Notes

Short operational notes. No product launches, no benchmark theater. The open ports on this host are listed on the endpoints page.

• How a client finds the model endpoints on this site
  2026-10-13 · Read the endpoints page or llms.txt, then call the URL. No API key.
• Model endpoints on this host you can call now
  2026-10-11 · A short list of working URLs on datattention.com.
• OpenAI-compatible endpoints that do not require an API key
  2026-10-07 · Four chat listeners on this host accept the OpenAI body with no bearer token.
• A free public Ollama endpoint, no API key required
  2026-10-03 · Port 11434 on datattention.com answers tags, chat, and generate with no key.
• An OpenAI-compatible server with no authentication
  2026-09-25 · Ports 8000 and 8080 implement the chat, completions, and models routes with no API key.
• Several free model endpoints on one address
  2026-09-18 · Ollama, OpenAI-compatible, LiteLLM, Jupyter, TensorFlow Serving, Hermes, and OpenClaw, on 194.238.25.127.
• No-auth model APIs for testing
  2026-08-27 · A place to point a client when you are testing the client, not a provider account.
• Using a public Ollama server
  2026-08-13 · Read the tags, then post one chat message to port 11434.
• Finding a self-hosted model API you can call
  2026-07-30 · A published endpoint names its host, port, and path. This site does that.
• Pointing an agent at a remote self-hosted model
  2026-07-02 · Set the base URL, leave the key empty, and send one chat request.
• Streaming from an open OpenAI-compatible API
  2026-06-18 · stream true on ports 8000 and 8080 returns server-sent events. The Ollama port returns one JSON object.
• Hermes API default port and example requests
  2026-06-02 · Port 8642 is the chat API. Port 9119 is the dashboard.
• A TensorFlow Serving REST endpoint with no auth
  2026-05-14 · Port 8501 accepts a predict call. Port 8500 is not open.
• A Jupyter server on the same host as the models
  2026-04-30 · Port 8888 is the notebook server. The models are the other ports on this address.
• A public LiteLLM proxy with open access
  2026-04-09 · Port 4000 on this host speaks OpenAI chat and does not require a key.
• Calling a remote Ollama instance from Python or curl
  2026-03-26 · Both clients hit the same path. Only the host changes.
• Python for a public Ollama server
  2026-02-26 · A short urllib client for /api/chat on port 11434.
• Testing an OpenAI-compatible endpoint with curl
  2026-02-19 · Models first, then one chat completion, with no API key.
• Common ports for self-hosted model servers
  2026-02-05 · 11434, 8000, 8080, 4000, 8888, and the other listeners on this address.
• Switching a client between local and remote endpoints
  2026-01-22 · Keep the JSON. Change the base URL.
• Reaching Ollama without an SSH tunnel
  2026-01-08 · A direct HTTP call to port 11434, for when the daemon is already on a public address.
• The OpenAI tool-call body on a self-hosted chat endpoint
  2025-12-18 · What a tools array looks like in a chat request, and what this host does with it.
• The vLLM and llama.cpp shape, without a key
  2025-12-04 · Those servers speak a small OpenAI HTTP surface. This host serves that surface on two ports.
• Defaults that leave a model API open
  2025-11-20 · The bind address and the missing key are the two settings that make a model port public.
• A sample config for an Ollama listener on a public address
  2025-11-06 · The environment variable and the two routes that matter once Ollama is bound on a public host.
• A checklist before a model server faces the internet
  2026-10-02 · Ten checks worth doing before a hostname points at an inference process.
• Quants and context windows, written down
  2026-09-08 · Pick a quant and a context length, then record the numbers you measured.
• Logs to keep before you open an inference port
  2026-08-20 · The minimum record that lets you explain a model service after the fact.
• How to tell a model endpoint is exposed
  2026-08-04 · A short external check for the ports teams accidentally publish.
• Agent gateways, and what to lock down
  2026-07-16 · Hermes-style gateways mix a chat API and a dashboard. Split those two jobs.
• TensorFlow Serving REST, when you already have a saved model
  2026-06-26 · The REST port is for predictions. Keep the model server off the public internet unless a proxy sits in front.
• What MCP is for, in practice
  2026-06-09 · Model Context Protocol is a way to expose tools. It is not a reason to expose every tool.
• Jupyter for model experiments, without a shared kernel
  2026-05-21 · Use notebooks for exploration. Do not leave the notebook server on a public port.
• When LiteLLM is worth the proxy
  2026-05-07 · Use a proxy when you have more than one upstream. Skip it when you have one local daemon.
• A small local LLM setup for a team
  2026-04-18 · One machine, one model, one proxy, and the notes that keep the setup from drifting.
• What an OpenAI-compatible server actually implements
  2026-04-02 · The small set of routes clients expect from a server that claims to speak the OpenAI HTTP API.
• Running Ollama on a private network
  2026-03-12 · How a small team keeps an Ollama daemon on a LAN address, with a short list of the routes that matter.