Models: C

`Call`

run_id: str (runId on the wire)
node_id: str | None (nodeId on the wire) (optional)
scope: str | None (optional)
tool_id: str (toolId on the wire)
arguments: Any | None (optional)
result: Any | None (optional)

`CallUsage`

usage: kindgi.client.ModelCallTokens
cost_usd: float | None (costUsd on the wire) (optional)
duration_ms: int | None (durationMs on the wire) (optional)
served_model: str | None (servedModel on the wire) (optional)

`CandidateItem`

An output's score: Σ yesWeight and Σ totalWeight over its judged items, and over those among the first k ranked items.

yes_weight: float (yesWeight on the wire)
total_weight: float (totalWeight on the wire)
items: int
judged_items: int (judgedItems on the wire)
top_k: kindgi.client.TopK (topK on the wire)
fresh: kindgi.client.Fresh | None (optional)

`Capability`

root: str

`Capability4`

Capability declaration — see @kindgi/capabilities. Additional properties are permitted so new capability kinds do not require a wire change.

`Capability5`

root: str

`CapabilityCollectionPage`

data: list[kindgi.client.CapabilityDescriptor]
next_cursor: str | None (nextCursor on the wire) (optional)
has_more: bool (hasMore on the wire)

`CapabilityDescriptor`

id: str
feature: Union[str, Annotated[Union[Literal['structured-output', 'vision', 'audio-input', 'audio-output', 'tool-use', 'parallel-tool-use', 'thinking', 'long-context', 'code-execution', 'web-search', 'file-search', 'streaming', 'batch'], str], <kindgi.client._open.OpenEnum object at 0x10700bbd0>]]
description: str
kind: str | None (optional)
params_schema: dict[str, Any] | None (paramsSchema on the wire) (optional)
providers: list[kindgi.client.CapabilityProvider] | None (optional)

`CapabilityProvider`

A provider of the tenant with a model that has the feature.

provider_id: str (providerId on the wire)
models: list[str]

`Change`

block_id: str (blockId on the wire)
content: kindgi.client.Content1

`Changes`

The first repetition's items against the judged ones.

kept: list[kindgi.client.KeptItem]
dropped: list[kindgi.client.DroppedItem]
new: list[kindgi.client.NewItem]

`ClaimMappingScopesSpec`

Claim path + optional delimiter for extracting Kindgi OAuth scopes from provider claims. When absent, scopes come from the token response scope field instead.

claim: str
delimiter: str | None (optional)

`ClaimMappingSpec`

Per-provider claim mapping. All fields optional; closer to OIDC defaults (sub/email/name) → fewer overrides needed.

user_id: str | None (userId on the wire) (optional)
email: str | None (optional)
display_name: str | None (displayName on the wire) (optional)
scopes: kindgi.client.ClaimMappingScopesSpec | None (optional)
metadata: list[kindgi.client.Metadatum] | None (optional)

`CodeArtifactRefFilesystem`

Handler-artifact pointer. Discriminated on kind: oci is the deploy-pipeline shape (image + module path + artifactVersion); filesystem is the local-development shape — absolute host path at the pack's on-disk handler file. Production servers SHOULD reject filesystem.

kind: Literal['filesystem']
module_path: str (modulePath on the wire)

`CodeArtifactRefOci`

kind: Literal['oci']
image_ref: str (imageRef on the wire)
module_path: str (modulePath on the wire)
artifact_version: str (artifactVersion on the wire)

`CodeArtifactRefUnrecognized`

A kind this client doesn't know yet, from a newer runtime: the object as it came, every field kept.

kind: str

`Comparison`

max_age_hours: float | None (maxAgeHours on the wire) (optional)
suite: kindgi.client.Suite | None (optional)

`Comparison1`

eval_run_id: uuid.UUID | None (evalRunId on the wire) (optional)
role: Union[Literal['reference', 'candidate', 'proof'], str]
part: Union[Literal['search', 'hold-out'], str]
block_id: str | None (blockId on the wire) (optional)
changed: dict[str, Any] | None (optional)
score: float | None (optional)
failed: str | None (optional)
refused: list[kindgi.client.RefusedItem] | None (optional)
hypothesis: str | None (optional)

`ComparisonCandidateAgent`

What ran on the cases: an agent version, or a flow version (with any versions it swapped in).

