metergraphrelay 0.1.0

Move LLM trace data between systems: pull from providers, push to metergraph.

metergraphrelay

Export OpenAI stored chat completions as trace records, and optionally generate demo conversations to try it out.

Purpose

Download the chat completions already stored in your own OpenAI account (completions created with store=True) as local JSONL trace records — useful for auditing, backups, or feeding into other tooling. This tool only reads what's associated with the API key you provide; it can't see or export anyone else's data.

Built on OpenAI's List Chat Completions API, which returns completions that were created with store: true. See the official OpenAI API reference for the underlying data model.

Setup

pip install -e ".[dev]"
cp .env.example .env
# edit .env and set OPENAI_API_KEY

Usage

Generate 1-2 demo conversations (stored with store=True) so there's data to pull:

metergraphrelay demo openai

Pull the 10 most recent stored chat completions into traces.jsonl, already shaped as metergraph trace records:

metergraphrelay pull openai

Options:

metergraphrelay pull openai -n 25 --output my-traces.jsonl --stdout --include-content --route my-app/support-bot
metergraphrelay demo openai --model gpt-4o-mini

Push a local JSONL file of traces to metergraph:

metergraphrelay push traces.jsonl

pull anthropic and pull langfuse accept the same shape but are not yet implemented in this version — they check for ANTHROPIC_API_KEY / LANGFUSE_PUBLIC_KEY+LANGFUSE_SECRET_KEY and report accordingly.

All subcommands accept --env-file PATH to point at a config file other than ./.env.

Using it against your real system (not just the demo)

metergraphrelay pull openai only finds completions that were created with store=True. The demo openai subcommand sets that flag for you, but for your own application to show up in a pull, its own OpenAI calls need the same flag. The only change required is adding store=True (and, optionally, metadata to help you tell requests apart later) to calls you're already making:

from openai import OpenAI

client = OpenAI()

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "..."}],
    store=True,                       # <-- persists this completion for later pulling
    metadata={"source": "my-app"},    # optional: filter/identify later via the API
)

No other code changes are needed — the request and response are handled exactly as before. Once your app is sending store=True, its completions become visible to metergraphrelay pull openai (or to the List Chat Completions API directly) using the same API key.

Two things worth knowing before flipping this on in production:

• Stored completions include full request/response content by default, so treat them as you would any other place your data is retained.
• Storage has no automatic expiry from this tool's side — deletion is a separate API call (Delete chat completion), not something metergraphrelay currently does.

Trace record shape

Each line of pull openai's output is a JSON object already shaped for metergraph's ingest API:

{
  "ts": "2026-07-30T12:00:00+00:00",
  "provider": "openai",
  "model": "gpt-4o-mini",
  "status": "success",
  "endpoint": "chat.completions",
  "input_tokens": 12,
  "output_tokens": 34,
  "error": false,
  "error_type": null,
  "request_id": "chatcmpl-...",
  "tags": {},
  "route": "openai/backfill",
  "content_opted_in": false,
  "request_json": null,
  "response_text": null,
  "sdk": "metergraphrelay",
  "sdk_version": "0.1.0"
}

request_json/response_text are populated only when --include-content is passed.

Every completion returned by the stored-completions list already succeeded, so status is always "success". error/error_type flag a partial record: --include-content was requested but the follow-up message fetch failed, so token counts are still real while the content is missing.

Running tests

pytest