Python SDK Reference

Shared Types

from llama_stack_client.types import (
    AgentConfig,
    BatchCompletion,
    CompletionMessage,
    ContentDelta,
    Document,
    InterleavedContent,
    InterleavedContentItem,
    Message,
    ParamType,
    QueryConfig,
    QueryResult,
    ReturnType,
    SafetyViolation,
    SamplingParams,
    ScoringResult,
    SystemMessage,
    ToolCall,
    ToolParamDefinition,
    ToolResponseMessage,
    URL,
    UserMessage,
)

Toolgroups

Types:

from llama_stack_client.types import (
    ListToolGroupsResponse,
    ToolGroup,
    ToolgroupListResponse,
)

Methods:

• client.toolgroups.list() -> ToolgroupListResponse

• client.toolgroups.get(toolgroup_id) -> ToolGroup

• client.toolgroups.register(**params) -> None

• client.toolgroups.unregister(toolgroup_id) -> None

Tools

Types:

from llama_stack_client.types import ListToolsResponse, Tool, ToolListResponse

Methods:

• client.tools.list(**params) -> ToolListResponse

• client.tools.get(tool_name) -> Tool

ToolRuntime

Types:

from llama_stack_client.types import ToolDef, ToolInvocationResult

Methods:

• client.tool_runtime.invoke_tool(**params) -> ToolInvocationResult

• client.tool_runtime.list_tools(**params) -> JSONLDecoder[ToolDef]

RagTool

Methods:

• client.tool_runtime.rag_tool.insert(**params) -> None

• client.tool_runtime.rag_tool.query(**params) -> QueryResult

Agents

DEPRECATED API

The Agents API is deprecated. Use the Responses API instead.

The Responses API provides equivalent functionality with an OpenAI-compatible interface. New applications should use client.responses.create() rather than the agents workflow below.

Types:

from llama_stack_client.types import (
    InferenceStep,
    MemoryRetrievalStep,
    ShieldCallStep,
    ToolExecutionStep,
    ToolResponse,
    AgentCreateResponse,
)

Methods:

• client.agents.create(**params) -> AgentCreateResponse

• client.agents.delete(agent_id) -> None

Session

Types:

from llama_stack_client.types.agents import Session, SessionCreateResponse

Methods:

• client.agents.session.create(agent_id, **params) -> SessionCreateResponse

• client.agents.session.retrieve(session_id, *, agent_id, **params) -> Session

• client.agents.session.delete(session_id, *, agent_id) -> None

Steps

Types:

from llama_stack_client.types.agents import StepRetrieveResponse

Methods:

• client.agents.steps.retrieve(step_id, *, agent_id, session_id, turn_id) -> StepRetrieveResponse

Turn

Types:

from llama_stack_client.types.agents import Turn, TurnCreateResponse

Methods:

• client.agents.turn.create(session_id, *, agent_id, **params) -> TurnCreateResponse

• client.agents.turn.retrieve(turn_id, *, agent_id, session_id) -> Turn

Datasets

Types:

from llama_stack_client.types import (
    ListDatasetsResponse,
    DatasetRetrieveResponse,
    DatasetListResponse,
)

Methods:

• client.datasets.retrieve(dataset_id) -> Optional[DatasetRetrieveResponse]

• client.datasets.list() -> DatasetListResponse

• client.datasets.register(**params) -> None

• client.datasets.unregister(dataset_id) -> None

Eval

Types:

from llama_stack_client.types import EvaluateResponse, Job

Methods:

• client.eval.evaluate_rows(benchmark_id, **params) -> EvaluateResponse

• client.eval.run_eval(benchmark_id, **params) -> Job

Jobs

Types:

from llama_stack_client.types.eval import JobStatusResponse

Methods:

• client.eval.jobs.retrieve(job_id, *, benchmark_id) -> EvaluateResponse

• client.eval.jobs.cancel(job_id, *, benchmark_id) -> None

• client.eval.jobs.status(job_id, *, benchmark_id) -> Optional[JobStatusResponse]

Inspect

Types:

from llama_stack_client.types import HealthInfo, ProviderInfo, RouteInfo, VersionInfo

Methods:

• client.inspect.health() -> HealthInfo

• client.inspect.version() -> VersionInfo

Inference

Types:

from llama_stack_client.types import (
    CompletionResponse,
    EmbeddingsResponse,
    TokenLogProbs,
    InferenceChatCompletionResponse,
    InferenceCompletionResponse,
)

Methods:

• client.inference.embeddings(**params) -> EmbeddingsResponse

VectorIo

DEPRECATED API

This API is deprecated and will be removed in a future version.

Use the OpenAI-compatible Vector Stores API instead:

• Instead of client.vector_io.insert(), use client.vector_stores.files.create() and client.vector_stores.files.chunks.create()
• Instead of client.vector_io.query(), use client.vector_stores.search()

See the RAG documentation for migration examples.

