Configuration | ARGUS Docs

Constructor Parameters

Everything is optional. Pass to the ArgusWatcher constructor to customize behavior per run.

Core

graphStateGraph

LangGraph graph to monitor. If passed, watch() is called automatically.

Default: None

max_field_sizeint

Max characters per field before truncation in stored outputs.

Default: 50_000

strictbool

Enable extra checks: nested error keys, rate-limit responses, empty lists, type mismatches. Recommended for CI/staging.

Default: False

investigatebool | "always"

LLM root-cause investigation. True = on failure only, "always" = every node, False = off.

Default: True

Security

redact_keysset[str]

Field names to redact from stored outputs (e.g. {"password", "api_key"}).

Default: None

validatorsdict

Per-node semantic validators. Use "*" as key to run on every node. Each validator is a (bool, str) callable.

Default: None

Latency

node_timeout_msfloat | None

Flag nodes that take ≥95% of this value as timeout-adjacent (likely truncated output). Pass via ArgusConfig.

Default: None

min_expected_msfloat | None

Flag LLM nodes completing faster than this as suspiciously fast (likely cached/stale). Pass via ArgusConfig.

Default: None

Replay & Evaluation

persist_statebool

Save run records to .argus/runs/. Set False for ephemeral monitoring.

Default: True

record_httpbool

Record all external HTTP/API calls for deterministic replay.

Default: True

semantic_judgebool

LLM-powered quality judge on every node output. Requires OPENAI_API_KEY.

Default: False

judge_modelstr

Model for the semantic judge and investigation.

Default: "gpt-4o"

Full Example

python

watcher = ArgusWatcher(
    graph,
    semantic_judge=True,
    judge_model="gpt-4o-mini",
    strict=True,
    record_http=True,
    redact_keys={"api_key", "token"},
    validators={
        "summarize": lambda o: (len(o.get("summary", "")) > 10, "Summary too short"),
    },
)

Environment Variables

Configuration can also be set via environment variables with the ARGUS_ prefix. Environment variables override config file values.

Core

ARGUS_STRICTbool

Halt execution when a detection fires. Useful in CI/CD to fail builds on quality regressions.

Default: false

ARGUS_INVESTIGATEbool | "always"

Run forensic root cause analysis. Set to "always" to analyze even when no detections fire.

Default: true

ARGUS_MAX_FIELD_SIZEint

Maximum character length for captured state fields.

Default: 50000

Semantic

ARGUS_SEMANTIC_JUDGEbool

Enable LLM-as-judge for semantic detection. Requires OPENAI_API_KEY.

Default: false

ARGUS_JUDGE_MODELstr

Model to use for LLM-as-judge evaluation.

Default: "gpt-4o"

Precedence

Configuration values are resolved in this order (highest priority first):

Constructor arguments — values passed directly to ArgusWatcher()
Environment variables — ARGUS_* vars
Config file — argus.yaml in project root
Defaults — built-in sensible defaults

Tip

Use constructor arguments for per-run overrides, environment variables for per-environment settings (dev vs prod), and the config file for project-wide defaults.

Validators

Validators catch semantic failures — when the structure is fine but the value is wrong. Each validator is a callable that returns (bool, str).

python

watcher = ArgusWatcher(graph, validators={
    "classify": lambda o: (o.get("label") in ["yes", "no"], "unexpected label"),
    "*":        lambda o: ("error" not in o, "error key present"),
})

"*" runs on every node.

Strict Mode

Strict mode enables additional detection patterns beyond the defaults: nested error keys, rate limit responses, empty required lists, and list[int] vs list[str] type mismatches.

python

# Recommended for CI/staging
watcher = ArgusWatcher(graph, strict=True)

Security

Always add sensitive field names to redact_keys. ARGUS captures full state at every step — without redaction, API keys and secrets will appear in your stored runs.