diff --git a/litellm/config.yaml b/litellm/config.yaml
deleted file mode 100644
index b8b6331..0000000
--- a/litellm/config.yaml
+++ /dev/null
@@ -1,233 +0,0 @@
-model_list:
-  # ═══════════════════════════════════════════════
-  # TIER 1: Free providers (try first)
-  # ═══════════════════════════════════════════════
-
-  # --- Groq (free tier, very fast) ---
-  - model_name: llama-3.3-70b
-    litellm_params:
-      model: groq/llama-3.3-70b-versatile
-      api_key: os.environ/GROQ_API_KEY
-
-  # --- Cerebras (free tier, very fast) ---
-  - model_name: llama-3.3-70b-cerebras
-    litellm_params:
-      model: cerebras/llama-3.3-70b
-      api_key: os.environ/CEREBRAS_API_KEY
-
-  # --- OpenRouter free models ---
-  - model_name: deepseek-v3-free
-    litellm_params:
-      model: openrouter/deepseek/deepseek-chat-v3-0324:free
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  # ═══════════════════════════════════════════════
-  # TIER 2: DeepSeek V3.2 (cheapest first)
-  # ═══════════════════════════════════════════════
-
-  # DeepSeek V3.2 via DeepInfra ($0.26 in / $0.38 out per M)
-  - model_name: deepseek-v3.2
-    litellm_params:
-      model: deepinfra/deepseek-ai/DeepSeek-V3.2
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # DeepSeek V3.2 fallback via SiliconFlow ($0.27 in / $0.42 out per M)
-  - model_name: deepseek-v3.2
-    litellm_params:
-      model: openai/deepseek-ai/DeepSeek-V3.2
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  # ═══════════════════════════════════════════════
-  # TIER 3: Ultra-cheap DeepInfra models
-  # ═══════════════════════════════════════════════
-
-  # GPT-OSS-120B — OpenAI open-weight MoE ($0.05 in / $0.24 out per M)
-  - model_name: gpt-oss
-    litellm_params:
-      model: deepinfra/openai/gpt-oss-120b
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # GPT-OSS-20B — lower latency variant ($0.04 in / $0.16 out per M)
-  - model_name: gpt-oss-20b
-    litellm_params:
-      model: deepinfra/openai/gpt-oss-20b
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # Nemotron Super 49B — near-flagship quality ($0.10 in / $0.40 out per M)
-  - model_name: nemotron-super
-    litellm_params:
-      model: deepinfra/nvidia/Llama-3.3-Nemotron-Super-49B-v1.5
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # Nemotron Nano 9B — dirt cheap for simple tasks ($0.04 in / $0.16 out per M)
-  - model_name: nemotron-nano
-    litellm_params:
-      model: deepinfra/nvidia/NVIDIA-Nemotron-Nano-9B-v2
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # ═══════════════════════════════════════════════
-  # TIER 4: Other DeepInfra models
-  # ═══════════════════════════════════════════════
-
-  - model_name: deepseek-r1
-    litellm_params:
-      model: deepinfra/deepseek-ai/DeepSeek-R1
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  - model_name: devstral
-    litellm_params:
-      model: deepinfra/mistralai/Devstral-Small-2505
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # ═══════════════════════════════════════════════
-  # TIER 5: GLM models (cheapest first)
-  # ═══════════════════════════════════════════════
-
-  # GLM-4.6 via DeepInfra ($0.60 in / $1.90 out per M)
-  - model_name: glm-4.6
-    litellm_params:
-      model: deepinfra/zai-org/GLM-4.6
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # GLM-4.7 via DeepInfra ($0.40 in / $1.75 out per M)
-  - model_name: glm-4.7
-    litellm_params:
-      model: deepinfra/zai-org/GLM-4.7
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # GLM-4.7 fallback via SiliconFlow
-  - model_name: glm-4.7
-    litellm_params:
-      model: openai/THUDM/GLM-4-32B-0414
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  # GLM-5 via DeepInfra ($0.80 in / $2.56 out per M)
-  - model_name: glm-5
-    litellm_params:
-      model: deepinfra/zai-org/GLM-5
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # ═══════════════════════════════════════════════
-  # TIER 6: Kimi K2 (cheapest first)
