记录一下目前在用的llm自建应用方案

stevessr

https://zeabur.com/templates/77FSH6
searXNG (feat: lobehub 但我用的是网上公开的)
https://github.com/open-webui/open-webui
open-webui (之前的ollama webui)
https://openwebui.com/f/coker/gemini
owu 插件函数，用于实现Gemini完全体
https://github.com/Calcium-Ion/new-api
new api 用于中转/合并/重定向/负载均衡
还是反向代理爽，直接加速窜稀
# /etc/nginx/nginx.conf or /usr/local/openresty/nginx/conf/nginx.conf

#----------------------------------------------------------------------
# Global Settings
#----------------------------------------------------------------------

# User and group Nginx runs as (often nginx or www-data)
# user nginx nginx;

# Automatically set worker processes based on CPU cores
worker_processes auto;

# Automatically bind worker processes to specific CPU cores
worker_cpu_affinity auto;

# Path to the PID file
pid /www/server/nginx/nginx.pid; # Make sure this path is writable by the Nginx user

# Error log configuration
# Levels: debug, info, notice, warn, error, crit, alert, emerg
error_log logs/error.log notice;

#----------------------------------------------------------------------
# Events Module Settings
#----------------------------------------------------------------------
events {
    # Max connections per worker process
    worker_connections 1024;

    # Use epoll for high performance I/O multiplexing on Linux
    use epoll;

    # Accept multiple connections at once if supported by the OS
    # multi_accept on;
}

#----------------------------------------------------------------------
# HTTP Block Settings
#----------------------------------------------------------------------
http {
    #------------------------------------------------------------------
    # Basic HTTP Settings
    #------------------------------------------------------------------
    include       mime.types;          # Include MIME type mappings
    default_type  application/octet-stream; # Default MIME type for unknown extensions

    #------------------------------------------------------------------
    # Logging Settings
    #------------------------------------------------------------------
    log_format main '[imath:0]remote_addr - [/imath:0]remote_user [[imath:0]time_local] "[/imath:0]request" '
                      '[imath:0]status [/imath:0]body_bytes_sent "[imath:0]http_referer" '
                      '"[/imath:0]http_user_agent" "[imath:0]http_x_forwarded_for"';

    access_log logs/access.log main; # Path to the access log

    #------------------------------------------------------------------
    # Performance Tuning
    #------------------------------------------------------------------
    sendfile        on;   # Enable efficient file transfer using sendfile()
    # tcp_nopush      on;   # Send HTTP response headers in one packet (useful with sendfile on)
    # tcp_nodelay     on;   # Send data immediately without buffering (good for low latency)

    keepalive_timeout  65;  # Timeout for keep-alive connections
    # keepalive_requests 100; # Max requests per keep-alive connection

    # client_max_body_size 100m; # Max allowed size of client request body

    #------------------------------------------------------------------
    # Gzip Compression Settings
    #------------------------------------------------------------------
    gzip              on;             # Enable gzip compression
    gzip_disable      "msie6";        # Disable for IE6 (buggy)
    gzip_vary         on;             # Add "Vary: Accept-Encoding" header
    gzip_proxied      any;            # Compress responses from proxied servers
    gzip_comp_level   6;              # Compression level (1-9)
    gzip_buffers      16 8k;          # Number and size of compression buffers
    gzip_http_version 1.1;            # Minimum HTTP version to compress
    gzip_min_length   1000;           # Minimum response size to compress
    gzip_types        text/plain text/css application/json application/javascript application/xml text/xml image/svg+xml; # MIME types to compress

    #------------------------------------------------------------------
    # Proxy Cache Settings
    #------------------------------------------------------------------
    # Defines a cache storage area on disk
    # levels: Directory structure levels (e.g., 1:2 means /cache/c/c4/...)
    # keys_zone: Name and size of the shared memory zone for cache keys
    # max_size: Maximum size of the cache on disk
    # inactive: Files not accessed for this duration are removed
    proxy_cache_path  cache levels=1:2 keys_zone=cache_zone:64m max_size=1000m inactive=10m use_temp_path=off;

    # Defines the default cache key format
    proxy_cache_key   "[/imath:0]scheme[imath:0]request_method[/imath:0]host[imath:0]request_uri[/imath:0]is_args$args";

