Configuration reference#

Maida is configured via environment variables and YAML config files. All settings are optional; defaults are tuned for safe, local-only tracing.

Precedence#

Configuration is merged in this order (highest wins):

Environment variables
Project config: .maida/config.yaml in project root (current working directory when config is loaded)
User config: ~/.maida/config.yaml
Defaults (see below)

Settings#

Data directory#

Env	YAML key	Default	Description
`MAIDA_DATA_DIR`	`data_dir`	`~/.maida`	Base directory for runs. Runs are stored under `<data_dir>/runs/<trace_id>/`.

Example (env):

export MAIDA_DATA_DIR=/path/to/my/maida/data

Example (YAML):

# ~/.maida/config.yaml or .maida/config.yaml
data_dir: /path/to/my/maida/data

Redaction#

Env	YAML key	Default	Description
`MAIDA_REDACT`	`redact`	`1` (on)	Enable redaction. Use `1`, `true`, or `yes` to enable; any other value disables.
`MAIDA_REDACT_KEYS`	`redact_keys`	`api_key,token,authorization,cookie,secret,password`	Comma-separated list of key patterns (case-insensitive substring match).
`MAIDA_MAX_FIELD_BYTES`	`max_field_bytes`	`20000`	Maximum size in bytes for a string/field before truncation. Minimum enforced: 100.

Redaction behavior:

Applied recursively to nested dicts and lists (e.g. span attributes, projected event payloads, and meta).
Key match: If a dict key contains any of the redact keys as a case-insensitive substring, the value is replaced with __REDACTED__ (the key is not redacted; the value is). Example: auth_token matches token; API_KEY matches api_key.
Recursion depth: Traversal is limited to depth 10 to avoid pathological structures; deeper values are replaced with the truncation marker.

Truncation behavior:

Strings (and other values serialized to strings) longer than MAIDA_MAX_FIELD_BYTES bytes (UTF-8) are truncated and suffixed with __TRUNCATED__.
At the recursion depth limit (10), the value is replaced with __TRUNCATED__.

Example (env):

export MAIDA_REDACT=1
export MAIDA_REDACT_KEYS="api_key,token,password,secret"
export MAIDA_MAX_FIELD_BYTES=10000

Example (YAML):

redact: true
redact_keys:
  - api_key
  - token
  - authorization
  - password
max_field_bytes: 10000

Loop detection#

Env	YAML key	Default	Description
`MAIDA_LOOP_WINDOW`	`loop_window`	`12`	Number of recent events to consider for pattern detection. Minimum: 4.
`MAIDA_LOOP_REPETITIONS`	`loop_repetitions`	`3`	Consecutive repetitions of a pattern required to emit `LOOP_WARNING`. Minimum: 2.

Example (env):

export MAIDA_LOOP_WINDOW=20
export MAIDA_LOOP_REPETITIONS=4

Example (YAML):

loop_window: 20
loop_repetitions: 4

Guardrails#

Guardrails are opt-in limits that stop a run after Maida has enough evidence to show why it was aborted. They are applied after events are recorded, so the trace still contains the event that crossed the threshold.

Env	YAML key	Default	Description
`MAIDA_STOP_ON_LOOP`	`guardrails.stop_on_loop`	`false`	Abort when loop detection emits `LOOP_WARNING`.
`MAIDA_STOP_ON_LOOP_MIN_REPETITIONS`	`guardrails.stop_on_loop_min_repetitions`	`3`	Minimum repeated pattern count required to abort when `stop_on_loop` is enabled. Minimum: 2.
`MAIDA_MAX_LLM_CALLS`	`guardrails.max_llm_calls`	`null`	Abort after more than N LLM calls. Triggers at N+1.
`MAIDA_MAX_TOOL_CALLS`	`guardrails.max_tool_calls`	`null`	Abort after more than N tool calls. Triggers at N+1.
`MAIDA_MAX_EVENTS`	`guardrails.max_events`	`null`	Abort after more than N total events.
`MAIDA_MAX_DURATION_S`	`guardrails.max_duration_s`	`null`	Abort when elapsed run time reaches the configured number of seconds.

Important behavior:

Existing event view only: guardrails use normal projected LOOP_WARNING, ERROR, and RUN_END records; they do not add special trace types.
Loop aborts: if stop_on_loop=true, Maida writes LOOP_WARNING first, then aborts.
Count-based limits: max_llm_calls=10 allows 10 calls and aborts after the 11th is recorded.
Exception propagation: guardrail aborts raise LoopAbort or GuardrailExceeded; they are not swallowed.
Decorator/context-manager args win: values passed to @trace(...) or traced_run(...) override env and YAML config.

Example (env):

export MAIDA_STOP_ON_LOOP=1
export MAIDA_STOP_ON_LOOP_MIN_REPETITIONS=3
export MAIDA_MAX_LLM_CALLS=50
export MAIDA_MAX_TOOL_CALLS=50
export MAIDA_MAX_EVENTS=200
export MAIDA_MAX_DURATION_S=60

Example (YAML):

guardrails:
  stop_on_loop: true
  stop_on_loop_min_repetitions: 3
  max_llm_calls: 50
  max_tool_calls: 50
  max_events: 200
  max_duration_s: 60

See Guardrails for usage examples and lifecycle details.

Run name (env only)#

Env	YAML	Default	Description
`MAIDA_RUN_NAME`	(not in YAML)	(derived)	Override the run name for the current process. Used when starting a run via `@trace`, `traced_run()`, or implicit run. If unset, run name is the explicit name argument, or a default like `path/to/script.py:main - 2026-02-18 14:12`.

Example:

export MAIDA_RUN_NAME="my-experiment-v1"

Implicit run (env only)#

Env	YAML	Default	Description
`MAIDA_IMPLICIT_RUN`	(not in YAML)	unset (off)	If set to `1`, the first `record_*` call with no active run creates an implicit run; all subsequent recorder calls attach to it until process exit.

This is useful for scripts without a single @trace entrypoint. Only read from the environment; not configurable via YAML.

Example:

export MAIDA_IMPLICIT_RUN=1

Full YAML example#

# ~/.maida/config.yaml or .maida/config.yaml
data_dir: ~/.maida
redact: true
redact_keys:
  - api_key
  - token
  - authorization
  - cookie
  - secret
  - password
max_field_bytes: 20000
loop_window: 12
loop_repetitions: 3
guardrails:
  stop_on_loop: false
  stop_on_loop_min_repetitions: 3
  max_llm_calls: null
  max_tool_calls: null
  max_events: null
  max_duration_s: null

Safe-by-default local traces#

Redaction is on by default so that common secret-key values are not written to disk.
Data directory defaults to ~/.maida so traces are local by default.
Maida.AI does not receive traces, prompts, outputs, source code, logs, or secrets by default. Local traces can still contain sensitive content depending on what your instrumentation records, so keep redaction on unless you are working in a trusted local setting.
No cloud storage is used by the default trace-storage path. Override only what you need (e.g. MAIDA_DATA_DIR for project-local storage, or MAIDA_REDACT=0 for trusted local inspection with full payloads).