-  # ═══════════════════════════════════════════════
-
-  # Kimi K2 via DeepInfra ($0.50 in / $2.00 out per M)
-  - model_name: kimi-k2
-    litellm_params:
-      model: deepinfra/moonshotai/Kimi-K2-Instruct-0905
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # Kimi K2 fallback via SiliconFlow ($0.58 in / $2.29 out per M)
-  - model_name: kimi-k2
-    litellm_params:
-      model: openai/moonshotai/Kimi-K2-Instruct-0905
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  # ═══════════════════════════════════════════════
-  # TIER 7: SiliconFlow (Qwen)
-  # ═══════════════════════════════════════════════
-
-  # Qwen3 Coder 480B MoE via SiliconFlow ($1.14 in / $2.28 out per M)
-  - model_name: qwen3-coder
-    litellm_params:
-      model: openai/Qwen/Qwen3-Coder-480B-A35B-Instruct
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  # Qwen3 Coder 30B — cheaper alternative for simpler tasks
-  - model_name: qwen3-coder-30b
-    litellm_params:
-      model: openai/Qwen/Qwen3-Coder-30B-A3B-Instruct
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  # ═══════════════════════════════════════════════
-  # TIER 8: OpenRouter (most expensive, widest selection)
-  # ═══════════════════════════════════════════════
-
-  # Kimi K2.5 — DeepInfra is cheapest ($0.45 in / $2.25 out per M)
-  - model_name: kimi-k2.5
-    litellm_params:
-      model: deepinfra/moonshotai/Kimi-K2.5
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # Kimi K2.5 fallback via OpenRouter
-  - model_name: kimi-k2.5
-    litellm_params:
-      model: openrouter/moonshotai/kimi-k2.5
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: minimax-m2.5
-    litellm_params:
-      model: openrouter/minimax/minimax-m2.5
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: gpt-4.1-mini
-    litellm_params:
-      model: openrouter/openai/gpt-4.1-mini
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: gemini-3-flash-preview
-    litellm_params:
-      model: openrouter/google/gemini-3-flash-preview
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: trinity-large-preview
-    litellm_params:
-      model: openrouter/arcee-ai/trinity-large-preview
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  # --- OpenRouter premium models ---
-  - model_name: gemini-2.5-pro
-    litellm_params:
-      model: openrouter/google/gemini-2.5-pro-preview
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: claude-sonnet
-    litellm_params:
-      model: openrouter/anthropic/claude-sonnet-4
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: gpt-4.1
-    litellm_params:
-      model: openrouter/openai/gpt-4.1
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  # DeepSeek V3.2 last-resort fallback via OpenRouter
-  - model_name: deepseek-v3.2
-    litellm_params:
-      model: openrouter/deepseek/deepseek-chat-v3-0324
-      api_key: os.environ/OPENROUTER_API_KEY
-
-general_settings:
-  master_key: os.environ/LITELLM_MASTER_KEY
-
-  # ── Model group fallbacks (when model misbehaves mid-stream) ──
-  fallbacks:
-    - deepseek-v3.2: [gpt-oss, kimi-k2]
-    - gpt-oss: [deepseek-v3.2, nemotron-super]
-    - kimi-k2: [deepseek-v3.2, gpt-oss]
-    - kimi-k2.5: [deepseek-v3.2, gpt-oss]
-    - glm-4.7: [deepseek-v3.2, gpt-oss]
-    - glm-5: [deepseek-v3.2, gpt-oss]
-
-litellm_settings:
-  drop_params: true
-  set_verbose: false
-  num_retries: 2
-  request_timeout: 600
-
-  # ── Response caching via Valkey (reuses SearXNG's instance) ──
-  cache: true
-  cache_params:
-    type: redis
-    host: valkey
-    port: 6379
-    ttl: 3600
-
-  # ── Budget limit: $3/day to prevent surprise bills ──
-  # max_budget: 3.0
-  # budget_duration: "1d"
diff --git a/new-api/CHANNELS.md b/new-api/CHANNELS.md
deleted file mode 100644
index 231b1a3..0000000
--- a/new-api/CHANNELS.md
+++ /dev/null
@@ -1,147 +0,0 @@
-# new-api Channel Configuration
-
-After first start, access the new-api web UI at `http://<server>:4000` to configure channels.
