Skip to the page

Providers

Another server

A model server you run yourself, or a service with an OpenAI-compatible address, as a provider of its own.

A model server you run yourself — llama.cpp, vLLM, Jan, LiteLLM — or a service with an OpenAI-compatible address. Give it a name, and its models join the picker.

  • Your own models
  • On your network too
  • Uses your integrations

Worth knowing first

  • Plain http only on this computer or your own network; anywhere else, the address must be https.
  • Tool-capable models can use Conch’s files and connected apps. Commands need the OS sandbox and run without network access; chat-only models cannot take actions.

Add one

  1. Open Settings → Providers and choose Another server.
  2. Type its address: localhost:8080, gpu-box:8000 or an https:// address. Conch says what it found, like “llama.cpp, with 3 models”.
  3. Give it a name, and a key if it asks. Press Add server.

A server already running here on its usual port is offered under Found on this computer.

Starting points

Choose llama.cpp, vLLM, Jan or LiteLLM on your computer, or Together AI, Fireworks, Hugging Face, NVIDIA or Venice online, and its address is filled in.

Good to know

  • Plain http stays at home: this computer and your own network, tailnet included. Anywhere else needs https.
  • A key goes to that server only.
  • A server at localhost answers offline, and can take over a chat when the internet drops.
  • Ollama and LM Studio here have their own cards, which do more. Type their address and Conch sends you there. One on another machine is added here.
  • Hugging Face and Venice need their key first before Conch calls them ready.
  • A backup names its servers. Restoring one that adds a server shows it first.