@hitechclaw/clawspark 2.0.0

package/lib/setup-messaging.sh

@@ -0,0 +1,242 @@
+#!/usr/bin/env bash
+# lib/setup-messaging.sh — Configures WhatsApp and/or Telegram messaging channels.
+# Reads FLAG_MESSAGING (set by install.sh or interactive prompt).
+set -euo pipefail
+
+setup_messaging() {
+    log_info "Setting up messaging channels..."
+    hr
+
+    local messaging_choice="${FLAG_MESSAGING:-}"
+
+    # ── Prompt if not set via flag ──────────────────────────────────────────
+    if [[ -z "${messaging_choice}" ]]; then
+        local options=("WhatsApp" "Telegram" "Both" "Skip")
+        messaging_choice=$(prompt_choice "Which messaging channel(s) would you like?" options 0)
+    fi
+
+    messaging_choice=$(to_lower "${messaging_choice}")
+    export MESSAGING_CHOICE="${messaging_choice}"
+
+    case "${messaging_choice}" in
+        whatsapp)
+            _setup_whatsapp
+            ;;
+        telegram)
+            _setup_telegram
+            ;;
+        both)
+            _setup_whatsapp
+            _setup_telegram
+            ;;
+        skip)
+            log_info "Messaging setup skipped."
+            ;;
+        *)
+            log_warn "Unknown messaging option '${messaging_choice}' -- skipping."
+            ;;
+    esac
+
+    # Always start the gateway (Chat UI needs it even without messaging)
+    _start_gateway
+}
+
+# ── WhatsApp ────────────────────────────────────────────────────────────────
+
+_setup_whatsapp() {
+    log_info "Configuring WhatsApp..."
+    printf '\n'
+    printf '  %s%sWhatsApp Setup%s\n' "${BOLD}" "${GREEN}" "${RESET}"
+    printf '  OpenClaw uses the Baileys library for WhatsApp Web linking.\n'
+    printf '  You will need to scan a QR code from your phone.\n'
+    printf '  Open WhatsApp on your phone > Linked Devices > Link a Device\n\n'
+    printf '  %s%sNote:%s WhatsApp linking depends on the upstream Baileys library.\n' "${BOLD}" "${YELLOW}" "${RESET}"
+    printf '  If linking fails, the Chat UI still works at:\n'
+    printf '    http://localhost:18789/__openclaw__/canvas/\n'
+    printf '  You can also use Telegram as an alternative messaging channel.\n\n'
+
+    if [[ "${CLAWSPARK_DEFAULTS}" == "true" ]]; then
+        log_warn "Cannot complete WhatsApp QR scan in --defaults mode."
+        log_info "Run 'openclaw channels login' later to link WhatsApp."
+        return 0
+    fi
+
+    # Ask for the user's WhatsApp phone number (for allowlist security)
+    printf '  Enter your WhatsApp phone number (with country code, e.g. 919878251239):\n'
+    printf '  > '
+    local wa_number=""
+    read -r wa_number </dev/tty || true
+    wa_number="${wa_number//[^0-9]/}"  # strip non-digits
+
+    # Enable WhatsApp channel in config with security + message batching
+    openclaw config set channels.whatsapp.enabled true >> "${CLAWSPARK_LOG}" 2>&1 || true
+    openclaw config set channels.whatsapp.debounceMs 8000 >> "${CLAWSPARK_LOG}" 2>&1 || true
+    openclaw config set channels.whatsapp.groupPolicy disabled >> "${CLAWSPARK_LOG}" 2>&1 || true
+
+    # Set allowlist so only the owner can interact (no auto-replies to strangers)
+    if [[ -n "${wa_number}" ]]; then
+        openclaw config set channels.whatsapp.allowFrom "[\"${wa_number}@s.whatsapp.net\"]" >> "${CLAWSPARK_LOG}" 2>&1 || true
+        openclaw config set channels.whatsapp.dmPolicy allowlist >> "${CLAWSPARK_LOG}" 2>&1 || true
+        log_success "WhatsApp locked to your number: +${wa_number}"
+    else
+        log_warn "No phone number provided. Using pairing mode (strangers will get auto-replies)."
