Python API reference

Import the supported user-facing surface from avalanche:

import avalanche as ava

Workflow authoring

API	Purpose
`ava.source`	Declare an ingest/create node
`ava.step`	Declare a deterministic node
`ava.transform`	Alias of `ava.step`
`ava.dest`	Declare an explicit destination/final-action node
`ava.workflow`	Declare a runnable workflow DAG
`ava.pipeline`	Declare a pipeline
`ava.input`	Select a runtime input field during graph construction
`ava.Workflow`, `ava.Pipeline`	Constructed workflow types

See Build a DAG.

Agent integration

API	Purpose
`ava.Agent`	Injected callable agent type
`ava.agent_step`, `ava.agent.step`	Declare an agent-backed workflow node
`ava.Signature`, `ava.InputField`, `ava.OutputField`	DSPy-compatible typed model contract helpers
`ava.agent.Skill`, `ava.agent.skills`	PredictRLM skill surface
`ava.agent.File`	PredictRLM file type exposed by agent integration
`ava.agent.capture_agent_evidence`	Capture agent evidence events
`ava.agent.AgentEvidenceEvent`, `AgentEvidenceListener`, `AgentEvidenceObserverEvent`	Agent evidence event interfaces
`ava.agent.AgentInvocationId`, `AgentTraceFinishedEvent`, `AgentTraceUnavailableEvent`, `ListenerErrorPolicy`	Agent evidence metadata and policies
`ava.agent.AgentStepError`, `AgentStepExecutionError`	Agent-step error types

See Declare agent steps.

Runtime values and providers

API	Purpose
`ava.BaseInput`, `ava.BaseContext`, `ava.RunContext`	Typed run input and context
`ava.File`, `ava.Workspace`	Portable file and directory values
`ava.RunHandle`	Awaitable embedded-run lifecycle handle
`ava.Rerun`	Partial rerun descriptor
`ava.Stream`, `ava.ModelStream`, `ava.consume_stream`	Table-backed runtime stream providers
`ava.Cursor`	Manual table checkpoint provider
`ava.Logger`	Injected runtime logging provider
`ava.ProgressStore`	Progress tracking store

The avalanche.runtime namespace additionally exposes get_current_run_context, run_with_context, and LoggerInstance.

Executors and execution services

API	Purpose
`ava.Executor`	Executor protocol
`ava.LocalExecutor`	Local execution
`ava.RayExecutor`	Optional Ray-backed execution
`ava.get_default_executor`	Resolve the configured default executor
`ava.ExecutionServices`, `ava.ExecutionServicesSpec`	Advanced executor-owned lifecycle provider and run specification
`ava.ExecutionTaskSpec`, `ava.ExecutionServiceReceipt`	Task identity and terminal receipt types
`ava.EXECUTION_SERVICES_V1`	Current execution-services protocol version

See Choose an executor and Integrate execution services.

Storage, tables, and lineage

API	Purpose
`ava.Namespace`, `ava.NamespaceConfig`	Backend-neutral namespace contract and configuration
`ava.Table`, `ava.TableGroup`, `ava.ScanResult`	Backend-neutral table interfaces and scan result
`ava.AppendResult`	Append result and typed conversion helpers
`ava.SnapshotState`, `ava.SnapshotMetadata`	Snapshot metadata types
`ava.Json`	Mark a model field for JSON storage

Iceberg and Lance

API	Purpose
`ava.IcebergNamespace`, `ava.IcebergNs`, `ava.IcebergNsConfig`	Iceberg namespace names and configuration
`ava.IcebergTable`, `ava.IcebergTableGroup`, `ava.IcebergAppendScan`	Iceberg tables and append scan provider
`ava.LanceNamespace`, `ava.LanceNs`, `ava.LanceNamespaceConfig`, `ava.LanceNsConfig`	Lance namespace names and configuration
`ava.LanceTable`	Lance-backed table

See schemas and tables and backend selection.

Other public types

API	Purpose
`ava.Webhook`	Workflow webhook declaration

Avalanche’s public surface is intentionally centered on the root package and ava.agent integration namespace. Operator, TUI, and browser implementation modules are documented as runtime behavior rather than supported application imports.

​Workflow authoring

​Agent integration

​Runtime values and providers

​Executors and execution services

​Storage, tables, and lineage

​Iceberg and Lance

​Other public types