Loading knowledge…
Fetching datasets and sources.
Fetching datasets and sources.
Source-linked operating notes for running your own model server, Next.js app, Postgres, and Cloudflare edge. Scoped to Next.js 16 and current vLLM; not a general knowledge corpus.
Agents can query it without an account. POST to /api/v1/retrieve, or connect an MCP client and call koskamo_retrieve.
curl --fail-with-body https://koskamo.com/api/v1/retrieve -H 'Content-Type: application/json' -d '{"query":"self-hosted-ai-reference","dataset":"self-hosted-ai-reference","top_k":3}'11 of 11 shown
after() from next/server schedules work to run once a response or prerender has finished, which suits logging and analytics that must not delay the caller. It works in Server…
Behind Cloudflare, the socket peer your origin sees is a Cloudflare edge address. CF-Connecting-IP carries the IP address of the client connecting to Cloudflare. Cloudflare…
Cloudflare returns HTTP 524 when it connected to the origin but the origin did not provide an HTTP response within the default Proxy Read Timeout of 125 seconds. Only Enterprise…
A mixture-of-experts (MoE) model must hold all of its parameters in memory, but each token only runs through some experts. Hugging Face's example is Mixtral 8x7B: it needs enough…
@neondatabase/serverless is configured to connect to Neon. Its README says it can reach your own Postgres only if you run your own WebSocket proxy. Koskamo field note: when moving…
In Next.js 16 the middleware file convention is deprecated and renamed proxy. Without a matcher, Proxy runs on every request, including _next/static, _next/image, and public…
When self-hosting Next.js with next start, put a reverse proxy such as nginx in front of the Node server instead of exposing it directly. The proxy handles malformed requests,…
On Linux, sun_path in a Unix domain socket address is 108 bytes and the pathname should be null-terminated, which leaves 107 usable bytes. PostgreSQL names its socket…
Next.js streams responses when self-hosted, but a buffering proxy holds the stream until it completes. For nginx, disable buffering by sending the response header…
When an app leaves a platform with built-in cron, a systemd .timer paired with a .service can call the same endpoint. OnCalendar= defines wall-clock schedules using calendar event…
vllm serve <model> starts an HTTP server that implements OpenAI-style APIs, including /v1/chat/completions (which needs a chat template), /v1/completions, /v1/embeddings for…