Weights & Biases — Weave¶

GigaFlow reads Weave calls via the W&B trace-server API. A template transform ships as gigaflow/transforms/wb_weave.yml — leave the transform blank in gigaflow setup to use it, but you'll likely need to tailor filter rules to your op names (Weave has no structural span-type field).

Prerequisites¶

WANDB_API_KEY (wandb.ai → Settings → API keys) — sent as HTTP Basic api:<key>.
Your Weave project id as entity/project (e.g. my-team/my-agent).
Base URL https://trace.wandb.ai (default; override for self-hosted).

Connect (API)¶

PID=$(curl -s -X POST "$GIGAFLOW_BACKEND_URL/projects/" -H "Authorization: Bearer $GIGAFLOW_API_KEY" \
  -H 'Content-Type: application/json' -d '{"name":"my-weave-project"}' | python3 -c 'import sys,json;print(json.load(sys.stdin)["project_id"])')

# upload your custom transform first, then:
curl -X POST "$GIGAFLOW_BACKEND_URL/datasources/" -H "Authorization: Bearer $GIGAFLOW_API_KEY" \
  -H 'Content-Type: application/json' -d "{
    \"project_id\": \"$PID\",
    \"name\": \"my-weave\",
    \"source_type\": \"wb_weave\",
    \"connection_url\": \"https://trace.wandb.ai\",
    \"source_table\": \"<entity>/<project>\",
    \"api_key\": \"<WANDB_API_KEY>\"
  }"

Transform¶

Weave calls normalize to nested dicts with attributes.*, inputs, output, op_name, display_name. Author a transform mapping these to primitives (template: gigaflow/transforms/arize_phoenix.yml). Upload via PUT /projects/$PID/transform.

Run¶

gigaflow sync
gigaflow compute "SELECT trace_id FROM trace_metrics WHERE run_id IS NULL"
gigaflow inspect <trace_id>