Reference

Config Reference

Quick answer

Prism stores its settings in config.json in its per-user config directory, and its model routing in a sibling model_remapping.json. Both files are created on first run, can be edited by hand or through the Admin UI, and are hot-reloaded — Prism picks up a change within seconds without a restart.

Last updated 2026-10-06Reviewed against Prism v0.3.26

File locations

Platform	Config directory
Windows	%APPDATA%\prism\
macOS	~/Library/Application Support/prism/
Linux	$XDG_CONFIG_HOME/prism/ (default ~/.config/prism/)

The same directory holds the proxy config, the model routing file, the stats database, and the managed SearXNG runtime:

config.json           # proxy, provider, search, MCP and telemetry settings
model_remapping.json  # default model, known models and aliases
stats.db              # request statistics (SQLite, WAL mode)
analytics_id.txt      # random anonymous telemetry id (only if telemetry ran)
codex_catalog.json    # model catalog written for Codex Desktop
searxng/              # managed SearXNG venv, settings and caches
logs/                 # log files (Linux; Windows and macOS log to the config dir)

Top-level keys

Key	Type	What it does
default_provider	string	Fallback provider id used when a model has no explicit provider. Usually ollama_cloud or opencode_go.
ollama_cloud	object	Credentials for the built-in Ollama Cloud provider: { id, name, base_url, api_key }.
opencode_go	object	Credentials for the built-in OpenCode Go provider: { id, name, base_url, api_key }.
custom_providers	array	Any number of OpenAI-compatible providers, each with id, name, base_url and api_key. IDs look like custom_openrouter_ab12cd.
oauth_accounts	array	Codex/OpenAI OAuth accounts. Prism manages access tokens, refresh tokens and usage for each entry.
agent_integrations	object	One-click agent setup state: claude_code_tiers { opus, sonnet, haiku, subagent } and auto_sync { <agent-id>: bool }.
searxng_autostart	boolean	Start the managed SearXNG instance together with Prism.
debug_logs	boolean	Verbose logging. Also toggled from the Proxy tab in the Admin UI.
search	object	Web search providers, fallback chain, limits, and custom REST providers. See below.
mcp	object	MCP gateway servers, per-agent access, registries and idle timeout. See below.
analytics_opt_out	boolean	Set true to disable the anonymous daily heartbeat. See Telemetry & Privacy.
analytics_notice_seen	boolean	Whether the first-run telemetry notice has been shown.

Example config.json

{
  "default_provider": "ollama_cloud",
  "ollama_cloud": {
    "id": "ollama_cloud",
    "name": "Ollama Cloud",
    "base_url": "https://ollama.com",
    "api_key": "your-ollama-cloud-key"
  },
  "custom_providers": [
    {
      "id": "custom_openrouter_ab12cd",
      "name": "OpenRouter",
      "base_url": "https://openrouter.ai/api/v1",
      "api_key": "sk-or-v1-..."
    }
  ],
  "agent_integrations": {
    "claude_code_tiers": {
      "opus": "glm-5.1:cloud",
      "sonnet": "deepseek-v4-flash:cloud",
      "haiku": "deepseek-v4-flash:cloud",
      "subagent": "deepseek-v4-flash:cloud"
    },
    "auto_sync": { "claude-code": true, "codex": true }
  },
  "searxng_autostart": true,
  "debug_logs": false,
  "analytics_opt_out": false
}

search — web search providers

Key	Default	What it does
active	searxng	The provider used for search interception. One of searxng, exa, tavily, brave or serper, or a custom provider id.
fallback	[]	Ordered list of provider ids to try when the active provider fails or returns nothing.
max_per_turn	5	How many search calls Prism will serve within a single agent turn.
timeout_ms	8000	Per-request timeout for a search call.
default_num_results	5	Results requested per query. Capped at 10.
providers	{}	Per-provider settings keyed by id, e.g. providers.exa.api_key or providers.searxng.base_url.
custom_providers	[]	User-defined declarative REST providers. See the Web Search page for the full field list.