-
-Default admin credentials: `root` / `123456` — **change immediately**.
-
-## API Token for Open WebUI
-
-Create an API token in new-api's token management. Use this token as `OPENWEBUI_API_KEY` in `.env`.
-
-## Channels to Create
-
-Configure each channel via **Channels > Add Channel** in the web UI.
-
-### 1. DeepInfra (Priority 1)
-
-| Field | Value |
-|---|---|
-| Name | DeepInfra |
-| Type | OpenAI |
-| Base URL | `https://api.deepinfra.com/v1/openai` |
-| Key | `$DEEPINFRA_API_KEY` |
-| Priority | 1 |
-| Models | See model mapping below |
-
-### 2. SiliconFlow (Priority 2)
-
-| Field | Value |
-|---|---|
-| Name | SiliconFlow |
-| Type | OpenAI |
-| Base URL | `https://api.siliconflow.com/v1` |
-| Key | `$SILICONFLOW_API_KEY` |
-| Priority | 2 |
-| Models | See model mapping below |
-
-### 3. OpenRouter (Priority 3)
-
-| Field | Value |
-|---|---|
-| Name | OpenRouter |
-| Type | OpenAI |
-| Base URL | `https://openrouter.ai/api/v1` |
-| Key | `$OPENROUTER_API_KEY` |
-| Priority | 3 |
-| Models | See model mapping below |
-
-### 4. Groq (Priority 1)
-
-| Field | Value |
-|---|---|
-| Name | Groq |
-| Type | OpenAI |
-| Base URL | `https://api.groq.com/openai/v1` |
-| Key | `$GROQ_API_KEY` |
-| Priority | 1 |
-| Models | `llama-3.3-70b` |
-
-### 5. Cerebras (Priority 1)
-
-| Field | Value |
-|---|---|
-| Name | Cerebras |
-| Type | OpenAI |
-| Base URL | `https://api.cerebras.ai/v1` |
-| Key | `$CEREBRAS_API_KEY` |
-| Priority | 1 |
-| Models | `llama-3.3-70b-cerebras` |
-
-## Model Mapping per Channel
-
-new-api uses model aliasing: the "model name" is what clients see, the "actual model" is what's sent to the provider.
-
-### DeepInfra Models
-
-| Client Model Name | Actual Provider Model |
-|---|---|
-| `deepseek-v3.2` | `deepseek-ai/DeepSeek-V3.2` |
-| `deepseek-r1` | `deepseek-ai/DeepSeek-R1` |
-| `gpt-oss` | `openai/gpt-oss-120b` |
-| `gpt-oss-20b` | `openai/gpt-oss-20b` |
-| `nemotron-super` | `nvidia/Llama-3.3-Nemotron-Super-49B-v1.5` |
-| `nemotron-nano` | `nvidia/NVIDIA-Nemotron-Nano-9B-v2` |
-| `devstral` | `mistralai/Devstral-Small-2505` |
-| `glm-4.6` | `zai-org/GLM-4.6` |
-| `glm-4.7` | `zai-org/GLM-4.7` |
-| `glm-5` | `zai-org/GLM-5` |
-| `kimi-k2` | `moonshotai/Kimi-K2-Instruct-0905` |
-| `kimi-k2.5` | `moonshotai/Kimi-K2.5` |
-| `deepseek-v3-free` | `deepseek-ai/DeepSeek-V3` |
-
-### SiliconFlow Models
-
-| Client Model Name | Actual Provider Model |
-|---|---|
-| `deepseek-v3.2` | `deepseek-ai/DeepSeek-V3.2` |
-| `glm-4.7` | `THUDM/GLM-4-32B-0414` |
-| `kimi-k2` | `moonshotai/Kimi-K2-Instruct-0905` |
-| `qwen3-coder` | `Qwen/Qwen3-Coder-480B-A35B-Instruct` |
-| `qwen3-coder-30b` | `Qwen/Qwen3-Coder-30B-A3B-Instruct` |
-
-### OpenRouter Models
-
-| Client Model Name | Actual Provider Model |
-|---|---|
-| `deepseek-v3.2` | `deepseek/deepseek-chat-v3-0324` |
-| `deepseek-v3-free` | `deepseek/deepseek-chat-v3-0324:free` |
-| `kimi-k2.5` | `moonshotai/kimi-k2.5` |
-| `minimax-m2.5` | `minimax/minimax-m2.5` |
-| `gpt-4.1-mini` | `openai/gpt-4.1-mini` |
-| `gpt-4.1` | `openai/gpt-4.1` |
-| `gemini-3-flash-preview` | `google/gemini-3-flash-preview` |