    #------------------------------------------------------------------
    # Server Block (Virtual Host)
    #------------------------------------------------------------------
    server {
        listen 80 default_server; # Listen on port 80 for IPv4
        # listen [::]:80 default_server; # Listen on port 80 for IPv6 (optional)
        server_name localhost your_domain.com; # Replace with your actual domain or IP

        # Default root directory for requests
        root /mnt/disk/development/frontend;
        # Default file to serve if directory is requested
        index index.html index.htm;

        #--------------------------------------------------------------
        # Location Blocks (Request Routing)
        #--------------------------------------------------------------

        # Serve static files from /op directory alias
        location /op {
            alias /mnt/disk/op/;      # Serve files directly from this path
            autoindex on;             # Optional: Enable directory listing if index file not found
            sendfile on;              # Use sendfile for these static assets
            tcp_nopush on;
            expires 30d;              # Set browser cache expiration
            add_header Cache-Control "public"; # Mark as publicly cacheable
        }

        # Serve static files from /admin directory alias (seems similar to root, maybe redundant?)
        # If /admin is part of your frontend SPA, the location block below might handle it better.
        # Consider if this specific alias is still needed. If not, remove it.
        location /admin {
            alias /mnt/disk/development/frontend/; # Alias to the same root? Check if this is correct.
            autoindex on;             # Optional: Enable directory listing
            sendfile on;
            tcp_nopush on;
            expires 30d;              # Set browser cache expiration
            add_header Cache-Control "public";
            # If /admin is part of the SPA, you might need try_files here too:
            # try_files $uri $uri/ /index.html;
        }

        # Proxy API requests under /api/v1 WITHOUT caching
        location /api/v1 {
            proxy_pass http://127.0.0.1:8088; # Proxy to backend service

            # Common Proxy Headers
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header X-Forwarded-Proto $scheme;
            proxy_set_header REMOTE-HOST $remote_addr; # Custom header? Ensure backend uses it.

            # Proxy Settings
            proxy_http_version 1.1;
            proxy_read_timeout 3600s; # Long timeout (1 hour), adjust if needed
            proxy_send_timeout 3600s; # Long timeout (1 hour), adjust if needed

            # Cache Control for this specific location (Bypass/No Cache)
            proxy_cache_bypass $http_upgrade; # Don't use cache for WebSocket upgrades
            proxy_no_cache $http_upgrade;     # Don't store cache for WebSocket upgrades
            # Explicitly disable caching for this path if needed beyond bypass/no_cache
            # proxy_cache off;

            # Add cache status header (will likely show BYPASS or MISS)
            add_header X-Cache $upstream_cache_status;

            # Security Headers (Consider placing globally if applicable to all proxied requests)
            add_header Strict-Transport-Security "max-age=31536000" always; # Use 'always' to ensure it's added

            # WebSocket Headers (Needed only if this endpoint handles WebSockets)
            # proxy_set_header Upgrade $http_upgrade;
            # proxy_set_header Connection "upgrade";
        }

        # Proxy API requests under /api WITH caching
        location /api {
            proxy_pass http://127.0.0.1:8088; # Proxy to backend service

            # Common Proxy Headers
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header X-Forwarded-Proto $scheme;
            proxy_set_header REMOTE-HOST $remote_addr;

            # Proxy Settings
            proxy_http_version 1.1;
            proxy_read_timeout 3600s; # Long timeout
            proxy_send_timeout 3600s; # Long timeout

            # Cache Settings for this location
            proxy_cache cache_zone;           # Use the defined cache zone
            proxy_cache_key $host$uri$is_args$args; # Use a specific cache key (or inherit global)
            proxy_cache_valid 200 304 1m;     # Cache 200/304 responses for 1 minute
            proxy_cache_valid 301 302 10m;    # Cache redirects for 10 minutes
            proxy_cache_valid any 1m;         # Cache other status codes (e.g., 404) for 1 minute (optional)
            proxy_cache_use_stale error timeout updating http_500 http_502 http_503 http_504; # Serve stale cache on errors
            proxy_cache_lock on;              # Prevent multiple requests for the same uncached item simultaneously

            # Cache Bypass
            proxy_cache_bypass $http_upgrade; # Bypass for WebSocket upgrades
            proxy_no_cache $http_upgrade;     # Don't store cache for WebSocket upgrades