mcp — gateway and servers

Key	Default	What it does
servers	[]	Upstream MCP servers. Each has id, name, transport (stdio \| http \| sse), enabled, and either command/args/env for stdio or url/headers for HTTP.
agent_servers	{}	Per-agent allowlist mapping an agent id to the server ids it may reach through /mcp/<agent>.
idle_timeout_sec	300	How long a warmed stdio server stays alive after its last use before Prism stops it.
registries	[]	Marketplace catalogs. The official registry is seeded as builtin and cannot be removed; private registries add auth_header.
auto_connect	true	Let Prism start an OAuth sign-in by itself the first time an agent uses a server that has no token yet.
client_id_metadata_url	—	Opt-in CIMD registration URL for OAuth servers that require a hosted client metadata document.

model_remapping.json

Model routing lives in a separate file. It has three sections: the fallback default_model, the known_models catalog (which carries each model's provider and capabilities), and aliases for name rewriting.

{
  "default_model": "glm-5.1:cloud",
  "known_models": [
    {
      "id": "glm-5.1:cloud",
      "provider": "ollama_cloud",
      "reasoning": true,
      "reasoning_effort": ["low", "medium", "high"],
      "context_length": 128000,
      "max_output_tokens": 16384,
      "capabilities": {
        "tool_calling": true,
        "structured_outputs": true,
        "vision": true
      },
      "api": "chat_completions"
    },
    {
      "id": "gpt-5-codex",
      "provider": "codex_ab12cd",
      "api": "responses"
    }
  ],
  "aliases": {
    "claude-3-5-sonnet-20241022": "glm-5.1:cloud",
    "claude-3-5-haiku-20241022": "deepseek-v4-flash:cloud"
  }
}

Model field	What it does
id	The model name clients send, and the name Prism forwards upstream.
provider	Which configured provider serves this model. This is what makes provider-per-model routing work.
reasoning	Whether the model accepts a reasoning/thinking parameter.
reasoning_effort	Allowed effort values for reasoning models. Prism strips or normalizes invalid values instead of failing the request.
context_length	Context window, used to fill agent-side model metadata.
max_output_tokens	Maximum output tokens, used to fill agent-side model metadata.
capabilities	tool_calling, structured_outputs and vision flags, so agents only advertise what the model can do.
api	chat_completions (default) or responses. Model-level override of which inbound-style protocol Prism uses upstream.

Environment variables

Variable	Default	What it does
PRISM_PORT	11434	Proxy listen port. Change it if a real Ollama already owns 11434.
PRISM_HOST	127.0.0.1	Proxy bind address.
PRISM_ADMIN_PORT	8765	Admin web UI port.
OLLAMA_API_KEY	—	Fallback Ollama Cloud key when the config file has none.
OPENCODE_GO_API_KEY	—	Fallback OpenCode Go key when the config file has none.
EXA_API_KEY	—	Auto-detected Exa key for web search.
TAVILY_API_KEY	—	Auto-detected Tavily key for web search.
BRAVE_SEARCH_API_KEY	—	Auto-detected Brave Search key.
SERPER_API_KEY	—	Auto-detected Serper key.
PRISM_ANALYTICS_DISABLED	—	Set to 1 to disable the anonymous heartbeat.
PRISM_ANALYTICS_DEBUG	—	Set to 1 to log telemetry payloads instead of sending them.
KIMI_CODE_HOME	—	Override where Prism looks for Kimi Code's config.
DSH_HOME	—	Override where Prism looks for DeepSeek Harness profiles.
HERMES_HOME	—	Override where Prism looks for Hermes' config.

Precedence

A value in config.json wins over the matching environment variable. Environment variables are the fallback, not an override.

Command-line flags

Prism has exactly one flag. Everything else is configured through the file, the environment, or the Admin UI.

prism --serve    # run the headless proxy in the console, without the tray icon

No autostart flags

Earlier builds documented --autostart, --no-autostart and --minimized. Those flags do not exist. Auto-start is toggled from the tray menu or the Admin UI Proxy tab, and auto-started instances start minimised in the tray on their own.

Config Reference

File locations

Top-level keys

Example config.json

search — web search providers

mcp — gateway and servers

model_remapping.json

Environment variables

Command-line flags

Related

Provider Setup

Model Remapping

Web Search

MCP Gateway