-| `gemini-2.5-pro` | `google/gemini-2.5-pro-preview` |
-| `claude-sonnet` | `anthropic/claude-sonnet-4` |
-| `trinity-large-preview` | `arcee-ai/trinity-large-preview` |
-
-### Groq Models
-
-| Client Model Name | Actual Provider Model |
-|---|---|
-| `llama-3.3-70b` | `llama-3.3-70b-versatile` |
-
-### Cerebras Models
-
-| Client Model Name | Actual Provider Model |
-|---|---|
-| `llama-3.3-70b-cerebras` | `llama-3.3-70b` |
-
-## Fallback Behavior
-
-new-api handles fallbacks via priority levels:
-- When a model exists on multiple channels, the highest priority (lowest number) channel is tried first
-- If it fails, it automatically falls back to the next priority level
-
-For example, `deepseek-v3.2` exists on:
-1. DeepInfra (priority 1) — tried first
-2. SiliconFlow (priority 2) — fallback
-3. OpenRouter (priority 3) — last resort
-
-## Grafana Setup
-
-After first start, access Grafana at `http://<server>:3001`:
-1. Login with `admin` / `$GRAFANA_ADMIN_PASSWORD`
-2. Add data source: **Prometheus** with URL `http://victoriametrics:8428`
-3. Import dashboards:
-   - Node Exporter Full: dashboard ID `1860`
-   - Redis: dashboard ID `763`
diff --git a/new-api/init-channels.sh b/new-api/init-channels.sh
deleted file mode 100755
index 06363e9..0000000
--- a/new-api/init-channels.sh
+++ /dev/null
@@ -1,218 +0,0 @@
-#!/usr/bin/env bash
-# Configures new-api channels and API token via the admin API.
-# Run once after first boot: ./new-api/init-channels.sh
-#
-# Requires these env vars (or .env file in project root):
-#   NEW_API_PASSWORD      - admin password
-#   DEEPINFRA_API_KEY
-#   SILICONFLOW_API_KEY
-#   OPENROUTER_API_KEY
-#   GROQ_API_KEY
-#   CEREBRAS_API_KEY
-#
-# Optional:
-#   NEW_API_USERNAME      - admin username (default: root)
-#   NEW_API_BASE          - API base URL (default: http://localhost:4000)
-
-set -euo pipefail
-
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-ENV_FILE="${SCRIPT_DIR}/../.env"
-
-# Load .env if present
-if [[ -f "$ENV_FILE" ]]; then
-  set -a
-  # shellcheck disable=SC1090
-  source "$ENV_FILE"
-  set +a
-fi
-
-API_BASE="${NEW_API_BASE:-http://localhost:4000}"
-USERNAME="${NEW_API_USERNAME:-root}"
-PASSWORD="${NEW_API_PASSWORD:?Set NEW_API_PASSWORD to the admin password}"
-COOKIE_JAR=$(mktemp)
-USER_ID=""
-trap 'rm -f "$COOKIE_JAR"' EXIT
-
-# ── Login and get user ID ───────────────────────────────
-login() {
-  echo "Logging in as ${USERNAME}..."
-  local resp
-  resp=$(curl -s -c "$COOKIE_JAR" "${API_BASE}/api/user/login" \
-    -H "Content-Type: application/json" \
-    -d "$(python3 -c "
-import json, sys
-print(json.dumps({'username': sys.argv[1], 'password': sys.argv[2]}))
-" "$USERNAME" "$PASSWORD")")
-
-  local success
-  success=$(echo "$resp" | python3 -c "import sys,json; print(json.load(sys.stdin).get('success', False))")
-  if [[ "$success" != "True" ]]; then
-    echo "ERROR: Login failed: ${resp}"
-    exit 1
-  fi
-
-  USER_ID=$(echo "$resp" | python3 -c "import sys,json; print(json.load(sys.stdin)['data']['id'])")
-  echo "  Logged in (user ID: ${USER_ID})."
