API reference¶

The reference is generated from the library's docstrings and type annotations. It documents each package's public API: the names in its __all__. Anything else is internal and may change without notice.

Packages¶

The packages are layers. Each depends only on the ones below it, so each is usable without the ones above (Architecture).

Package	What it holds	Install
`artifactr.core`	Every rule, as pure functions over immutable values: artifact types, actors, commands, events, patches, rejections, change notes and the protocol's frames	core
`artifactr.telemetry`	Tracing and metrics through the OpenTelemetry API: span attribution, the metric registry, and untraced polling	core
`artifactr.workspace`	Tenant-scoped workspace handles, the storage protocol and in-memory storage	core
`artifactr.agent`	The pydantic-ai capability, the session, the runner and live output	core
`artifactr.scores`	Feedback as scores: the mirror that follows the log, on evalr's mapping and ports	`langfuse` or `evals` extra
`artifactr.sql`	SQL storage on PostgreSQL and SQLite, and its migrations	`sql`, `postgres` or `sqlite` extra
`artifactr.fastapi`	The thread protocol over WebSocket, and REST, as a FastAPI router	`fastapi` extra
`artifactr.otel`	`configure_telemetry`: the OpenTelemetry SDK, exporters and instrumentations, and metric views, composed with other libraries' contributions	`otel` extra
`artifactr.langfuse`	Langfuse behind evalr's score ports, a span filter for whole traces, and a turn's trace attributes	`langfuse` extra
`artifactr.litellm`	A model over a LiteLLM proxy, and a capability for each request's metadata, key and guardrails	`litellm` extra
`artifactr.mcp`	An MCP server for external agents	`mcp` extra
`artifactr.evals`	evalr for artifactr: datasets from the log, experiments that replay turns, online evaluation of turns, and the end-to-end measures	`evals` extra

The wire formats have their own pages: the thread protocol and its JSON Schema.

The top-level package¶

artifactr re-exports the names most applications need, so from artifactr import Workspaces, Runner works. Each is documented in its layer:

Name	Documented in
`Artifact`, `MarkdownArtifact`, `Versioned`, `WritePolicy`	`artifactr.core`: Artifacts
`new_id`	`artifactr.core`: Identifiers
`Actor`, `UserActor`, `AgentActor`, `ExternalAgentActor`, `SystemActor`	`artifactr.core`: Actors
`Applied`, `Proposed`, `Resolved`, `Recorded`	`artifactr.core`: Outcomes
`Rejection`, `VersionConflict`, `NotFound`	`artifactr.core`: Rejections
`JsonPatch`, `TextEdit`, `TextEdits`	`artifactr.core`: Patches
`Workspaces`, `Workspace`, `InMemoryStorage`	`artifactr.workspace`
`ArtifactWorkspace`, `Session`, `Runner`, `RunHandle`, `ArtifactDraft`	`artifactr.agent`

artifactr.__version__ is the installed version.