Governance - MeshAI Documentation

classify_risk()

Assign an EU AI Act risk level to an agent.

classification = client.classify_risk(
    agent_id="agt_abc123",
    risk_level="high",
    justification="Processes employee performance data for HR decisions",
    assessed_by="compliance-team",
)

string

required

Agent ID

string

required

Risk level: minimal, limited, high, unacceptable

string

required

Reason for the classification

string

required

Identifier of the person or system performing the assessment

array

Optional list of domain tags associated with the classification

array

Optional list of EU AI Act categories associated with the classification

See Risk Classification for details on each level.

create_policy()

Create a governance policy.

policy = client.create_policy(
    name="production-model-allowlist",
    policy_type="model_allowlist",
    rules={"allowed_models": ["gpt-4o", "claude-sonnet-4-20250514"]},
    conditions={"environments": ["production"]},
    enabled=True,
)

string

required

Policy name

string

required

One of: model_allowlist, block_provider, require_approval, budget_limit, rate_limit, require_human_review, prompt_injection, pii_filter

dict

required

Policy rules (varies by type; must be non-empty)

dict

Optional scoping: agent_ids, team_ids, environments, risk_levels

integer

default:"100"

Evaluation priority, 1 to 1000

boolean

default:"true"

Whether the policy is active

See Policies for all policy types, rules keys, and the enforcement model.

list_approvals()

List HITL approval items.

# All pending approvals
pending = client.list_approvals(status="pending")

# All approvals for an agent
agent_approvals = client.list_approvals(agent_id="agt_abc123")

string

Filter by status: pending, approved, denied, expired

string

Filter by agent ID

integer

default:"20"

Results per page

See HITL Approvals for the full workflow.

decide_approval()

Approve or deny a pending approval item.

# Approve
client.decide_approval(
    12345,
    decision="approved",
    reviewer_id="reviewer-jane",
    reason="Agent configuration verified",
)

# Deny
client.decide_approval(
    12345,
    decision="denied",
    reviewer_id="reviewer-jane",
    reason="Agent not authorized for production",
)

integer

required

Approval request ID

string

required

approved or denied

string

required

Identifier of the reviewer making the decision

string

Reason for the decision (recorded in audit trail)

Additional Methods

One-line reference for other governance methods.

Method	Description
`get_risk_classification(agent_id)`	Get current risk classification for an agent.
`get_risk_suggestion(agent_id)`	Get AI-assisted risk level suggestion based on agent metadata.
`list_risk_classifications(**params)`	List all risk classifications. Supports risk_level, page, limit.
`get_policy(policy_id)`	Get a single policy.
`update_policy(policy_id, **fields)`	Update policy fields such as name, enabled, priority, conditions, and rules.
`delete_policy(policy_id)`	Delete a governance policy.
`list_policies(**params)`	List governance policies. Supports policy_type, enabled, page, limit.
`evaluate_policies(agent_id, provider, model, team_id=None, environment="production")`	Dry-run evaluate all policies against a request context.
`get_pending_count()`	Get count of pending approval requests.

list_audit_events()

Query the immutable audit trail.

# Recent events
events = client.list_audit_events(limit=50)

# Filtered
events = client.list_audit_events(
    event_type="policy.violated",
    target_id="agt_abc123",
    from_date="2026-03-01T00:00:00Z",
    to_date="2026-03-17T23:59:59Z",
    limit=100,
)

string

Filter by event type (e.g., agent.registered, policy.violated)

string

Filter by target entity ID

string

Start of time range (ISO 8601)

string

End of time range (ISO 8601)

integer

default:"50"

Results per page

See Audit Trail for all event types and export options.

​classify_risk()

​create_policy()

​list_approvals()

​decide_approval()

​Additional Methods

​list_audit_events()

classify_risk()

create_policy()

list_approvals()

decide_approval()

Additional Methods

list_audit_events()