#!/usr/bin/env bash
# embed.sh — Generate an embedding vector from text via the configured embedding endpoint.
#
# Defaults to local llama-server (EMBED_BASE_URL in .env).
# Set EMBED_API_KEY in .env for remote providers that require auth.
#
# Usage:
#   ./embed.sh "text to embed"
#   echo "text to embed" | ./embed.sh -
#
# Output: JSON array of floats (the embedding vector)

. "$(dirname "$0")/common.sh"

# Read input
if [ "${1:-}" = "-" ]; then
    INPUT_TEXT=$(cat)
elif [ -n "${1:-}" ]; then
    INPUT_TEXT="$1"
else
    echo "Usage: embed.sh <text> | embed.sh -" >&2
    exit 1
fi

# Escape for JSON
JSON_INPUT=$(python3 -c "import json,sys; print(json.dumps(sys.argv[1]))" "$INPUT_TEXT")

# Call embedding API (OpenAI-compatible: local llama-server or remote)
RESPONSE=$(curl -s -X POST "$EMBED_API_URL" \
    -H "Authorization: Bearer ${EMBED_API_KEY:-}" \
    -H "Content-Type: application/json" \
    -d "{\"model\": \"$EMBED_MODEL\", \"input\": $JSON_INPUT, \"dimensions\": $EMBED_DIMENSIONS}")

# Extract embedding, error if missing
python3 -c "
import json, sys
r = json.loads(sys.argv[1])
if 'error' in r:
    print(f'API error: {r[\"error\"]}', file=sys.stderr)
    sys.exit(1)
if 'data' not in r or len(r['data']) == 0:
    print(f'Unexpected response: {json.dumps(r)[:200]}', file=sys.stderr)
    sys.exit(1)
print(json.dumps(r['data'][0]['embedding']))
" "$RESPONSE"