool calls, and tool results.
+Set `DATABASE_URL` in `.env` (or the process environment) to a Postgres connection string. The proxy creates its `requests` table and index automatically on startup. Only requests to the OpenAI Chat Completions and Responses endpoints and the Anthropic Messages endpoints are stored. Their complete responses—including streaming responses and errors—are stored with headers, status, model, client IP, and timestamp. On startup, stored requests for all other endpoints are deleted. Successful chat completion requests also store a normalized training conversation containing the full chat history and generated assistant output, including reasoning, tool calls, and tool results.
 
 ```sh
 npm start
@@ -67,7 +67,9 @@ For Claude Code, either set `ANTHROPIC_BASE_URL=http://localhost:4005/ant` or us
 
 ## Exporting training data
 
-Exports use JSONL: each line is one inference request containing its complete input history followed by the generated assistant message. The normalized OpenAI-style `messages` preserve system/user/assistant roles, `reasoning_content`, assistant `tool_calls`, and `tool` results. Non-chat routes and failed inference requests are excluded.
+Each export creates two JSONL files. The regular `export-<timestamp>.jsonl` remains unchanged: each line is one inference request containing its complete input history followed by the generated assistant message. The additional `export-<timestamp>-traces.jsonl` removes cumulative intermediate snapshots and keeps each complete conversation branch. If a conversation is rewound and continued in multiple ways, every branch leaf is retained as its own full trace.
+
+Chat Completions, Responses, and Anthropic Messages requests are all exported in the same normalized OpenAI-style `messages` format. It preserves system/user/assistant roles, `reasoning_content`, assistant `tool_calls`, and `tool` results. Non-chat routes and failed inference requests are excluded.
 
 Each line also has `metadata` with the exact tool definitions supplied on that request, tool choice, model, API format, endpoint, streaming mode, generation parameters, caller-supplied request metadata, response ID/model, finish reason, usage, HTTP status, duration, request ID, and timestamp. Tool metadata remains in its original OpenAI, Responses, or Anthropic format so no provider-specific schema information is lost.
 
@@ -78,17 +80,17 @@ Each line also has `metadata` with the exact tool definitions supplied on that r
 Omit the limit to export every training request, newest first:
 
 ```sh
-npm run --silent export > requests.jsonl
+npm run --silent export
 ```
 
 Pass a positive limit to export that many of the most recent training requests:
 
 ```sh
-npm run --silent export -- 100 > recent-requests.jsonl
+npm run --silent export -- 100
 # Equivalent: npm run --silent export -- --limit 100
 ```
 
-The export command uses the same `DATABASE_URL` as the server. `--silent` suppresses npm's banner, and the script's progress message is written to standard error, so redirected JSONL remains valid. Rows collected before normalized training storage was added are normalized from their saved raw request and response during export.
+The export command uses the same `DATABASE_URL` as the server and saves both files in the project root. The limit applies to the regular per-request export; the traces file contains the complete branch leaves found among those requests. Rows collected before normalized training storage was added are normalized from their saved raw request and response during export.
 
 ## Tests