            # Headers
            add_header X-Cache $upstream_cache_status; # Add cache status header
            add_header Strict-Transport-Security "max-age=31536000" always;
            proxy_ignore_headers Set-Cookie Cache-Control expires; # Ignore backend cache control headers

            # WebSocket Headers (Needed only if this endpoint handles WebSockets)
            # proxy_set_header Upgrade $http_upgrade;
            # proxy_set_header Connection "upgrade";
        }

        # Proxy WebSocket requests under /ws
        location /ws {
            proxy_pass http://127.0.0.1:8088; # Proxy to backend WebSocket service

            # Proxy Settings for WebSocket
            proxy_http_version 1.1;
            proxy_set_header Upgrade $http_upgrade; # Essential for WebSocket handshake
            proxy_set_header Connection "upgrade";   # Essential for WebSocket handshake
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header X-Forwarded-Proto $scheme;
            proxy_read_timeout 86400s; # Very long timeout for persistent connections (24 hours)
            proxy_send_timeout 86400s; # Very long timeout for persistent connections

            # Caching is generally NOT desired for WebSockets. Removed cache directives.
            # add_header X-Cache $upstream_cache_status; # This header is less relevant here
            add_header Strict-Transport-Security "max-age=31536000" always;
        }

        # Proxy requests under /ollama (Assuming potentially long-running requests, maybe SSE or similar)
        location /ollama {
            proxy_pass http://127.0.0.1:8088; # Proxy to Ollama backend service

            # Common Proxy Headers
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header X-Forwarded-Proto $scheme;

            # Proxy Settings
            proxy_http_version 1.1;
            proxy_read_timeout 3600s; # Long timeout, adjust as needed for Ollama responses
            proxy_send_timeout 3600s; # Long timeout
            proxy_buffering off;      # Disable buffering for streaming responses (like SSE)

            # WebSocket Headers (Include if /ollama uses WebSockets)
            # proxy_set_header Upgrade $http_upgrade;
            # proxy_set_header Connection "upgrade";

            # Caching is likely NOT desired for dynamic AI responses. Removed cache directives.
            # add_header X-Cache $upstream_cache_status;
            add_header Strict-Transport-Security "max-age=31536000" always;
        }

        # Handle Single Page Application (SPA) routing for specific paths
        # This ensures that direct navigation to /c/some/path, /channels/etc, /workspace/view, /admin/panel
        # serves the main index.html, letting the frontend router take over.
        location ~ ^/(c|channels|workspace|admin)/ {
            # Serve static assets if they exist directly (e.g., /admin/logo.png)
            # Otherwise, fall back to serving /index.html for the SPA router
            try_files $uri $uri/ /index.html;

            # No proxying here unless specific API calls under these paths are intended
            # and not covered by /api or /api/v1. The original 'if' logic based on
            # Accept header was complex and potentially problematic.
            # If you need to proxy specific non-HTML requests under these paths,
            # add more specific location blocks *before* this one. Example:
            # location /admin/api/users { proxy_pass ...; }
        }

        # Catch-all for other requests - handles root SPA routing and static files
        # This block should usually come last or after more specific locations.
        location / {
            try_files $uri $uri/ /index.html; # Serve file, directory, or fallback to index.html for SPA
            expires -1; # Prevent caching of index.html itself (usually desired for SPAs)
            add_header Cache-Control "no-store"; # Stronger cache prevention
        }

        # Example Lua block (OpenResty specific)
        location = /hello { # Use exact match '=' for performance
            default_type text/html;
            content_by_lua_block {
                ngx.say("<h1>Hello, OpenResty!</h1>")
            }
        }

        # Optional: Deny access to hidden files (e.g., .htaccess, .git)
        location ~ /\. {
            deny all;
        }

        # Optional: Favicon and robots.txt handling
        location = /favicon.ico {
            log_not_found off;
            access_log off;
        }
        location = /robots.txt {
            log_not_found off;
            access_log off;
        }
    }

    #------------------------------------------------------------------
    # Include other configuration files (optional)
    #------------------------------------------------------------------
    # include /etc/nginx/conf.d/*.conf;
    # include /etc/nginx/sites-enabled/*;
}
部分自定义配置
https://linux.do/t/topic/506938
还有在论坛里面帮别人解决问题时用deno写的反向代理
https://github.com/stevessr/small-things/blob/main/reverse_http.ts
可部署在
https://dash.deno.com/
支持http&https、ws&wss，用于隐藏原服务器