Skip to content

Glossary

Key terms and concepts used in the IVCAP Lambda SDK.

Artifact

A file or data object stored in the IVCAP platform. Lambda tools accept artifact URNs as input parameters and upload new artifacts as results.

Lambda Service

A service that exposes one or more tool functions as HTTP endpoints. Each invocation is independent and returns a result (or defers via the try-later pattern). Contrast with a batch service, which processes jobs from a queue.

Batch Service

A service built with ivcap-service that processes discrete jobs asynchronously from a platform queue. ivcap-lambda extends ivcap-service to add HTTP scaffolding.

JobContext

The context object injected into your tool function by the framework. Provides access to the job ID, EventReporter, and the IVCAP client for platform APIs.

Tool Function

A Python function decorated with @ivcap_lambda that becomes an HTTP endpoint. Accepts a Pydantic request model and returns a Pydantic result model.

ToolOptions

A Pydantic model that configures per-tool behaviour: wait timeout, retry interval, OpenAPI tags, service ID, executor options, and readiness check.

Try-Later / 204 Semantics

The asynchronous response pattern used by lambda services. When a job exceeds max_wait_time, the server returns 204 No Content with Location and Retry-Later headers so the caller can poll GET /jobs/{job_id} later.

Event / Progress Event

A message sent during tool execution to report progress, errors, or other status information back to the IVCAP platform. Emitted via JobContext.report.

Step

A named phase of tool execution. Steps are reported as structured events with start, optional info updates, and finish markers.

MCP (Model Context Protocol)

An open standard for AI agents to discover and call tools. The --with-mcp flag enables an /mcp endpoint on the lambda server that speaks the MCP protocol.

Request / Result Models

Pydantic BaseModel subclasses that define the input (Request) and output (Result) of a tool. Must be decorated with @with_schema from ivcap_service.

Schema URN

A urn: identifier attached to request/result models via @with_schema. Used by the IVCAP platform to identify payload types. Format: urn:{namespace}:schema:{name}.{version}.

ExecutorOpts

Configuration for the per-tool Executor: thread pool size, job result cache size, and cache TTL.

Sidecar Reporter

The component (from ivcap-service) that forwards progress events from the tool's execution thread to the IVCAP platform sidecar via HTTP.

OpenTelemetry / OTEL

An open standard for exporting logs, metrics, and traces. Enabled with --with-telemetry; the FastAPI app is automatically instrumented.

URN

Uniform Resource Name. IVCAP uses URNs to identify resources. Format: urn:ivcap:{type}:{uuid}.

See Also