Workflows

Workflows let you chain AI models, connect datasets, and build intelligent pipelines. Automate complex multi-step processes visually or programmatically.

What Are Workflows?

A workflow is a directed graph of nodes connected by edges. Each node performs an operation—running a model, reading a dataset, transforming data—and passes results to the next node.

graph LR
    A[Start] --> B[Speech-to-Text]
    B --> C[NER Model]
    C --> D[LLM Analysis]
    D --> E[Output Dataset]

Why Use Workflows?

Use Case	Example
Multi-model pipelines	OCR → NER → Classification
Document processing	Audio → Transcript → Summary
Data enrichment	Images → Descriptions → Metadata
Quality assurance	Multiple models voting on results

Quick Example

Create a Workflow

from seeme import Client

client = Client()

## Create workflow
workflow = client.create_workflow(
    name="Document Analysis Pipeline",
    description="STT → NER → Summary"
)

# Create a version
version = client.create_workflow_version(
    workflow_id=workflow.id,
    name="v1"
)

# Add nodes
stt_node = client.create_workflow_node(
    version_id=version.id,
    name="Transcribe Audio",
    entity_type="model",
    entity_id=stt_model.id,
    config={"input_template": "{{input}}"}
)

ner_node = client.create_workflow_node(
    version_id=version.id,
    name="Extract Entities",
    entity_type="model",
    entity_id=ner_model.id,
    config={"input_template": "{{stt_node}}"}
)

# Connect nodes
client.create_workflow_edge(
    version_id=version.id,
    begin_node_id=stt_node.id,
    end_node_id=ner_node.id,
    edge_type="data"
)

Execute a Workflow

# Execute workflow with text input
execution = client.execute_workflow(
    workflow_id=workflow.id,
    input_mode="single",
    single_input="Analyze this customer call recording..."
)

# Or with a file
execution = client.execute_workflow(
    workflow_id=workflow.id,
    input_mode="single",
    item="./recording.mp3"
)

# Monitor progress
import time
while execution.status in ["pending", "running"]:
    time.sleep(5)
    execution = client.get_workflow_execution(
        workflow_id=workflow.id,
        execution_id=execution.id
    )
    print(f"Status: {execution.status}, Progress: {execution.progress}")

# Get results
print(execution.results)

# Start execution
curl -X POST "https://api.seeme.ai/api/v1/workflows/WORKFLOW_ID/execute" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input_mode": "single", "single_input": "Your text here"}'

# Check status
curl "https://api.seeme.ai/api/v1/workflows/WORKFLOW_ID/executions/EXEC_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

Key Concepts

Concept	Description
Workflow	Container for versions and configurations
Version	Specific configuration of nodes and edges
Node	Individual processing step (model, dataset, transform)
Edge	Connection between nodes (data or context)
Execution	Single run of a workflow with input

Workflow Sections

Concepts Building Workflows Node Types Edge Types Execution & Monitoring Examples

Common Patterns

Document Processing Pipeline

graph LR
    A[Upload PDF] --> B[OCR]
    B --> C[NER]
    C --> D[Classification]
    D --> E[Store Results]

Multi-Model Ensemble

graph TD
    A[Input Image] --> B[Model A]
    A --> C[Model B]
    A --> D[Model C]
    B --> E[Aggregate Results]
    C --> E
    D --> E

RAG Pipeline

graph LR
    A[User Query] --> B[Embedding Model]
    B --> C[Vector Search]
    C --> D[Context Assembly]
    D --> E[LLM Response]