kind: Literal['agent']
agent_id: str (agentId on the wire)
version: str
pins_digest: str | None (pinsDigest on the wire) (optional)
overrides: kindgi.client.Overrides | None (optional)

`ComparisonCandidateFlow`

What ran on the cases: an agent version, or a flow version (with any versions it swapped in).

kind: Literal['flow']
flow_id: str (flowId on the wire)
version: str
versions: kindgi.client.FlowVersionOverrides | None (optional)

`ComparisonCaseResult`

One case of a comparison: its replay runs, the scores, the items kept, dropped and new, the tool calls, and why it didn't run when it didn't.

case_id: str (caseId on the wire)
run_ids: list[str] (runIds on the wire)
baseline: kindgi.client.Baseline2
candidate: list[kindgi.client.CandidateItem]
changes: kindgi.client.Changes | None (optional)
tools: list[kindgi.client.ComparisonCaseResultTool] | None (optional)
diverged: bool
refused_writes: int (refusedWrites on the wire)
no_context: bool (noContext on the wire)
approval_skipped: bool (approvalSkipped on the wire)
error: str | None (optional)
stopped: kindgi.client.Stopped | None (optional)
rescored: Optional[Literal[False]] (optional)

`ComparisonCaseResultTool`

step: int
call_id: str (callId on the wire)
tool_id: str (toolId on the wire)
tool_version: str (toolVersion on the wire)
arguments: Any
source: Literal['live', 'recorded', 'refused']
recomputed: bool | None (optional)
reason: str | None (optional)

`ComparisonMetric`

One metric, the recorded runs beside the candidate. null where a side had no judged evidence; n / weight are the candidate's evidence (cases with judged items, and the judgment weight behind them), baselineN / baselineWeight the recorded side's.

baseline: float | None
candidate: float | None
delta: float | None
n: int
weight: float
baseline_n: int (baselineN on the wire)
baseline_weight: float (baselineWeight on the wire)
direction: Literal['higher']
k: int | None (optional)
spread: float | None (optional)
fresh_weight: float | None (freshWeight on the wire) (optional)

`CompleteApprovalBody`

decision: Literal['approve', 'reject', 'escalate', 'withdraw']
rationale: str | None (optional)
value: Any | None (optional)

`CompleteApprovalResult`

A recorded decision, and what it did to the run waiting on the approval. runStatus is the run's status when the decision couldn't resolve its waitpoint because the run had already ended (e.g. cancelled after the decision was recorded); the decision stands.

kind: Union[Literal['terminal', 'escalated'], str]
approval: kindgi.client.Approval
decision: kindgi.client.ReviewDecision
next_approval: kindgi.client.Approval | None (nextApproval on the wire) (optional)
waitpoint_resolved: bool (waitpointResolved on the wire)
run_status: Optional[Annotated[Union[Literal['pending', 'running', 'suspended', 'completed', 'failed', 'cancelled'], str], <kindgi.client._open.OpenEnum object at 0x10700bbd0>]] (runStatus on the wire) (optional)
resume: Optional[Annotated[kindgi.client.Resume | kindgi.client.Resume1 | kindgi.client.CompleteApprovalResultResumeUnrecognized, <kindgi.client._open.OpenUnion object at 0x107a703d0>]] (optional)

`CompleteApprovalResultResumeUnrecognized`

A kind this client doesn't know yet, from a newer runtime: the object as it came, every field kept.

kind: str

`ComplianceEvidence`

One evidence record — matches @kindgi/specs/compliance-evidence.schema.json (payload is a versioned document opaque to the wire schema).

id: str
tenant_id: uuid.UUID (tenantId on the wire)
project_id: uuid.UUID | None (projectId on the wire) (optional)
kind: str
timestamp: AwareDatetime
actor: kindgi.client.Actor | None (optional)
subject: kindgi.client.Subject1 | None (optional)
outcome: Optional[Annotated[Union[Literal['allowed', 'denied', 'succeeded', 'failed', 'escalated'], str], <kindgi.client._open.OpenEnum object at 0x10700bbd0>]] (optional)
payload: dict[str, Any]
provenance_ref: kindgi.client.ProvenanceRef2 | None (provenanceRef on the wire) (optional)
signature: kindgi.client.Signature1 | None (optional)