Configuration reference
Prism reads config.toml once at startup. On its first start Prism moves every setting except the startup keys (storage.dir, console.*, personal.enabled, the license and trust keys, sidecar.*) into the console’s settings store, settings.toml in the storage directory, and from then on the console is where settings are changed. The file keeps the moved keys only so an older version finds them after a rollback. A key or [[providers]] block added to or changed in the file later is imported into the console at the next start, unless the console has changed that value since; the start log names every imported or ignored key, and prism config check shows what the next start will do. To add a provider without a restart, use the console (Connections) or POST /prism/personal/custom-providers.
An unknown or misspelled key does not stop Prism: it prints config warning: unknown key "<key>" (ignored) at startup and uses the default. prism config check validates a file without starting Prism.
Defaults are the values Prism applies to an empty file. A blank default means the key is unset; where Prism then uses a built-in value, the description says so.
| Section | What it controls |
|---|---|
[server] |
Addresses Prism listens on for model requests, request limits, browser origins and the optional managed and control listeners. |
[console] |
The web console listener, its TLS, the owner token file, restarts and updates from the console. |
[personal] |
Owner profile, users, sign-in sessions and passkeys behind the web console. |
[routing] |
Default routing strategy, model denylist, role aliases, provider failover and account selection. |
[profiles] |
Named routing profiles that an agent or request can select. |
[providers] |
Model providers: name, address, key source, routing role, limits and declared models. |
[local] |
Local model servers given by address only. |
[budget] |
Spending limits per period and what Prism does when a threshold is reached. |
[health] |
Provider health checks, rate-limit cooldowns and funding probes. |
[runtime_policy] |
Timeouts of upstream requests and streams, and internal buffer sizes. |
[admission] |
Queue limits for concurrent requests (off by default). |
[storage] |
Where Prism keeps its state and how long it keeps request history. |
[catalog] |
Sources of the model capability and class catalogs. |
[pricing] |
Source of the model price catalog. |
[metrics] |
An optional separate listener for Prometheus metrics. |
[harness] |
Served-model metadata for agent harnesses and prompt cache refreshes. |
[v2] |
The orchestration layer for selector-routed requests (on by default). |
[traffic_guard] |
Optional checks of model traffic for credentials and personal data (off by default). |
[implicit_outcomes] |
Judging whether a served turn worked from the traffic that follows it (on by default, nothing stored but hashes). |
[mail] |
SMTP server for sign-in links, invitations and notifications. |
[inference] |
Model servers and machines that Prism starts, stops and powers (lifecycle control needs Pro). |
[secrets] |
Reading provider keys from 1Password instead of environment variables. |
[egress] |
Outbound proxy for requests to providers. |
[license] |
The subscription service of Pro, Teams and Enterprise. |
[statistics] |
Optional anonymous usage statistics (off by default). |