How cairndev-sh/cairn works
Cairn is a self-hosted control plane for running AI/ML work as governed pipelines: users declare and launch flows or multi-step pipelines, and Cairn applies admission, budget, policy, and approval controls before execution, then records run state and audit evidence. The map emphasizes that workflow and its major documented capabilities, including dataflow engines, evaluation, and retrieval-augmented generation. Several links are documented in the README or SOURCE INDEX; unsampled implementation details are not treated as inspected evidence.
Main entry points: leg_executor.py Leg executor · pipeline_launch.py Pipeline launch · pipeline.py Pipeline definitions · pipeline_bridge.py Pipeline coordinator
Components 20
Pipeline authoring
- CLI commands
src/cairn/cli/commands/pipeline.py - Pipeline definitions
src/cairn/core/engine/pipeline.py - Input presets
src/cairn/core/pipeline_presets.py - Domain pack registry
src/cairn/core/pack/registry.py
Governance and control
- Pipeline launch
src/cairn/core/dispatch/pipeline_launch.py - Policy and approvals
src/cairn/core/policy/pipeline_gates.py - Budget controls
src/cairn/core/budget_reservations.py - HTTP pipeline routes
src/cairn/core/triggers/routers/pipelines.py
Execution
- Pipeline coordinator
src/cairn/core/dispatch/pipeline_bridge.py - Leg executor
src/cairn/core/dispatch/leg_executor.py - Flow definitions
src/cairn/core/dataflow/defs.py - Engine selection
src/cairn/core/dataflow/engine_registry.py - Dataflow engine adapters
packages/cairn-operator-dataflow/src/operator_dataflow/engines.py - Local training operator
packages/cairn-operator-train-local/src/operator_train_local/operators/lora.py
State and evidence
- Pipeline records
src/cairn/core/storage/pipeline_store.py - Workspace isolation
src/cairn/core/workspace.py - Audit and run records
src/cairn/core/obs/audit.py
AI capabilities
- Model evaluation
src/cairn/core/eval/scoring.py - Retrieval augmented generation
src/cairn/services/rag/query.py
Other components
- User
Connections 20
- User → CLI commands: invokes
- CLI commands → Pipeline launch: launches
src/cairn/core/dispatch/pipeline_launch.py - HTTP pipeline routes → Pipeline launch: launches
src/cairn/core/dispatch/pipeline_launch.py - Pipeline launch → Pipeline definitions: resolves
src/cairn/core/dispatch/pipeline_launch.py - Pipeline launch → Pipeline coordinator: starts
src/cairn/core/dispatch/pipeline_launch.py - Pipeline launch → Policy and approvals: checks gates
src/cairn/core/dispatch/pipeline_launch.py - Pipeline launch → Input presets: loads inputs
src/cairn/core/dispatch/pipeline_launch.py - Pipeline definitions → Domain pack registry: uses pack
src/cairn/core/engine/pipeline.py - Pipeline coordinator → Leg executor: submits steps
src/cairn/core/dispatch/pipeline_bridge.py - Pipeline coordinator → Pipeline records: persists state
src/cairn/core/dispatch/pipeline_bridge.py - Leg executor → Flow definitions: resolves flow
src/cairn/core/dispatch/leg_executor.py - Leg executor → Engine selection: selects engine
src/cairn/core/dispatch/leg_executor.py - Engine selection → Dataflow engine adapters: loads adapter
src/cairn/core/dataflow/engine_registry.py - Leg executor → Budget controls: reserves cost
src/cairn/core/dispatch/leg_executor.py - Leg executor → Policy and approvals: enforces policy
- Leg executor → Local training operator: runs operator
- Pipeline records → Audit and run records: records outcomes
- Pipeline coordinator → Workspace isolation: scopes work
src/cairn/core/dispatch/pipeline_bridge.py - Pipeline definitions → Model evaluation: can define evaluation
- Pipeline definitions → Retrieval augmented generation: can use retrieval