improved free-models-collector with liveness checks

free-model-collector.sh

@@ -11,7 +11,7 @@ TCONFIG="$(mktemp)"
 $REQ https://opencode.ai/zen/v1/models | jq -r '.data[] | select(.id | test("-free")) | "\(.id),-,https://opencode.ai/zen/v1,"' >> "$TCONFIG"
 
 # Write Kilo free models
-$REQ https://api.kilo.ai/api/openrouter/models | jq -r '.data[] | select(.isFree) | "\(.id),-,https://api.kilo.ai/api/openrouter,"' >> "$TCONFIG"
+$REQ https://api.kilo.ai/api/gateway/models | jq -r '.data[] | select(.isFree) | "\(.id),-,https://api.kilo.ai/api/gateway,"' >> "$TCONFIG"
 
 # Write OVHCloud free trial models (2 req per minute per IP keyless), commented out by default
 # $REQ https://oai.endpoints.kepler.ai.cloud.ovh.net/v1/models | jq -r '.data[] | "\(.id),-,https://oai.endpoints.kepler.ai.cloud.ovh.net,"' | grep -Ev 'tts-|whisper-|bge-' >> "$TCONFIG"
@@ -22,8 +22,8 @@ $REQ https://api.llm7.io/v1/models | jq -r '.data[] | select(.usage_based_only |
 # Write UncloseAI models
 $REQ https://hermes.ai.unturf.com/v1/models | jq -r '.data[] | "\(.id),-,https://hermes.ai.unturf.com/v1,"' >> "$TCONFIG"
 
-# Write AIHorde models, slow but community-powered
-$REQ https://oai.aihorde.net/v1/models | jq -r '.data[] | "\(.id),0000000000,https://oai.aihorde.net/v1,"' >> "$TCONFIG"
+# Write AIHorde models, slow but community-powered, commented out by default
+# $REQ https://oai.aihorde.net/v1/models | jq -r '.data[] | "\(.id),0000000000,https://oai.aihorde.net/v1,"' >> "$TCONFIG"
 
 # Write Pekpik models (need to be rotated frequently, highly experimental, commented out by default)
 # $REQ https://aiapiv2.pekpik.com/free-keys.json | jq -r '.keys[] | "\(.model),\(.key),https://aiapiv2.pekpik.com/v1,"' >> "$TCONFIG"
@@ -42,10 +42,74 @@ $REQ https://zsnonsky-nim-proxy.hf.space/v1/models | jq -r '.data[] | "\(.id),-,
 # Write XORTRON as hardcode
 echo 'xortron,-,https://darkc0de-chat.hf.space/v1,'>> "$TCONFIG"
 
+# Verify each collected model for responsiveness under the actual /chat/completions
+# (prompt it to respond with OK). Unresponsive models are dropped from the final file.
+# For non-timeout drops, the HTTP response code is reported for diagnostics.
+VCONFIG="$(mktemp)"
+SKIPF="$(mktemp)"
+while IFS=',' read -r model key endpoint extra; do
+  [ -z "$model" ] && continue
+  case "$model" in *embed*) echo "Skipped (embedding model): $model @ $endpoint" >&2; continue;; esac
+  if grep -qxF "$endpoint" "$SKIPF" 2>/dev/null; then
+    echo "Skipped (endpoint unreachable): $model @ $endpoint" >&2
+    continue
+  fi
+  auth=''
+  [ "$key" != '-' ] && auth="-H Authorization: Bearer $key"
+  payload="$(jq -nc --arg m "$model" '{model:$m,messages:[{role:"user",content:"Reply with the word OK only."}],max_tokens:8,stream:true}')"
+  bodyf="$(mktemp)"
+  attempt=0
+  max_attempts=3
+  ok=0
+  reason=''
+  while [ "$attempt" -lt "$max_attempts" ]; do
+    attempt=$((attempt+1))
+    out="$(curl -sS -m 8 -o "$bodyf" -w '%{http_code}|%{time_starttransfer}' \
+         -X POST "$endpoint/chat/completions" \
+         -H 'Content-Type: application/json' $auth \
+         -d "$payload" 2>/dev/null)"
+    rc=$?
+    http_code="${out%%|*}"
+    ttft="${out##*|}"
+    if [ "$http_code" -ge 200 ] && [ "$http_code" -lt 300 ] && awk "BEGIN{exit !($ttft>0)}"; then
+      ok=1
+      break
+    fi
+    if [ "$rc" -eq 28 ]; then
+      reason="Timeout error (no first token within 8s)"
+    elif [ "$rc" -ne 0 ]; then
+      reason="curl error $rc"
+    else
+      reason="HTTP $http_code"
+    fi
+    retryable=0
+    if [ "$rc" -eq 28 ] || [ "$rc" -ne 0 ] || [ "$http_code" -eq 429 ] || { [ "$http_code" -ge 500 ] && [ "$http_code" -lt 600 ]; }; then
+      retryable=1
+    fi
+    if [ "$retryable" -eq 1 ] && [ "$attempt" -lt "$max_attempts" ]; then
+      sleep 3
+    else
+      break
+    fi
+  done
+  if [ "$ok" -eq 1 ]; then
+    echo "Added: $model @ $endpoint (TTFT: ${ttft}s)"
+    printf '%s,%s,%s,%s\n' "$model" "$key" "$endpoint" "$extra" >> "$VCONFIG"
+  else
+    echo "Unresponsive ($reason), dropped: $model @ $endpoint" >&2
+    if [ "$rc" -eq 6 ] || [ "$rc" -eq 7 ]; then
+      printf '%s\n' "$endpoint" >> "$SKIPF"
+    fi
+  fi
+  rm -f "$bodyf"
+done < "$TCONFIG"
+rm -f "$TCONFIG"
+rm -f "$SKIPF"
+
 # Write the config header
 printf 'model,key,endpoint,extra\n' > "$CONFIG"
 # Finalize the config
-sort -u "$TCONFIG" | grep -v embed >> "$CONFIG"
-rm -f "$TCONFIG"
+sort -u "$VCONFIG" >> "$CONFIG"
+rm -f "$VCONFIG"
 
 exit 0