Skip to content

Quick start

This page assumes Prism is installed and running (Install Prism).

The console runs on 127.0.0.1:18081. prism console prints its URL and the owner token. It reads config.toml in the current directory, so run it where the config is:

18081/prism/ui/
cd ~/.prism && prism console # one-command installer
# token: 6f1c…

With the Debian package the config is /etc/prism/config.toml and the token file belongs to the prism user:

Terminal window
sudo -u prism sh -c 'cd /etc/prism && prism console'

Open the URL in a browser on the same machine, choose Sign in with owner secret and paste the token. To open the console from another machine, see Open the console from another machine.

In the console open AI providers and connect one of:

  • an API key of a provider (OpenAI, Anthropic, Google, OpenRouter, DeepSeek and others from the list);
  • a subscription account through OAuth (Claude, Gemini, Cursor and Grok CLI forbid third-party programs and stay closed unless the central Prism administrator allows them for your installation);
  • your own OpenAI-compatible endpoint, for example a local Ollama, vLLM or llama.cpp server.

Prism then discovers the provider’s models. Test request on a connection sends one short real request through that connection (mode Only this connection does not fall through to another account) and shows the model that served it, latency, tokens and cost.

Without the console, add a block per provider to config.toml and put the key in the environment Prism starts with. Known providers need only a name:

[[providers]]
name = "openrouter" # key from OPENROUTER_API_KEY

The built-in key variables are OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY (provider google), OPENROUTER_API_KEY, GROQ_API_KEY and DEEPSEEK_API_KEY; any provider can name its own with api_key_env. With the installer, keys go into ~/.config/prism/prism.env; with the Debian package, into /etc/prism/prism.env. A key in the environment without a [[providers]] block is not used. Restart Prism after editing either file. After its first start Prism keeps settings in the console; a block added to config.toml later is imported into the console at the next start, and the start log says so (or why not). prism config check lists the providers Prism will start with.

A keyless local server needs only its address:

[[providers]]
name = "local-llama"
base_url = "http://127.0.0.1:8080/v1"
routing_role = "local"
Terminal window
curl -s http://127.0.0.1:18000/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{"model": "auto", "messages": [{"role": "user", "content": "Hello"}]}'

"model": "auto" lets Prism choose the model. The response headers X-Prism-Provider and X-Prism-Model name the provider and model that answered, and X-Prism-Cost-USD the cost Prism computed. The request also appears in the console under Usage history → Request details.

The main address http://127.0.0.1:18000 takes no Prism key. Clients that require a key accept any non-empty value:

Terminal window
export OPENAI_BASE_URL=http://127.0.0.1:18000/v1
export OPENAI_API_KEY=prism

Claude Code, Codex CLI, Gemini CLI, the Anthropic SDK and Ollama clients need other settings: see Connect clients.