+        openclaw config set channels.whatsapp.dmPolicy pairing >> "${CLAWSPARK_LOG}" 2>&1 || true
+    fi
+
+    printf '\n'
+    printf '  Now scan the QR code with your phone:\n'
+    printf '  Open WhatsApp > Linked Devices > Link a Device\n\n'
+    printf '  Press Enter when ready...'
+    read -r </dev/tty || true
+
+    # Source Ollama env so the gateway can start if needed
+    local env_file="${HOME}/.openclaw/gateway.env"
+    if [[ -f "${env_file}" ]]; then set +e; set -a; source "${env_file}" 2>/dev/null; set +a; set -e; fi
+
+    # Use the channels login command for WhatsApp
+    openclaw channels login --channel whatsapp --verbose 2>&1 | tee -a "${CLAWSPARK_LOG}" || {
+        log_warn "WhatsApp linking exited with an error."
+    }
+
+    printf '\n'
+    if prompt_yn "  Did you successfully link WhatsApp?" "y"; then
+        log_success "WhatsApp linked."
+    else
+        log_warn "WhatsApp not linked. You can try again later: openclaw channels login"
+        log_info "The Chat UI is always available at: http://localhost:18789/__openclaw__/canvas/"
+    fi
+}
+
+# ── Telegram ────────────────────────────────────────────────────────────────
+
+_setup_telegram() {
+    log_info "Configuring Telegram..."
+    printf '\n'
+    printf '  %s%sTelegram Setup%s\n' "${BOLD}" "${GREEN}" "${RESET}"
+    printf '  The OpenClaw wizard will guide you through Telegram bot setup.\n'
+    printf '  You will need a bot token from @BotFather on Telegram.\n\n'
+
+    if [[ "${CLAWSPARK_DEFAULTS}" == "true" ]]; then
+        log_warn "Cannot configure Telegram bot in --defaults mode."
+        log_info "Run 'openclaw configure --section channels' later."
+        return 0
+    fi
+
+    # Source Ollama env
+    local env_file="${HOME}/.openclaw/gateway.env"
+    if [[ -f "${env_file}" ]]; then set +e; set -a; source "${env_file}" 2>/dev/null; set +a; set -e; fi
+
+    openclaw configure --section channels 2>&1 | tee -a "${CLAWSPARK_LOG}" || {
+        log_warn "Channel setup exited with an error. You can configure Telegram manually later."
+        return 0
+    }
+    log_success "Telegram bot configured."
+}
+
+# ── Gateway ─────────────────────────────────────────────────────────────────
+
+_start_gateway() {
+    log_info "Starting OpenClaw gateway..."
+    local gateway_log="${CLAWSPARK_DIR}/gateway.log"
+    local gateway_pid_file="${CLAWSPARK_DIR}/gateway.pid"
+
+    # Kill existing gateway if running
+    if [[ -f "${gateway_pid_file}" ]]; then
+        local old_pid
+        old_pid=$(cat "${gateway_pid_file}")
+        if kill -0 "${old_pid}" 2>/dev/null; then
+            log_info "Stopping existing gateway (PID ${old_pid})..."
+            kill "${old_pid}" 2>/dev/null || true
+            sleep 1
+        fi
+    fi
+
+    # Source Ollama provider credentials
+    local env_file="${HOME}/.openclaw/gateway.env"
+    if [[ -f "${env_file}" ]]; then set +e; set -a; source "${env_file}" 2>/dev/null; set +a; set -e; fi
+
+    nohup openclaw gateway run --bind loopback > "${gateway_log}" 2>&1 &
+    local gw_pid=$!
+    echo "${gw_pid}" > "${gateway_pid_file}"
+
+    sleep 3
+    if kill -0 "${gw_pid}" 2>/dev/null; then
+        log_success "Gateway running (PID ${gw_pid}). Logs: ${gateway_log}"
+    else
+        log_warn "Gateway process exited unexpectedly. Check ${gateway_log}."
+        log_warn "You can start it manually later: clawspark restart"
+        return 0
+    fi
+
+    # Node host will be started later (after all config changes are done)
+    # to avoid gateway restart (1012) killing it mid-setup.
+}
+
+_start_node_host() {
+    log_info "Starting node host (agent execution engine)..."
