Configuration

All configuration keys, defaults, and examples for jido and jido_ai.

Jump To Section
jido — Core jido_ai — AI integration Next steps

Configuration is set in your config/*.exs files. Runtime secrets belong in config/runtime.exs. All keys have sensible defaults — you only need to configure what you want to change.

jido — Core

Telemetry

Controls structured logging for agent commands, signal processing, and directive execution.

config :jido, :telemetry,
  log_level: :debug,                    # :trace | :debug | :info | :warning | :error
  log_args: :keys_only,                 # :keys_only | :full | :none
  slow_signal_threshold_ms: 10,         # log signals slower than this
  slow_directive_threshold_ms: 5,       # log directives slower than this
  interesting_signal_types: [           # always log these signal types at debug level
    "jido.strategy.init",
    "jido.strategy.complete"
  ]
Key Type Default Description
log_level atom :debug Minimum level for telemetry logs. :trace logs everything; :info logs only request start/stop.
log_args atom :keys_only How action arguments appear in logs. :full logs values, :none suppresses entirely.
slow_signal_threshold_ms integer 10 Signals exceeding this duration are always logged at debug level.
slow_directive_threshold_ms integer 5 Directives exceeding this duration are always logged at debug level.
interesting_signal_types list ["jido.strategy.init", "jido.strategy.complete"] Signal types always logged at debug level regardless of duration.

Observability

Controls the Jido.Observe instrumentation layer.

config :jido, :observability,
  log_level: :info,            # Logger level for observe spans
  debug_events: :off,          # :off | :minimal | :all
  redact_sensitive: false       # redact sensitive fields in telemetry metadata
Key Type Default Description
log_level Logger.level :info Logger level for observe-layer span logging.
debug_events atom :off Buffer debug events in-process. :minimal captures key events, :all captures everything.
redact_sensitive boolean false When true, scrubs keys like api_key, password, secret, and *_token from telemetry metadata.

Timeouts

Default timeouts for agent operations. Override per-call where needed.

config :jido, :timeouts,
  agent_server_shutdown_ms: 5_000,
  agent_server_call_ms: 5_000,
  agent_server_await_ms: 10_000,
  worker_pool_checkout_ms: 5_000,
  worker_pool_call_ms: 5_000
Key Type Default Description
agent_server_shutdown_ms integer 5_000 Graceful shutdown timeout for AgentServer workers.
agent_server_call_ms integer 5_000 Timeout for synchronous AgentServer calls.
agent_server_await_ms integer 10_000 Timeout for AgentServer.await_completion/2.
worker_pool_checkout_ms integer 5_000 Checkout timeout for worker pool agents.
worker_pool_call_ms integer 5_000 Call timeout when signaling pooled agents.

Per-instance configuration

Each Jido instance can override global config under its own module key:

# Global default
config :jido, :telemetry, log_level: :info

# Instance override — this instance gets trace-level logging
config :my_app, MyApp.Jido,
  telemetry: [log_level: :trace],
  observability: [debug_events: :all]

Resolution order (highest priority first):

  1. Jido.Debug runtime override (persistent_term, per-instance)
  2. Per-instance app config (config :my_app, MyApp.Jido, ...)
  3. Global app config (config :jido, :telemetry / config :jido, :observability)
  4. Hardcoded default

jido_ai — AI integration

Model aliases

Map semantic names to provider model strings. Override defaults or add your own:

config :jido_ai,
  model_aliases: %{
    fast: "anthropic:claude-haiku-4-5",
    capable: "anthropic:claude-sonnet-4-20250514",
    thinking: "anthropic:claude-sonnet-4-20250514",
    reasoning: "anthropic:claude-sonnet-4-20250514",
    planning: "anthropic:claude-sonnet-4-20250514",
    image: "openai:gpt-image-1",
    embedding: "openai:text-embedding-3-small"
  }

The table above shows built-in defaults. Your config merges on top — define only the aliases you want to change. Use aliases in code with Jido.AI.resolve_model(:fast).

A full list of supported provider/model IDs is available at llmdb.xyz.

LLM defaults

Role-based defaults for the generate_text/2, generate_object/3, and stream_text/2 facade functions:

config :jido_ai,
  llm_defaults: %{
    text: %{model: :fast, temperature: 0.2, max_tokens: 1024, timeout: 30_000},
    object: %{model: :thinking, temperature: 0.0, max_tokens: 1024, timeout: 30_000},
    stream: %{model: :fast, temperature: 0.2, max_tokens: 1024, timeout: 30_000}
  }
Kind Used by Default model Default temperature
:textJido.AI.generate_text/2:fast 0.2
:objectJido.AI.generate_object/3:thinking 0.0
:streamJido.AI.stream_text/2:fast 0.2

Per-kind config merges with defaults, so you only need to specify fields you want to override.

Provider API keys

Set provider API keys in config/runtime.exs (never commit these to source control):

# config/runtime.exs
config :req_llm,
  anthropic_api_key: System.get_env("ANTHROPIC_API_KEY"),
  openai_api_key: System.get_env("OPENAI_API_KEY")

See req_llm HexDocs for the full list of supported providers and their key names.

Next steps

Edit this page →