-}
-
-# ── API call helper (cookie + New-Api-User header) ─────
-api_call() {
-  local endpoint="$1"
-  shift
-  curl -s -b "$COOKIE_JAR" \
-    -H "New-Api-User: ${USER_ID}" \
-    -H "Content-Type: application/json" \
-    "${API_BASE}${endpoint}" "$@"
-}
-
-# ── Create channel ──────────────────────────────────────
-create_channel() {
-  local name="$1" type="$2" key="$3" base_url="$4" priority="$5" models="$6" model_mapping="$7"
-
-  echo "Creating channel: ${name} (priority ${priority})..."
-
-  local payload
-  payload=$(python3 -c "
-import json, sys
-print(json.dumps({
-    'type': int(sys.argv[1]),
-    'name': sys.argv[2],
-    'key': sys.argv[3],
-    'base_url': sys.argv[4],
-    'models': sys.argv[5],
-    'model_mapping': sys.argv[6],
-    'priority': int(sys.argv[7]),
-    'status': 1,
-    'group': 'default',
-    'weight': 1,
-    'auto_ban': 1
-}))
-" "$type" "$name" "$key" "$base_url" "$models" "$model_mapping" "$priority")
-
-  local resp success
-  resp=$(api_call "/api/channel/" -d "$payload")
-
-  success=$(echo "$resp" | python3 -c "import sys,json; print(json.load(sys.stdin).get('success', False))")
-  if [[ "$success" == "True" ]]; then
-    echo "  OK"
-  else
-    echo "  FAILED: ${resp}" | head -c 500
-    echo
-  fi
-}
-
-# ── Wait for new-api ────────────────────────────────────
-echo "Waiting for new-api at ${API_BASE}..."
-for i in $(seq 1 30); do
-  if curl -sf "${API_BASE}/" > /dev/null 2>&1; then
-    echo "new-api is ready."
-    break
-  fi
-  if [[ "$i" == "30" ]]; then
-    echo "ERROR: new-api did not become ready in time."
-    exit 1
-  fi
-  sleep 2
-done
-
-# ── Login ───────────────────────────────────────────────
-login
-
-# ── Generate system access token for future use ─────────
-echo ""
-echo "Generating system access token..."
-ACCESS_TOKEN_RESP=$(api_call "/api/user/token")
-ACCESS_TOKEN=$(echo "$ACCESS_TOKEN_RESP" | python3 -c "
-import sys, json
-data = json.load(sys.stdin)
-if data.get('success'):
-    print(data.get('data', ''))
-else:
-    print('')
-" 2>/dev/null || echo "")
-
-if [[ -n "$ACCESS_TOKEN" ]]; then
-  echo "  Access token: ${ACCESS_TOKEN}"
-  echo "  Save as NEW_API_ACCESS_TOKEN in .env for future API use."
-  echo "  Usage: -H 'Authorization: Bearer ${ACCESS_TOKEN}' -H 'New-Api-User: ${USER_ID}'"
-else
-  echo "  Could not generate access token (non-critical, using session)."