+    local node_log="${CLAWSPARK_DIR}/node.log"
+    local node_pid_file="${CLAWSPARK_DIR}/node.pid"
+
+    # Ensure workspace directory exists and is writable
+    local workspace="${HOME}/workspace"
+    mkdir -p "${workspace}"
+    openclaw config set agents.defaults.workspace "${workspace}" >> "${CLAWSPARK_LOG}" 2>&1 || true
+
+    # Kill existing node host if running
+    if [[ -f "${node_pid_file}" ]]; then
+        local old_pid
+        old_pid=$(cat "${node_pid_file}")
+        kill "${old_pid}" 2>/dev/null || true
+        sleep 1
+    fi
+
+    local env_file="${HOME}/.openclaw/gateway.env"
+    if [[ -f "${env_file}" ]]; then set +e; set -a; source "${env_file}" 2>/dev/null; set +a; set -e; fi
+
+    nohup openclaw node run --host 127.0.0.1 --port 18789 > "${node_log}" 2>&1 &
+    local node_pid=$!
+    echo "${node_pid}" > "${node_pid_file}"
+
+    # Wait for the node to register and create a pairing request
+    sleep 5
+
+    # Auto-approve pending device requests (this is a local single-user install)
+    local attempt
+    for attempt in 1 2 3; do
+        local device_output
+        device_output=$(openclaw devices list 2>/dev/null || echo "")
+
+        # Only extract UUIDs from lines that contain "pending" or "Pending"
+        # to avoid approving already-paired devices or matching non-request UUIDs
+        local request_ids
+        request_ids=$(echo "${device_output}" | grep -iE 'pending|request' | grep -oE '[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}' || true)
+        # Fallback: if no "pending" lines but there are UUIDs in a "Requests" section, try those
+        if [[ -z "${request_ids}" ]]; then
+            request_ids=$(echo "${device_output}" | sed -n '/^Request/,/^$/p' | grep -oE '[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}' || true)
+        fi
+
+        if [[ -n "${request_ids}" ]]; then
+            local rid
+            for rid in ${request_ids}; do
+                openclaw devices approve "${rid}" >> "${CLAWSPARK_LOG}" 2>&1 && \
+                    log_info "Auto-approved device: ${rid}" || true
+            done
+            # Restart node host after approval so it reconnects with proper auth
+            kill "${node_pid}" 2>/dev/null || true
+            sleep 2
+            nohup openclaw node run --host 127.0.0.1 --port 18789 > "${node_log}" 2>&1 &
+            node_pid=$!
+            echo "${node_pid}" > "${node_pid_file}"
+            sleep 3
+            break
+        fi
+        sleep 2
+    done
+
+    if kill -0 "${node_pid}" 2>/dev/null; then
+        log_success "Node host running (PID ${node_pid})."
+    else
+        log_warn "Node host failed to start. Agent will work without exec/browser tools."
+        log_warn "You can start it manually: openclaw node run --host 127.0.0.1 --port 18789"
+    fi
+}

package/lib/setup-models.sh

@@ -0,0 +1,121 @@
+#!/usr/bin/env bash
+# lib/setup-models.sh -- Multi-model configuration for OpenClaw.
+# Configures vision and image generation model slots alongside the
+# primary chat model that was already selected during install.
+set -euo pipefail
+
+setup_models() {
+    log_info "Configuring multi-model support..."
+
+    # The primary model is already set during install (SELECTED_MODEL_ID).
+    # Now configure additional model slots: vision and image generation.
+
+    # ── Vision model ──────────────────────────────────────────────────────
+    # Check for vision-capable models already pulled in Ollama
+    local vision_models=("qwen2.5vl" "qwen2.5-vl" "llava" "minicpm-v" "llama3.2-vision" "moondream")
+    local found_vision=""
+
+    for vm in "${vision_models[@]}"; do
+        if ollama list 2>/dev/null | grep -qi "${vm}"; then
+            found_vision="${vm}"
+            break
+        fi
+    done
+
+    if [[ -n "${found_vision}" ]]; then
+        # Get the full model tag from ollama list output
+        local full_tag
+        full_tag=$(ollama list 2>/dev/null | grep -i "${found_vision}" | head -1 | awk '{print $1}')
+        log_success "Found vision model: ${full_tag}"
+        openclaw config set agents.defaults.imageModel "ollama/${full_tag}" >> "${CLAWSPARK_LOG}" 2>&1 || true
+    else
+        # No vision model found -- decide whether to auto-pull based on VRAM
+        local vram_gb=$(( ${HW_GPU_VRAM_MB:-0} / 1024 ))
+        local skip_vision=false
+
+        # RTX with <=12GB VRAM: pulling a 5GB vision model alongside the chat
+        # model will cause OOM when both are loaded simultaneously
+        if [[ "${HW_PLATFORM:-}" == "rtx" ]] && (( vram_gb > 0 && vram_gb <= 12 )); then
+            skip_vision=true
+            log_info "Skipping vision model auto-pull (${vram_gb}GB VRAM -- dual-model loading would OOM)."