Related: Issue #2981

Types:

from llama_stack_client.types import QueryChunksResponse

Methods:

• client.vector_io.insert(**params) -> None

• client.vector_io.query(**params) -> QueryChunksResponse

VectorDBs

DEPRECATED API

This API is deprecated and will be removed in a future version.

Use the OpenAI-compatible Vector Stores API instead:

• Instead of client.vector_dbs.register(), use client.vector_stores.create()
• Instead of client.vector_dbs.list(), use client.vector_stores.list()
• Instead of client.vector_dbs.retrieve(), use client.vector_stores.retrieve()
• Instead of client.vector_dbs.unregister(), use client.vector_stores.delete()

See the RAG documentation for migration examples.

Related: Issue #2981

Types:

from llama_stack_client.types import (
    ListVectorDBsResponse,
    VectorDBRetrieveResponse,
    VectorDBListResponse,
    VectorDBRegisterResponse,
)

Methods:

• client.vector_dbs.retrieve(vector_db_id) -> Optional[VectorDBRetrieveResponse]

• client.vector_dbs.list() -> VectorDBListResponse

• client.vector_dbs.register(**params) -> VectorDBRegisterResponse

• client.vector_dbs.unregister(vector_db_id) -> None

Models

Types:

from llama_stack_client.types import ListModelsResponse, Model, ModelListResponse

Methods:

• client.models.retrieve(model_id) -> Optional[Model]

• client.models.list() -> ModelListResponse

• client.models.register(**params) -> Model

• client.models.unregister(model_id) -> None

PostTraining

UNAVAILABLE API

The Post Training API is not currently available in Llama Stack. There are no active providers implementing this API. The SDK types remain for forward compatibility but these endpoints are non-functional.

Types:

from llama_stack_client.types import ListPostTrainingJobsResponse, PostTrainingJob

Methods:

• client.post_training.preference_optimize(**params) -> PostTrainingJob

• client.post_training.supervised_fine_tune(**params) -> PostTrainingJob

Job

Types:

from llama_stack_client.types.post_training import (
    JobListResponse,
    JobArtifactsResponse,
    JobStatusResponse,
)

Methods:

• client.post_training.job.list() -> JobListResponse

• client.post_training.job.artifacts(**params) -> Optional[JobArtifactsResponse]

• client.post_training.job.cancel(**params) -> None

• client.post_training.job.status(**params) -> Optional[JobStatusResponse]

Providers

Types:

from llama_stack_client.types import ListProvidersResponse, ProviderListResponse

Methods:

• client.providers.list() -> ProviderListResponse

Routes

Types:

from llama_stack_client.types import ListRoutesResponse, RouteListResponse

Methods:

• client.routes.list() -> RouteListResponse

Safety

Types:

from llama_stack_client.types import RunShieldResponse

Methods:

• client.safety.run_shield(**params) -> RunShieldResponse

Shields

Types:

from llama_stack_client.types import ListShieldsResponse, Shield, ShieldListResponse

Methods:

• client.shields.retrieve(identifier) -> Optional[Shield]

• client.shields.list() -> ShieldListResponse

• client.shields.register(**params) -> Shield

SyntheticDataGeneration

UNAVAILABLE API

The Synthetic Data Generation API is not currently available in Llama Stack. There are no active providers implementing this API. The SDK types remain for forward compatibility but these endpoints are non-functional.

Types:

from llama_stack_client.types import SyntheticDataGenerationResponse

Methods:

• client.synthetic_data_generation.generate(**params) -> SyntheticDataGenerationResponse

Datasetio

Types:

from llama_stack_client.types import PaginatedRowsResult

Methods:

• client.datasetio.append_rows(**params) -> None

• client.datasetio.get_rows_paginated(**params) -> PaginatedRowsResult

Scoring

Types:

from llama_stack_client.types import ScoringScoreResponse, ScoringScoreBatchResponse

Methods:

• client.scoring.score(**params) -> ScoringScoreResponse

• client.scoring.score_batch(**params) -> ScoringScoreBatchResponse

ScoringFunctions

Types:

from llama_stack_client.types import (
    ListScoringFunctionsResponse,
    ScoringFn,
    ScoringFunctionListResponse,
)

Methods:

• client.scoring_functions.retrieve(scoring_fn_id) -> Optional[ScoringFn]

• client.scoring_functions.list() -> ScoringFunctionListResponse

• client.scoring_functions.register(**params) -> None

Benchmarks

Types:

from llama_stack_client.types import (
    Benchmark,
    ListBenchmarksResponse,
    BenchmarkListResponse,
)

Methods:

• client.benchmarks.retrieve(benchmark_id) -> Optional[Benchmark]

• client.benchmarks.list() -> BenchmarkListResponse

• client.benchmarks.register(**params) -> None