-fi
-
-# ── Channels ────────────────────────────────────────────
-
-create_channel "DeepInfra" 1 \
-  "${DEEPINFRA_API_KEY:?}" \
-  "https://api.deepinfra.com/v1/openai" \
-  1 \
-  "deepseek-v3.2,deepseek-r1,gpt-oss,gpt-oss-20b,nemotron-super,nemotron-nano,devstral,glm-4.6,glm-4.7,glm-5,kimi-k2,kimi-k2.5" \
-  '{"deepseek-v3.2":"deepseek-ai/DeepSeek-V3.2","deepseek-r1":"deepseek-ai/DeepSeek-R1","gpt-oss":"openai/gpt-oss-120b","gpt-oss-20b":"openai/gpt-oss-20b","nemotron-super":"nvidia/Llama-3.3-Nemotron-Super-49B-v1.5","nemotron-nano":"nvidia/NVIDIA-Nemotron-Nano-9B-v2","devstral":"mistralai/Devstral-Small-2505","glm-4.6":"zai-org/GLM-4.6","glm-4.7":"zai-org/GLM-4.7","glm-5":"zai-org/GLM-5","kimi-k2":"moonshotai/Kimi-K2-Instruct-0905","kimi-k2.5":"moonshotai/Kimi-K2.5"}'
-
-create_channel "SiliconFlow" 1 \
-  "${SILICONFLOW_API_KEY:?}" \
-  "https://api.siliconflow.com/v1" \
-  2 \
-  "deepseek-v3.2,glm-4.7,kimi-k2,qwen3-coder,qwen3-coder-30b" \
-  '{"deepseek-v3.2":"deepseek-ai/DeepSeek-V3.2","glm-4.7":"THUDM/GLM-4-32B-0414","kimi-k2":"moonshotai/Kimi-K2-Instruct-0905","qwen3-coder":"Qwen/Qwen3-Coder-480B-A35B-Instruct","qwen3-coder-30b":"Qwen/Qwen3-Coder-30B-A3B-Instruct"}'
-
-create_channel "OpenRouter" 1 \
-  "${OPENROUTER_API_KEY:?}" \
-  "https://openrouter.ai/api/v1" \
-  3 \
-  "deepseek-v3.2,deepseek-v3-free,kimi-k2.5,minimax-m2.5,gpt-4.1-mini,gpt-4.1,gemini-3-flash-preview,gemini-2.5-pro,claude-sonnet,trinity-large-preview" \
-  '{"deepseek-v3.2":"deepseek/deepseek-chat-v3-0324","deepseek-v3-free":"deepseek/deepseek-chat-v3-0324:free","kimi-k2.5":"moonshotai/kimi-k2.5","minimax-m2.5":"minimax/minimax-m2.5","gpt-4.1-mini":"openai/gpt-4.1-mini","gpt-4.1":"openai/gpt-4.1","gemini-3-flash-preview":"google/gemini-3-flash-preview","gemini-2.5-pro":"google/gemini-2.5-pro-preview","claude-sonnet":"anthropic/claude-sonnet-4","trinity-large-preview":"arcee-ai/trinity-large-preview"}'
-
-create_channel "Groq" 1 \
-  "${GROQ_API_KEY:?}" \
-  "https://api.groq.com/openai/v1" \
-  1 \
-  "llama-3.3-70b" \
-  '{"llama-3.3-70b":"llama-3.3-70b-versatile"}'
-
-create_channel "Cerebras" 1 \
-  "${CEREBRAS_API_KEY:?}" \
-  "https://api.cerebras.ai/v1" \
-  1 \
-  "llama-3.3-70b-cerebras" \
-  '{"llama-3.3-70b-cerebras":"llama-3.3-70b"}'
-
-# ── Create API token for Open WebUI ─────────────────────
-echo ""
-echo "Creating API token for Open WebUI..."
-TOKEN_RESP=$(api_call "/api/token/" -d "$(python3 -c "
-import json
-print(json.dumps({
-    'name': 'open-webui',
-    'remain_quota': 0,
-    'unlimited_quota': True
-}))
-")")
-
-TOKEN_KEY=$(echo "$TOKEN_RESP" | python3 -c "
-import sys, json
-data = json.load(sys.stdin)
-if data.get('success'):
-    print(data['data']['key'])
-else:
-    print('FAILED: ' + data.get('message', 'unknown error'))
-" 2>/dev/null || echo "FAILED: could not parse response")
-
-echo ""
-echo "══════════════════════════════════════"
-echo "Setup complete!"
-echo ""
-if [[ "$TOKEN_KEY" != FAILED* ]]; then
-  echo "Open WebUI API key: ${TOKEN_KEY}"
-  echo "Set OPENWEBUI_API_KEY=${TOKEN_KEY} in your .env"
-  echo ""
-  echo "Test:"
-  echo "  curl ${API_BASE}/v1/chat/completions \\"
-  echo "    -H 'Authorization: Bearer ${TOKEN_KEY}' \\"
-  echo "    -H 'Content-Type: application/json' \\"
-  echo "    -d '{\"model\":\"deepseek-v3.2\",\"messages\":[{\"role\":\"user\",\"content\":\"hi\"}]}'"
-else
-  echo "Token creation: ${TOKEN_KEY}"
-  echo "Create a token manually in the new-api UI."
-fi
-echo "══════════════════════════════════════"