+            log_info "Add one later when needed: ollama pull moondream (1.5GB, fits alongside chat model)"
+        fi
+
+        if [[ "${skip_vision}" != "true" ]]; then
+            local vision_choice="qwen2.5vl:7b"
+            log_info "No vision model found. Pulling ${vision_choice} for image analysis (~5GB)..."
+            (ollama pull "${vision_choice}") >> "${CLAWSPARK_LOG}" 2>&1 &
+            spinner $! "Pulling ${vision_choice}..."
+            # Wait a moment for Ollama to index the model, then check with retries
+            local _vision_ok=false
+            local _retry
+            for _retry in 1 2 3; do
+                if ollama list 2>/dev/null | grep -qi "qwen2.5vl"; then
+                    _vision_ok=true
+                    break
+                fi
+                sleep 2
+            done
+            if [[ "${_vision_ok}" == "true" ]]; then
+                found_vision="qwen2.5vl"
+                openclaw config set agents.defaults.imageModel "ollama/${vision_choice}" >> "${CLAWSPARK_LOG}" 2>&1 || true
+                log_success "Vision model configured: ollama/${vision_choice}"
+            else
+                log_warn "Vision model pull failed. You can add one later: ollama pull qwen2.5vl:7b"
+            fi
+        fi
+    fi
+
+    # ── Register vision model with explicit input modality ─────────────────
+    # OpenClaw discovers Ollama models via /v1/models which doesn't report
+    # vision capability. We must explicitly register the model with
+    # input: ["text", "image"] so the image tool works.
+    local config_file="${HOME}/.openclaw/openclaw.json"
+    if [[ -n "${found_vision}" && -f "${config_file}" ]]; then
+        local vision_model_id="${found_vision}"
+        # Get the full tag
+        local vision_full_tag
+        vision_full_tag=$(ollama list 2>/dev/null | grep -i "${vision_model_id}" | head -1 | awk '{print $1}')
+        if [[ -n "${vision_full_tag}" ]]; then
+            python3 -c "
+import json, sys
+path = sys.argv[1]
+model_id = sys.argv[2]
+with open(path) as f:
+    cfg = json.load(f)
+cfg.setdefault('models', {}).setdefault('providers', {}).setdefault('ollama', {})
+cfg['models']['providers']['ollama'].setdefault('baseUrl', 'http://127.0.0.1:11434/v1')
+# Use openai-completions api -- Ollama's /v1 endpoint is OpenAI-compatible
+# and the built-in openai-completions provider is always registered,
+# unlike the dynamic 'ollama' api which is missing from the vision code path.
+cfg['models']['providers']['ollama']['api'] = 'openai-completions'
+cfg['models']['providers']['ollama']['models'] = [
+    {
+        'id': model_id,
+        'name': model_id.replace(':', ' ').title(),
+        'api': 'openai-completions',
+        'input': ['text', 'image'],
+        'contextWindow': 32768,
+        'maxTokens': 8192
+    }
+]
+with open(path, 'w') as f:
+    json.dump(cfg, f, indent=2)
+print('ok')
+" "${config_file}" "${vision_full_tag}" 2>> "${CLAWSPARK_LOG}" || {
+                log_warn "Could not register vision model modality."
+            }
+            log_success "Vision model registered with image input capability."
+        fi
+    fi
+
+    # ── Image generation model ────────────────────────────────────────────
+    # Image generation (text-to-image) is optional and more complex.
+    # It typically requires ComfyUI, Stable Diffusion, or an external API.
+    # For now, log a message about how to enable it later.
+    log_info "Image generation: not configured (optional)."
+    log_info "  To enable later, set up ComfyUI or a text-to-image API and run:"
+    log_info "  openclaw config set agents.defaults.imageGenerationModel <provider>/<model>"
+
+    log_success "Model configuration complete."
+}